The Inference Report

The reliability of multilingual Large Language Model (LLM) evaluation is currently compromised by the inconsistent quality of translated benchmarks. Existing resources often suffer from semantic drift and context loss, which can lead to misleading performance metrics. In this work, we present a fully automated framework designed to address these challenges by enabling scalable, high-quality translation of datasets and benchmarks. We demonstrate that adapting test-time compute scaling strategies, specifically Universal Self-Improvement (USI) and our proposed multi-round ranking method, T-RANK, allows for significantly higher quality outputs compared to traditional pipelines. Our framework ensures that benchmarks preserve their original task structure and linguistic nuances during localization. We apply this approach to translate popular benchmarks and datasets into eight Eastern and Southern European languages (Ukrainian, Bulgarian, Slovak, Romanian, Lithuanian, Estonian, Turkish, Greek). Evaluations using both reference-based metrics and LLM-as-a-judge show that our translations surpass existing resources, resulting in more accurate downstream model assessment. We release both the framework and the improved benchmarks to facilitate robust and reproducible multilingual AI development.

Sumerian transliteration is a conventional system for representing a scholar's interpretation of a tablet in the Latin script. Thanks to visionary digital Assyriology projects such as ETCSL, CDLI, and Oracc, a large number of Sumerian transliterations have been published online, and these data are well-structured for a variety of search and analysis tasks. However, the absence of a comprehensive, accessible dataset pairing transliterations with a digital representation of the tablet's cuneiform glyphs has prevented the application of modern Natural Language Processing (NLP) methods to the task of Sumerian transliteration. To address this gap, we present SumTablets, a dataset pairing Unicode representations of 91,606 Sumerian cuneiform tablets (totaling 6,970,407 glyphs) with the associated transliterations published by Oracc. We construct SumTablets by first preprocessing and standardizing the Oracc transliterations before mapping each reading back to the Unicode representation of the source glyph. Further, we retain parallel structural information (e.g., surfaces, newlines, broken segments) through the use of special tokens. We release SumTablets as a Hugging Face Dataset (CC BY 4.0) and open source data preparation code via GitHub. Additionally, we leverage SumTablets to implement and evaluate two transliteration baselines: (1) weighted sampling from a glyph's possible readings, and (2) fine-tuning an autoregressive language model. Our fine-tuned language model achieves an average transliteration character-level F-score (chrF) of 97.55, demonstrating the immediate potential of transformer-based transliteration models in allowing experts to rapidly verify generated transliterations rather than manually transliterating tablets one-by-one.

Advances in Generative AI (GenAI) have led to the development of various protection strategies to prevent the unauthorized use of images. These methods rely on adding imperceptible protective perturbations to images to thwart misuse such as style mimicry or deepfake manipulations. Although previous attacks on these protections required specialized, purpose-built methods, we demonstrate that this is no longer necessary. We show that off-the-shelf image-to-image GenAI models can be repurposed as generic ``denoisers" using a simple text prompt, effectively removing a wide range of protective perturbations. Across 8 case studies spanning 6 diverse protection schemes, our general-purpose attack not only circumvents these defenses but also outperforms existing specialized attacks while preserving the image's utility for the adversary. Our findings reveal a critical and widespread vulnerability in the current landscape of image protection, indicating that many schemes provide a false sense of security. We stress the urgent need to develop robust defenses and establish that any future protection mechanism must be benchmarked against attacks from off-the-shelf GenAI models. Code is available in this repository: https://github.com/mlsecviswanath/img2imgdenoiser

We study reasoning for accessing world knowledge stored in a language model's parameters. For example, recalling that Canberra is Australia's capital may benefit from thinking through major cities and the concept of purpose-built capitals. While reasoning language models are trained via reinforcement learning to produce reasoning traces on tasks such as mathematics, they may not reason well for accessing their own world knowledge. We first find that models do not generate their best world knowledge reasoning by default: adding a simple "think step-by-step" cue demonstrates statistically significant improvement in knowledge recall but not math. Motivated by this, we propose training models to reason over their parametric knowledge using world-knowledge question answering as a verifiable reward. After reinforcement learning on TriviaQA (+9.9%), performance also improves on Natural Questions, HotpotQA, SimpleQA, and StrategyQA by 4.2%, 2.1%, 0.6%, and 3.0%, respectively. Reasoning models are under-optimized for parametric knowledge access, but can be easily trained to reason better.

Open-source native GUI agents still lag behind closed-source systems on long-horizon navigation tasks. This gap stems from two limitations: a shortage of high-quality, action-aligned reasoning data, and the direct adoption of generic post-training pipelines that overlook the unique challenges of GUI agents. We identify two fundamental issues in these pipelines: (i) standard SFT with CoT reasoning often hurts grounding, and (ii) step-wise RLVR-tyle training faces partial verifiability, where multiple actions can be correct but only a single demonstrated action is used for verification. This makes offline step-wise metrics weak predictors of online task success. In this work, we present GUI-Libra, a tailored training recipe that addresses these challenges. First, to mitigate the scarcity of action-aligned reasoning data, we introduce a data construction and filtering pipeline and release a curated 81K GUI reasoning dataset. Second, to reconcile reasoning with grounding, we propose action-aware SFT that mixes reasoning-then-action and direct-action data and reweights tokens to emphasize action and grounding. Third, to stabilize RL under partial verifiability, we identify the overlooked importance of KL regularization in RLVR and show that a KL trust region is critical for improving offline-to-online predictability; we further introduce success-adaptive scaling to downweight unreliable negative gradients. Across diverse web and mobile benchmarks, GUI-Libra consistently improves both step-wise accuracy and end-to-end task completion. Our results suggest that carefully designed post-training and data curation can unlock significantly stronger task-solving capabilities without costly online data collection. We release our dataset, code, and models to facilitate further research on data-efficient post-training for reasoning-capable GUI agents.

Modelling rock-fluid interaction requires solving a set of partial differential equations (PDEs) to predict the flow behaviour and the reactions of the fluid with the rock on the interfaces. Conventional high-fidelity numerical models require a high resolution to obtain reliable results, resulting in huge computational expense. This restricts the applicability of these models for multi-query problems, such as uncertainty quantification and optimisation, which require running numerous scenarios. As a cheaper alternative to high-fidelity models, this work develops eight surrogate models for predicting the fluid flow in porous media. Four of these are reduced-order models (ROM) based on one neural network for compression and another for prediction. The other four are single neural networks with the property of grid-size invariance; a term which we use to refer to image-to-image models that are capable of inferring on computational domains that are larger than those used during training. In addition to the novel grid-size-invariant framework for surrogate models, we compare the predictive performance of UNet and UNet++ architectures, and demonstrate that UNet++ outperforms UNet for surrogate models. Furthermore, we show that the grid-size-invariant approach is a reliable way to reduce memory consumption during training, resulting in good correlation between predicted and ground-truth values and outperforming the ROMs analysed. The application analysed is particularly challenging because fluid-induced rock dissolution results in a non-static solid field and, consequently, it cannot be used to help in adjustments of the future prediction.

Over the last twenty years, significant progress has been made in designing and implementing Question Answering (QA) systems. However, addressing complex questions, the answers to which are spread across multiple documents, remains a challenging problem. Recent QA systems that are designed to handle complex questions work either on the basis of knowledge graphs, or utilise contem- porary neural models that are expensive to train, in terms of both computational resources and the volume of training data required. In this paper, we present LiCQA, an unsupervised question answer- ing model that works primarily on the basis of corpus evidence. We empirically compare the effectiveness and efficiency of LiCQA with two recently presented QA systems, which are based on different underlying principles. The results of our experiments show that LiCQA significantly outperforms these two state-of-the-art systems on benchmark data with noteworthy reduction in latency.

In many applications, it is important to identify subpopulations that survive longer or shorter than the rest of the population. In medicine, for example, it allows determining which patients benefit from treatment, and in predictive maintenance, which components are more likely to fail. Existing methods for discovering subgroups with exceptional survival characteristics require restrictive assumptions about the survival model (e.g. proportional hazards), pre-discretized features, and, as they compare average statistics, tend to overlook individual deviations. In this paper, we propose Sysurv, a fully differentiable, non-parametric method that leverages random survival forests to learn individual survival curves, automatically learns conditions and how to combine these into inherently interpretable rules, so as to select subgroups with exceptional survival characteristics. Empirical evaluation on a wide range of datasets and settings, including a case study on cancer data, shows that Sysurv reveals insightful and actionable survival subgroups.

Understanding and reasoning over long contexts is a crucial capability for language models (LMs). Although recent models support increasingly long context windows, their accuracy often deteriorates as input length grows. In practice, models often struggle to keep attention aligned with the most relevant context throughout decoding. In this work, we propose DySCO, a novel decoding algorithm for improving long-context reasoning. DySCO leverages retrieval heads--a subset of attention heads specialized for long-context retrieval--to identify task-relevant tokens at each decoding step and explicitly up-weight them. By doing so, DySCO dynamically adjusts attention during generation to better utilize relevant context. The method is training-free and can be applied directly to any off-the-shelf LMs. Across multiple instruction-tuned and reasoning models, DySCO consistently improves performance on challenging long-context reasoning benchmarks, yielding relative gains of up to 25% on MRCR and LongBenchV2 at 128K context length with modest additional compute. Further analysis highlights the importance of both dynamic attention rescaling and retrieval-head-guided selection for the effectiveness of the method, while providing interpretability insights into decoding-time attention behavior. Our code is available at https://github.com/princeton-pli/DySCO.

The inability of Large Language Models (LLMs) to modulate their personality expression in response to evolving dialogue dynamics hinders their performance in complex, interactive contexts. We propose a model-agnostic framework for dynamic personality simulation that employs state machines to represent latent personality states, where transition probabilities are dynamically adapted to the conversational context. Part of our architecture is a modular pipeline for continuous personality scoring that evaluates dialogues along latent axes while remaining agnostic to the specific personality models, their dimensions, transition mechanisms, or LLMs used. These scores function as dynamic state variables that systematically reconfigure the system prompt, steering behavioral alignment throughout the interaction.We evaluate this framework by operationalizing the Interpersonal Circumplex (IPC) in a medical education setting. Results demonstrate that the system successfully adapts its personality state to user inputs, but also influences user behavior, thereby facilitating de-escalation training. Notably, the scoring pipeline maintains comparable precision even when utilizing lightweight, fine-tuned classifiers instead of large-scale LLMs. This work demonstrates the feasibility of modular, personality-adaptive architectures for education, customer support, and broader human-computer interaction.

Cardiovascular disease (CVD) remains one of the leading global health challenges, accounting for more than 19 million deaths worldwide. To address this, several tools that aim to predict CVD risk and support clinical decision making have been developed. In particular, the Framingham Risk Score (FRS) is one of the most widely used and recommended worldwide. However, it does not explain why a patient was assigned to a particular risk category nor how it can be reduced. Due to this lack of transparency, we present a logical explainer for the FRS. Based on first-order logic and explainable artificial intelligence (XAI) fundaments, the explainer is capable of identifying a minimal set of patient attributes that are sufficient to explain a given risk classification. Our explainer also produces actionable scenarios that illustrate which modifiable variables would reduce a patient's risk category. We evaluated all possible input combinations of the FRS (over 22,000 samples) and tested them with our explainer, successfully identifying important risk factors and suggesting focused interventions for each case. The results may improve clinician trust and facilitate a wider implementation of CVD risk assessment by converting opaque scores into transparent and prescriptive insights, particularly in areas with restricted access to specialists.

Reinforcement Learning from Human Feedback (RLHF) plays a significant role in aligning Large Language Models (LLMs) with human preferences. While RLHF with expected reward constraints can be formulated as a primal-dual optimization problem, standard primal-dual methods only guarantee convergence with a distributional policy where the saddle-point problem is in convex-concave form. Moreover, standard primal-dual methods may exhibit instability or divergence in the last iterate under policy parameterization in practical applications. In this work, we propose a universal primal-dual framework for safe RLHF that unifies a broad class of existing alignment algorithms, including safe-RLHF, one-shot, and multi-shot based methods. Building on this framework, we introduce an optimistic primal-dual (OPD) algorithm that incorporates predictive updates for both primal and dual variables to stabilize saddle-point dynamics. We establish last-iterate convergence guarantees for the proposed method, covering both exact policy optimization in the distributional space and convergence to a neighborhood of the optimal solution whose gap is related to approximation error and bias under parameterized policies. Our analysis reveals that optimism plays a crucial role in mitigating oscillations inherent to constrained alignment objectives, thereby closing a key theoretical gap between constrained RL and practical RLHF.

Large Language Models (LLMs) are increasingly used to ``professionalize'' workplace communication, often at the cost of linguistic identity. We introduce "Cultural Ghosting", the systematic erasure of linguistic markers unique to non-native English varieties during text processing. Through analysis of 22,350 LLM outputs generated from 1,490 culturally marked texts (Indian, Singaporean,& Nigerian English) processed by five models under three prompt conditions, we quantify this phenomenon using two novel metrics: Identity Erasure Rate (IER) & Semantic Preservation Score (SPS). Across all prompts, we find an overall IER of 10.26%, with model-level variation from 3.5% to 20.5% (5.9x range). Crucially, we identify a Semantic Preservation Paradox: models maintain high semantic similarity (mean SPS = 0.748) while systematically erasing cultural markers. Pragmatic markers (politeness conventions) are 1.9x more vulnerable than lexical markers (71.5% vs. 37.1% erasure). Our experiments demonstrate that explicit cultural-preservation prompts reduce erasure by 29% without sacrificing semantic quality.

Object hallucination is a critical issue in Large Vision-Language Models (LVLMs), where outputs include objects that do not appear in the input image. A natural question arises from this phenomenon: Which component of the LVLM pipeline primarily contributes to object hallucinations? The vision encoder to perceive visual information, or the language decoder to generate text responses? In this work, we strive to answer this question through designing a systematic experiment to analyze the roles of the vision encoder and the language decoder in hallucination generation. Our observations reveal that object hallucinations are predominantly associated with the strong priors from the language decoder. Based on this finding, we propose a simple and training-free framework, No-Language-Hallucination Decoding, NoLan, which refines the output distribution by dynamically suppressing language priors, modulated based on the output distribution difference between multimodal and text-only inputs. Experimental results demonstrate that NoLan effectively reduces object hallucinations across various LVLMs on different tasks. For instance, NoLan achieves substantial improvements on POPE, enhancing the accuracy of LLaVA-1.5 7B and Qwen-VL 7B by up to 6.45 and 7.21, respectively. The code is publicly available at: https://github.com/lingfengren/NoLan.

Deep neural networks (DNNs) are essential for performing advanced tasks on edge or mobile devices, yet their deployment is often hindered by severe resource constraints, including limited memory, energy, and computational power. While uniform quantization provides a straightforward approach to compress model and reduce hardware requirement, it fails to fully leverage the varying robustness across layers, and often lead to accuracy degradation or suboptimal resource usage, particularly at low bitwidths. In contrast, heterogeneous quantization, which allocates different bitwidths to individual layers, can mitigate these drawbacks. Nonetheless, current heterogeneous quantization methods either needs huge brute-force design space search or lacks the adaptability to meet different hardware conditions, such as memory size, energy budget, and latency requirement. Filling these gaps, this work introduces \textbf{\textit{SigmaQuant}}, an adaptive layer-wise heterogeneous quantization framework designed to efficiently balance accuracy and resource usage for varied edge environments without exhaustive search.

We study the basic task of mean estimation in the presence of mean-shift contamination. In the mean-shift contamination model, an adversary is allowed to replace a small constant fraction of the clean samples by samples drawn from arbitrarily shifted versions of the base distribution. Prior work characterized the sample complexity of this task for the special cases of the Gaussian and Laplace distributions. Specifically, it was shown that consistent estimation is possible in these cases, a property that is provably impossible in Huber's contamination model. An open question posed in earlier work was to determine the sample complexity of mean estimation in the mean-shift contamination model for general base distributions. In this work, we study and essentially resolve this open question. Specifically, we show that, under mild spectral conditions on the characteristic function of the (potentially multivariate) base distribution, there exists a sample-efficient algorithm that estimates the target mean to any desired accuracy. We complement our upper bound with a qualitatively matching sample complexity lower bound. Our techniques make critical use of Fourier analysis, and in particular introduce the notion of a Fourier witness as an essential ingredient of our upper and lower bounds.

Instruction-following benchmarks remain predominantly English-centric, leaving a critical evaluation gap for the hundreds of millions of Indic language speakers. We introduce IndicIFEval, a benchmark evaluating constrained generation of LLMs across 14 Indic languages using automatically verifiable, rule-based instructions. It comprises around 800 human-verified examples per language spread across two complementary subsets: IndicIFEval-Ground, translated prompts from IFEval (Zhou et al., 2023) carefully localized for Indic contexts, and IndicIFEval-Ground, synthetically generated instructions grounded in native Indic content. We conduct a comprehensive evaluation of major open-weight and proprietary models spanning both reasoning and non-reasoning models. While models maintain strong adherence to formatting constraints, they struggle significantly with lexical and cross-lingual tasks -- and despite progress in high-resource languages, instruction-following across the broader Indic family lags significantly behind English. We release IndicIFEval and its evaluation scripts to support progress on multilingual constrained generation (http://github.com/ai4bharat/IndicIFEval).

Small language models (SLMs) offer compelling advantages in cost, latency, and adaptability, but have so far lagged behind larger models on long-horizon software engineering tasks such as SWE-bench, where they suffer from pervasive action looping and low resolution rates. We introduce SWE-Protégé, a post-training framework that reframes software repair as an expert-protégé collaboration problem. In SWE-Protégé, an SLM remains the sole decision-maker while learning to selectively seek guidance from a strong expert model, recognize stalled states, and follow through on expert feedback. Our approach combines supervised fine-tuning on expert-augmented trajectories with agentic reinforcement learning that explicitly discourages degenerative looping and unproductive expert collaboration. We lightly post-train Qwen2.5-Coder-7B-Instruct to achieve 42.4% Pass@1 on SWE-bench Verified, a +25.4% improvement over the prior SLM state of the art, while using expert assistance sparsely (~4 calls per task and 11% of total tokens).

Understanding the geometry of learned distributions is fundamental to improving and interpreting diffusion models, yet systematic tools for exploring their landscape remain limited. Standard latent-space interpolations fail to respect the structure of the learned distribution, often traversing low-density regions. We introduce a framework based on the string method that computes continuous paths between samples by evolving curves under the learned score function. Operating on pretrained models without retraining, our approach interpolates between three regimes: pure generative transport, which yields continuous sample paths; gradient-dominated dynamics, which recover minimum energy paths (MEPs); and finite-temperature string dynamics, which compute principal curves -- self-consistent paths that balance energy and entropy. We demonstrate that the choice of regime matters in practice. For image diffusion models, MEPs contain high-likelihood but unrealistic ''cartoon'' images, confirming prior observations that likelihood maxima appear unrealistic; principal curves instead yield realistic morphing sequences despite lower likelihood. For protein structure prediction, our method computes transition pathways between metastable conformers directly from models trained on static structures, yielding paths with physically plausible intermediates. Together, these results establish the string method as a principled tool for probing the modal structure of diffusion models -- identifying modes, characterizing barriers, and mapping connectivity in complex learned distributions.

Neural networks (NNs) are pervasive across various domains but often lack interpretability. To address the growing need for explanations, logic-based approaches have been proposed to explain predictions made by NNs, offering correctness guarantees. However, scalability remains a concern in these methods. This paper proposes an approach leveraging domain slicing to facilitate explanation generation for NNs. By reducing the complexity of logical constraints through slicing, we decrease explanation time by up to 40\% less time, as indicated through comparative experiments. Our findings highlight the efficacy of domain slicing in enhancing explanation efficiency for NNs.

Despite the extensive literature on training loss functions, the evaluation of generalization on the validation set remains underexplored. In this work, we conduct a systematic empirical and statistical study of how the validation criterion used for model selection affects test performance in neural classifiers, with attention to early stopping. Using fully connected networks on standard benchmarks under $k$-fold evaluation, we compare: (i) early stopping with patience and (ii) post-hoc selection over all epochs (i.e. no early stopping). Models are trained with cross-entropy, C-Loss, or PolyLoss; the model parameter selection on the validation set is made using accuracy or one of the three loss functions, each considered independently. Three main findings emerge. (1) Early stopping based on validation accuracy performs worst, consistently selecting checkpoints with lower test accuracy than both loss-based early stopping and post-hoc selection. (2) Loss-based validation criteria yield comparable and more stable test accuracy. (3) Across datasets and folds, any single validation rule often underperforms the test-optimal checkpoint. Overall, the selected model typically achieves test-set performance statistically lower than the best performance across all epochs, regardless of the validation criterion. Our results suggest avoiding validation accuracy (in particular with early stopping) for parameter selection, favoring loss-based validation criteria.

Many real-world applications provide a continuous stream of data that is subsequently used by machine learning models to solve regression tasks of interest. Hoeffding trees and their variants have a long-standing tradition due to their effectiveness, either alone or as base models in broader ensembles. At the same time a recent line of work in batch learning has shown that kernel density estimation (KDE) is an effective approach for smoothed predictions in imbalanced regression tasks [Yang et al., 2021]. Moreover, another recent line of work for batch learning, called hierarchical shrinkage (HS) [Agarwal et al., 2022], has introduced a post-hoc regularization method for decision trees that does not alter the structure of the learned tree. Using a telescoping argument we cast KDE to streaming environments and extend the implementation of HS to incremental decision tree models. Armed with these extensions we investigate the performance of decision trees that may enjoy such options in datasets commonly used for regression in online settings. We conclude that KDE is beneficial in the early parts of the stream, while HS hardly, if ever, offers performance benefits. Our code is publicly available at: https://github.com/marinaAlchirch/DSFA_2026.

Plans often change due to changes in the situation or our understanding of the situation. Sometimes, a feasible plan may not even exist, and identifying such infeasibilities is useful to determine when requirements need adjustment. Common planning approaches focus on efficient one-shot planning in feasible cases rather than updating domains or detecting infeasibility. We propose a Petri net reachability relaxation to enable robust invariant synthesis, efficient goal-unreachability detection, and helpful infeasibility explanations. We further leverage incremental constraint solvers to support goal and constraint updates. Empirically, compared to baselines, our system produces a comparable number of invariants, detects up to 2 times more infeasibilities, performs competitively in one-shot planning, and outperforms in sequential plan updates in the tested domains.

Large Language Models (LLMs) have revolutionized inference across diverse natural language tasks, with larger models performing better but at higher computational costs. We propose a confidence-driven strategy that dynamically selects the most suitable model based on confidence estimates. By assessing a model's confidence in handling the task and response accuracy, tasks that are likely to be solved correctly are retained, while more uncertain or complex cases are delegated to a larger model, ensuring reliability while minimizing computation. Specifically, we evaluate a model's likelihood of knowing the correct answer and the probability that its response is accurate. Experiments on the Massive Multitask Language Understanding (MMLU) benchmark show that our approach achieves accuracy comparable to the largest model while reducing computational costs by 20\% to 40\%. When applied to GPT-4o API calls, it reduces token usage by approximately 60\%, further improving cost efficiency. These findings indicate the potential of confidence-based model selection to enhance real-world LLM deployment, particularly in resource-constrained settings such as edge devices and commercial API applications.

Van der Waals (vdW) interactions are essential for describing molecules and materials, from drug design and catalysis to battery applications. These omnipresent interactions must also be accurately included in machine-learned force fields. The many-body dispersion (MBD) method stands out as one of the most accurate and transferable approaches to capture vdW interactions, requiring only atomic $C_6$ coefficients and polarizabilities as input. We present MBD-ML, a pretrained message passing neural network that predicts these atomic properties directly from atomic structures. Through seamless integration with libMBD, our method enables the immediate calculation of MBD-inclusive total energies, forces, and stress tensors. By eliminating the need for intermediate electronic structure calculations, MBD-ML offers a practical and streamlined tool that simplifies the incorporation of state-of-the-art vdW interactions into any electronic structure code, as well as empirical and machine-learned force fields.

A class of causal effect functionals requires integration over conditional densities of continuous variables, as in mediation effects and nonparametric identification in causal graphical models. Estimating such densities and evaluating the resulting integrals can be statistically and computationally demanding. A common workaround is to discretize the variable and replace integrals with finite sums. Although convenient, discretization alters the population-level functional and can induce non-negligible approximation bias, even under correct identification. Under smoothness conditions, we show that this coarsening bias is first order in the bin width and arises at the level of the target functional, distinct from statistical estimation error. We propose a simple bias-reduced functional that evaluates the outcome regression at within-bin conditional means, eliminating the leading term and yielding a second-order approximation error. We derive plug-in and one-step estimators for the bias-reduced functional. Simulations demonstrate substantial bias reduction and near-nominal confidence interval coverage, even under coarse binning. Our results provide a simple framework for controlling the impact of variable discretization on parameter approximation and estimation.

This paper explains the Visual Milestone Planning (VMP) method using an agile vocabulary to facilitate its adoption by agile practitioners as a front end for a hybrid development process. VMP is a visual and collaborative planning approach which promotes a shared understanding of the work approach and commitment through the direct manipulation by team members of the reified planning constructs involved in the development of the plan. Once the product backlog has been established and relevant milestones identified, a novel construct called the milestone planning matrix is used to document the allocation of product backlog items to milestones. The milestones due dates are later determined by grouping sticky notes representing the work to be performed into time-boxes called work packages and accommodating them on a resource and time scaled scheduling canvas very much as it would be done in a Tetris game.

Theory of Mind (ToM) refers to an agent's ability to model the internal states of others. Contributing to the debate whether large language models (LLMs) exhibit genuine ToM capabilities, our study investigates their ToM robustness using perturbations on false-belief tasks and examines the potential of Chain-of-Thought prompting (CoT) to enhance performance and explain the LLM's decision. We introduce a handcrafted, richly annotated ToM dataset, including classic and perturbed false belief tasks, the corresponding spaces of valid reasoning chains for correct task completion, subsequent reasoning faithfulness, task solutions, and propose metrics to evaluate reasoning chain correctness and to what extent final answers are faithful to reasoning traces of the generated CoT. We show a steep drop in ToM capabilities under task perturbation for all evaluated LLMs, questioning the notion of any robust form of ToM being present. While CoT prompting improves the ToM performance overall in a faithful manner, it surprisingly degrades accuracy for some perturbation classes, indicating that selective application is necessary.

Large language models are increasingly used in decision-making tasks that require them to process information from a variety of sources, including both human experts and other algorithmic agents. How do LLMs weigh the information provided by these different sources? We consider the well-studied phenomenon of algorithm aversion, in which human decision-makers exhibit bias against predictions from algorithms. Drawing upon experimental paradigms from behavioural economics, we evaluate how eightdifferent LLMs delegate decision-making tasks when the delegatee is framed as a human expert or an algorithmic agent. To be inclusive of different evaluation formats, we conduct our study with two task presentations: stated preferences, modeled through direct queries about trust towards either agent, and revealed preferences, modeled through providing in-context examples of the performance of both agents. When prompted to rate the trustworthiness of human experts and algorithms across diverse tasks, LLMs give higher ratings to the human expert, which correlates with prior results from human respondents. However, when shown the performance of a human expert and an algorithm and asked to place an incentivized bet between the two, LLMs disproportionately choose the algorithm, even when it performs demonstrably worse. These discrepant results suggest that LLMs may encode inconsistent biases towards humans and algorithms, which need to be carefully considered when they are deployed in high-stakes scenarios. Furthermore, we discuss the sensitivity of LLMs to task presentation formats that should be broadly scrutinized in evaluation robustness for AI safety.

Grounding is a critical step in classical planning, yet it often becomes a computational bottleneck due to the exponential growth in grounded actions and atoms as task size increases. Recent advances in partial grounding have addressed this challenge by incrementally grounding only the most promising operators, guided by predictive models. However, these approaches primarily rely on relational features or learned embeddings and do not leverage the textual and structural cues present in PDDL descriptions. We propose SPG-LLM, which uses LLMs to analyze the domain and problem files to heuristically identify potentially irrelevant objects, actions, and predicates prior to grounding, significantly reducing the size of the grounded task. Across seven hard-to-ground benchmarks, SPG-LLM achieves faster grounding-often by orders of magnitude-while delivering comparable or better plan costs in some domains.

Time-series foundation models (TSFMs) have achieved strong univariate forecasting through large-scale pre-training, yet effectively extending this success to multivariate forecasting remains challenging. To address this, we propose DualWeaver, a novel framework that adapts univariate TSFMs (Uni-TSFMs) for multivariate forecasting by using a pair of learnable, structurally symmetric surrogate series. Generated by a shared auxiliary feature-fusion module that captures cross-variable dependencies, these surrogates are mapped to TSFM-compatible series via the forecasting objective. The symmetric structure enables parameter-free reconstruction of final predictions directly from the surrogates, without additional parametric decoding. A theoretically grounded regularization term is further introduced to enhance robustness against adaptation collapse. Extensive experiments on diverse real-world datasets show that DualWeaver outperforms state-of-the-art multivariate forecasters in both accuracy and stability. We release the code at https://github.com/li-jinpeng/DualWeaver.

Generative models for quantum data pose significant challenges but hold immense potential in fields such as chemoinformatics and quantum physics. Quantum denoising diffusion probabilistic models (QuDDPMs) enable efficient learning of quantum data distributions by progressively scrambling and denoising quantum states; however, existing implementations typically rely on circuit-based random unitary dynamics that can be costly to realize and sensitive to control imperfections, particularly on analog quantum hardware. We propose the chaotic quantum diffusion model, a framework that generates projected ensembles via chaotic Hamiltonian time evolution, providing a flexible and hardware-compatible diffusion mechanism. Requiring only global, time-independent control, our approach substantially reduces implementation overhead across diverse analog quantum platforms while achieving accuracy comparable to QuDDPMs. This method improves trainability and robustness, broadening the applicability of quantum generative modeling.

Neural operators have emerged as an efficient paradigm for solving PDEs, overcoming the limitations of traditional numerical methods and significantly improving computational efficiency. However, due to the diversity and complexity of PDE systems, existing neural operators typically rely on a single network architecture, which limits their capacity to fully capture heterogeneous features and complex system dependencies. This constraint poses a bottleneck for large-scale PDE pre-training based on neural operators. To address these challenges, we propose a large-scale PDE pre-trained neural operator based on a nested Mixture-of-Experts (MoE) framework. In particular, the image-level MoE is designed to capture global dependencies, while the token-level Sub-MoE focuses on local dependencies. Our model can selectively activate the most suitable expert networks for a given input, thereby enhancing generalization and transferability. We conduct large-scale pre-training on twelve PDE datasets from diverse sources and successfully transfer the model to downstream tasks. Extensive experiments demonstrate the effectiveness of our approach.

Generative manipulation policies can fail catastrophically under deployment-time distribution shift, yet many failures are near-misses: the robot reaches almost-correct poses and would succeed with a small corrective motion. We present FlowCorrect, a deployment-time correction framework that converts near-miss failures into successes using sparse human nudges, without full policy retraining. During execution, a human provides brief corrective pose nudges via a lightweight VR interface. FlowCorrect uses these sparse corrections to locally adapt the policy, improving actions without retraining the backbone while preserving the model performance on previously learned scenarios. We evaluate on a real-world robot across three tabletop tasks: pick-and-place, pouring, and cup uprighting. With a low correction budget, FlowCorrect improves success on hard cases by 85\% while preserving performance on previously solved scenarios. The results demonstrate clearly that FlowCorrect learns only with very few demonstrations and enables fast and sample-efficient incremental, human-in-the-loop corrections of generative visuomotor policies at deployment time in real-world robotics.

Accurate prediction of shaft rotational speed, shaft power, and fuel consumption is crucial for enhancing operational efficiency and sustainability in maritime transportation. Conventional physics-based models provide interpretability but struggle with real-world variability, while purely data-driven approaches achieve accuracy at the expense of physical plausibility. This paper introduces a Physics-Informed Kolmogorov-Arnold Network (PI-KAN), a hybrid method that integrates interpretable univariate feature transformations with a physics-informed loss function and a leakage-free chained prediction pipeline. Using operational and environmental data from five cargo vessels, PI-KAN consistently outperforms the traditional polynomial method and neural network baselines. The model achieves the lowest mean absolute error (MAE) and root mean squared error (RMSE), and the highest coefficient of determination (R^2) for shaft power and fuel consumption across all vessels, while maintaining physically consistent behavior. Interpretability analysis reveals rediscovery of domain-consistent dependencies, such as cubic-like speed-power relationships and cosine-like wave and wind effects. These results demonstrate that PI-KAN achieves both predictive accuracy and interpretability, offering a robust tool for vessel performance monitoring and decision support in operational settings.

We introduce DLT-Corpus, the largest domain-specific text collection for Distributed Ledger Technology (DLT) research to date: 2.98 billion tokens from 22.12 million documents spanning scientific literature (37,440 publications), United States Patent and Trademark Office (USPTO) patents (49,023 filings), and social media (22 million posts). Existing Natural Language Processing (NLP) resources for DLT focus narrowly on cryptocurrencies price prediction and smart contracts, leaving domain-specific language under explored despite the sector's ~$3 trillion market capitalization and rapid technological evolution. We demonstrate DLT-Corpus' utility by analyzing technology emergence patterns and market-innovation correlations. Findings reveal that technologies originate in scientific literature before reaching patents and social media, following traditional technology transfer patterns. While social media sentiment remains overwhelmingly bullish even during crypto winters, scientific and patent activity grow independently of market fluctuations, tracking overall market expansion in a virtuous cycle where research precedes and enables economic growth that funds further innovation. We publicly release the full DLT-Corpus; LedgerBERT, a domain-adapted model achieving 23% improvement over BERT-base on a DLT-specific Named Entity Recognition (NER) task; and all associated tools and code.

Low-resource automatic speech recognition (ASR) continues to pose significant challenges, primarily due to the limited availability of transcribed data for numerous languages. While a wealth of spoken content is accessible in television dramas and online videos, Taiwanese Hokkien exemplifies this issue, with transcriptions often being scarce and the majority of available subtitles provided only in Mandarin. To address this deficiency, we introduce TG-ASR for Taiwanese Hokkien drama speech recognition, a translation-guided ASR framework that utilizes multilingual translation embeddings to enhance recognition performance in low-resource environments. The framework is centered around the parallel gated cross-attention (PGCA) mechanism, which adaptively integrates embeddings from various auxiliary languages into the ASR decoder. This mechanism facilitates robust cross-linguistic semantic guidance while ensuring stable optimization and minimizing interference between languages. To support ongoing research initiatives, we present YT-THDC, a 30-hour corpus of Taiwanese Hokkien drama speech with aligned Mandarin subtitles and manually verified Taiwanese Hokkien transcriptions. Comprehensive experiments and analyses identify the auxiliary languages that most effectively enhance ASR performance, achieving a 14.77% relative reduction in character error rate and demonstrating the efficacy of translation-guided learning for underrepresented languages in practical applications.

Metro trains often operate in highly complex environments, characterized by illumination variations, high-speed motion, and adverse weather conditions. These factors pose significant challenges for visual perception systems, especially those relying solely on conventional RGB cameras. To tackle these difficulties, we explore the integration of event cameras into the perception system, leveraging their advantages in low-light conditions, high-speed scenarios, and low power consumption. Specifically, we focus on Kilometer Marker Recognition (KMR), a critical task for autonomous metro localization under GNSS-denied conditions. In this context, we propose a robust baseline method based on a pre-trained RGB OCR foundation model, enhanced through multi-modal adaptation. Furthermore, we construct the first large-scale RGB-Event dataset, EvMetro5K, containing 5,599 pairs of synchronized RGB-Event samples, split into 4,479 training and 1,120 testing samples. Extensive experiments on EvMetro5K and other widely used benchmarks demonstrate the effectiveness of our approach for KMR. Both the dataset and source code will be released on https://github.com/Event-AHU/EvMetro5K_benchmark

Integrated Development Environments shape developers' daily experience, yet the empirical study of their usability and user experience (UX) remains limited. This work presents an LLM-assisted approach to detecting UX smells in Visual Studio Code by mining and classifying user-reported issues from the GitHub repository. Using a validated taxonomy and expert review, we identified recurring UX problems that affect the developer experience. Our results show that the majority of UX smells are concentrated in informativeness, clarity, intuitiveness, and efficiency, qualities that developers value most.

Disease progression modeling provides a robust framework to identify long-term disease trajectories from short-term biomarker data. It is a valuable tool to gain a deeper understanding of diseases with a long disease trajectory, such as Alzheimer's disease. A key limitation of most disease progression models is that they are specific to a single data type (e.g., continuous data), thereby limiting their applicability to heterogeneous, real-world datasets. To address this limitation, we propose the Mixed Events model, a novel disease progression model that handles both discrete and continuous data types. This model is implemented within the Subtype and Stage Inference (SuStaIn) framework, resulting in Mixed-SuStaIn, enabling subtype and progression modeling. We demonstrate the effectiveness of Mixed-SuStaIn through simulation experiments and real-world data from the Alzheimer's Disease Neuroimaging Initiative, showing that it performs well on mixed datasets. The code is available at: https://github.com/ucl-pond/pySuStaIn.

Bayesian deep learning (BDL) has emerged as a principled approach to produce reliable uncertainty estimates by integrating deep neural networks with Bayesian inference, and the selection of informative prior distributions remains a significant challenge. Various function-space variational inference (FSVI) regularisation methods have been presented, assigning meaningful priors over model predictions. However, these methods typically rely on a Gaussian prior, which fails to capture the heavy-tailed statistical characteristics inherent in neural network outputs. By contrast, this work proposes a novel function-space empirical Bayes regularisation framework -- termed ST-FS-EB -- which employs heavy-tailed Student's $t$ priors in both parameter and function spaces. Also, we approximate the posterior distribution through variational inference (VI), inducing an evidence lower bound (ELBO) objective based on Monte Carlo (MC) dropout. Furthermore, the proposed method is evaluated against various VI-based BDL baselines, and the results demonstrate its robust performance in in-distribution prediction, out-of-distribution (OOD) detection and handling distribution shifts.

Diversity has been gaining interest in the NLP community in recent years. At the same time, state-of-the-art transformer models such as ModernBERT use very large pre-training datasets, which are driven by size rather than by diversity. This summons for an investigation of the impact of diversity on the ModernBERT pre-training. We do so in this study, with the express intent of reducing pre-training dataset size, while retaining at least comparable performance. We compare diversity-driven sampling algorithms, so as to pick the best one. We find that diversity-driven sampling allows in some tasks to gain 10 points relative to randomly-sampled pre-training data of commensurate size. We also see that a model pre-trained for 483h on a diversity-driven dataset of 150M tokens can yield a commensurate performance to a model pre-trained for 1,775h on a randomly-driven dataset of 2.4B tokens.

In recent years, the machine learning community has increasingly embraced the optimal transport (OT) framework for modeling distributional relationships. In this work, we introduce a sample-based neural solver for computing the Wasserstein geodesic between a source and target distribution, along with the associated velocity field. Building on the dynamical formulation of the optimal transport (OT) problem, we recast the constrained optimization as a minimax problem, using deep neural networks to approximate the relevant functions. This approach not only provides the Wasserstein geodesic but also recovers the OT map, enabling direct sampling from the target distribution. By estimating the OT map, we obtain velocity estimates along particle trajectories, which in turn allow us to learn the full velocity field. The framework is flexible and readily extends to general cost functions, including the commonly used quadratic cost. We demonstrate the effectiveness of our method through experiments on both synthetic and real datasets.

Automated test generation is essential for software quality assurance, with coverage rate serving as a key metric to ensure thorough testing. Recent advancements in Large Language Models (LLMs) have shown promise in improving test generation, particularly in achieving higher coverage. However, while existing LLM-based test generation solutions perform well on small, isolated code snippets, they struggle when applied to complex methods under test. To address these issues, we propose a scalable LLM-based unit test generation method. Our approach consists of two key steps. The first step is context information retrieval, which uses both LLMs and static analysis to gather relevant contextual information associated with the complex methods under test. The second step, iterative test generation with code elimination, repeatedly generates unit tests for the code slice, tracks the achieved coverage, and selectively removes code segments that have already been covered. This process simplifies the testing task and mitigates issues arising from token limits or reduced reasoning effectiveness associated with excessively long contexts. Through comprehensive evaluations on open-source projects, our approach outperforms state-of-the-art LLM-based and search-based methods, demonstrating its effectiveness in achieving high coverage on complex methods.

The optimization of complex medical appointment scheduling remains a significant operational challenge in multi-center healthcare environments, where clinical safety protocols and patient logistics must be reconciled. This study proposes and evaluates a Genetic Algorithm (GA) framework designed to automate the scheduling of multiple medical acts while adhering to rigorous inter-procedural incompatibility rules. Using a synthetic dataset encompassing 50 medical acts across four healthcare facilities, we compared two GA variants, Pre-Ordered and Unordered, against deterministic First-Come, First-Served (FCFS) and Random Choice baselines. Our results demonstrate that the GA framework achieved a 100% constraint fulfillment rate, effectively resolving temporal overlaps and clinical incompatibilities that the FCFS baseline failed to address in 60% and 40% of cases, respectively. Furthermore, the GA variants demonstrated statistically significant improvements (p < 0.001) in patient-centric metrics, achieving an Idle Time Ratio (ITR) frequently below 0.4 and reducing inter-healthcenter trips. While the GA (Ordered) variant provided a superior initial search locus, both evolutionary models converged to comparable global optima by the 100th generation. These findings suggest that transitioning from manual, human-mediated scheduling to an automated metaheuristic approach enhances clinical integrity, reduces administrative overhead, and significantly improves the patient experience by minimizing wait times and logistical burdens.

Medical images are essential for diagnosis, treatment planning, and research, but their quality is often degraded by noise from low-dose acquisition, patient motion, or scanner limitations, affecting both clinical interpretation and downstream analysis. Traditional filtering approaches often over-smooth and lose fine anatomical details, while deep learning methods, including CNNs, GANs, and transformers, may struggle to preserve such details or require large, computationally expensive models, limiting clinical practicality. We propose PatchDenoiser, a lightweight, energy-efficient multi-scale patch-based denoising framework. It decomposes denoising into local texture extraction and global context aggregation, fused via a spatially aware patch fusion strategy. This design enables effective noise suppression while preserving fine structural and anatomical details. PatchDenoiser is ultra-lightweight, with far fewer parameters and lower computational complexity than CNN-, GAN-, and transformer-based denoisers. On the 2016 Mayo Low-Dose CT dataset, PatchDenoiser consistently outperforms state-of-the-art CNN- and GAN-based methods in PSNR and SSIM. It is robust to variations in slice thickness, reconstruction kernels, and HU windows, generalizes across scanners without fine-tuning, and reduces parameters by ~9x and energy consumption per inference by ~27x compared with conventional CNN denoisers. PatchDenoiser thus provides a practical, scalable, and computationally efficient solution for medical image denoising, balancing performance, robustness, and clinical deployability.

Recent work has examined language models from a linguistic perspective to better understand how they acquire language. Most existing benchmarks focus on judging grammatical acceptability, whereas the ability to interpret meanings conveyed by grammatical forms has received much less attention. We introduce the Linguistic Minimal-Pair Benchmark for Evaluating Constructional Understanding in Language Models (CxMP), a benchmark grounded in Construction Grammar that treats form-meaning pairings, or constructions, as fundamental linguistic units. CxMP evaluates whether models can interpret the semantic relations implied by constructions, using a controlled minimal-pair design across nine construction types, including the let-alone, caused motion, and ditransitive constructions. Our results show that while syntactic competence emerges early, constructional understanding develops more gradually and remains limited even in large language models (LLMs). CxMP thus reveals persistent gaps in how language models integrate form and meaning, providing a framework for studying constructional understanding and learning trajectories in language models.

Critical applications in areas such as medicine, robotics and autonomous systems require compact (i.e., memory efficient), uncertainty-aware neural networks suitable for edge and other resource-constrained deployments. We study compact spectral circulant and block-circulant-with-circulant-blocks (BCCB) layers: FFT-diagonalizable circular convolutions whose weights live directly in the real FFT (RFFT) half (1D) or half-plane (2D). Parameterizing filters in the frequency domain lets us impose simple spectral structure, perform structured variational inference in a low-dimensional weight space, and calculate exact layer spectral norms, enabling inexpensive global Lipschitz bounds and margin-based robustness diagnostics. By placing independent complex Gaussians on the Hermitian support we obtain a discrete instance of the spectral representation of stationary kernels, inducing an exact stationary Gaussian-process prior over filters on the discrete circle/torus. We exploit this to define a practical spectral prior and a Hermitian-aware low-rank-plus-diagonal variational posterior in real coordinates. Empirically, spectral circulant/BCCB layers are effective compact building blocks in both (variational) Bayesian and point estimate regimes: compact Bayesian neural networks on MNIST->Fashion-MNIST, variational heads on frozen CIFAR-10 features, and deterministic ViT projections on CIFAR-10/Tiny ImageNet; spectral layers match strong baselines while using substantially fewer parameters and with tighter Lipschitz certificates.

We investigate the robustness of sparse artificial neural networks trained with adaptive topology. We focus on a simple yet effective architecture consisting of three sparse layers with 99% sparsity followed by a dense layer, applied to image classification tasks such as MNIST and Fashion MNIST. By updating the topology of the sparse layers between each epoch, we achieve competitive accuracy despite the significantly reduced number of weights. Our primary contribution is a detailed analysis of the robustness of these networks, exploring their performance under various perturbations including random link removal, adversarial attack, and link weight shuffling. Through extensive experiments, we demonstrate that adaptive topology not only enhances efficiency but also maintains robustness. This work highlights the potential of adaptive sparse networks as a promising direction for developing efficient and reliable deep learning models.

To reduce carbon emissions and minimize shipping costs, improving the fuel efficiency of ships is crucial. Various measures are taken to reduce the total fuel consumption of ships, including optimizing vessel parameters and selecting routes with the lowest fuel consumption. Different estimation methods are proposed for predicting fuel consumption, while various optimization methods are proposed to minimize fuel oil consumption. This paper provides a comprehensive review of methods for estimating and optimizing fuel oil consumption in maritime transport. Our novel contributions include categorizing fuel oil consumption \& estimation methods into physics-based, machine-learning, and hybrid models, exploring their strengths and limitations. Furthermore, we highlight the importance of data fusion techniques, which combine AIS, onboard sensors, and meteorological data to enhance accuracy. We make the first attempt to discuss the emerging role of Explainable AI in enhancing model transparency for decision-making. Uniquely, key challenges, including data quality, availability, and the need for real-time optimization, are identified, and future research directions are proposed to address these gaps, with a focus on hybrid models, real-time optimization, and the standardization of datasets.

Federated recommendation facilitates collaborative model training across distributed clients while keeping sensitive user interaction data local. Conventional approaches typically rely on synchronizing high-dimensional item representations between the server and clients. This paradigm implicitly assumes that precise geometric alignment of embedding coordinates is necessary for collaboration across clients. We posit that establishing relative semantic relationships among items is more effective than enforcing shared representations. Specifically, global semantic relations serve as structural constraints for items. Within these constraints, the framework allows item representations to vary locally on each client, which flexibility enables the model to capture fine-grained user personalization while maintaining global consistency. To this end, we propose Cluster-Guided FedRec framework (CGFedRec), a framework that transforms uploaded embeddings into compact cluster labels. In this framework, the server functions as a global structure discoverer to learn item clusters and distributes only the resulting labels. This mechanism explicitly cuts off the downstream transmission of item embeddings, relieving clients from maintaining global shared item embeddings. Consequently, CGFedRec achieves the effective injection of global collaborative signals into local item representations without transmitting full embeddings. Extensive experiments demonstrate that our approach significantly improves communication efficiency while maintaining superior recommendation accuracy across multiple datasets.

Knowledge graph reasoning (KGR) infers missing facts, with recent advances increasingly harnessing the semantic priors and reasoning abilities of Large Language Models (LLMs). However, prevailing generative paradigms are prone to memorizing surface-level co-occurrences rather than learning genuine relational semantics, limiting out-of-distribution generalization. To address this, we propose RADAR, which reformulates KGR from generative pattern matching to discriminative relational reasoning. We recast KGR as discriminative entity selection, where reinforcement learning enforces relative entity separability beyond token-likelihood imitation. Leveraging this separability, inference operates directly in representation space, ensuring consistency with the discriminative optimization and bypassing generation-induced hallucinations. Across four benchmarks, RADAR achieves 5-6% relative gains on link prediction and triple classification over strong LLM baselines, while increasing task-relevant mutual information in intermediate representations by 62.9%, indicating more robust and transferable relational reasoning.

Multimodal large language models (MLLMs) have shown great potential in medical applications, yet existing benchmarks inadequately capture real-world clinical complexity. We introduce MEDSYN, a multilingual, multimodal benchmark of highly complex clinical cases with up to 7 distinct visual clinical evidence (CE) types per case. Mirroring clinical workflow, we evaluate 18 MLLMs on differential diagnosis (DDx) generation and final diagnosis (FDx) selection. While top models often match or even outperform human experts on DDx generation, all MLLMs exhibit a much larger DDx--FDx performance gap compared to expert clinicians, indicating a failure mode in synthesis of heterogeneous CE types. Ablations attribute this failure to (i) overreliance on less discriminative textual CE ($\it{e.g.}$, medical history) and (ii) a cross-modal CE utilization gap. We introduce Evidence Sensitivity to quantify the latter and show that a smaller gap correlates with higher diagnostic accuracy. Finally, we demonstrate how it can be used to guide interventions to improve model performance. We will open-source our benchmark and code.

Generative Adversarial Networks (GAN) have been used in many studies to synthesise mixed tabular data. Conditional tabular GAN (CTGAN) have been the most popular variant but struggle to effectively navigate the risk-utility trade-off. Bayesian GAN have received less attention for tabular data, but have been explored with unstructured data such as images and text. The most used technique employed in Bayesian GAN is Markov Chain Monte Carlo (MCMC), but it is computationally intensive, particularly in terms of weight storage. In this paper, we introduce Gaussian Approximation of CTGAN (GACTGAN), an integration of the Bayesian posterior approximation technique using Stochastic Weight Averaging-Gaussian (SWAG) within the CTGAN generator to synthesise tabular data, reducing computational overhead after the training phase. We demonstrate that GACTGAN yields better synthetic data compared to CTGAN, achieving better preservation of tabular structure and inferential statistics with less privacy risk. These results highlight GACTGAN as a simpler, effective implementation of Bayesian tabular synthesis.

Large language models (LLMs) demonstrate remarkable breadth of knowledge, yet their ability to reason about computational processes remains poorly understood. Closing this gap matters for practitioners who rely on LLMs to guide algorithm selection and deployment. We address this limitation using causal discovery as a testbed and evaluate eight frontier LLMs against ground truth derived from large-scale algorithm executions and find systematic, near-total failure. Models produce ranges far wider than true confidence intervals yet still fail to contain the true algorithmic mean in the majority of instances; most perform worse than random guessing and the marginal above-random performance of the best model is most consistent with benchmark memorization rather than principled reasoning. We term this failure algorithmic blindness and argue it reflects a fundamental gap between declarative knowledge about algorithms and calibrated procedural prediction.

How can researchers identify beliefs that large language models (LLMs) hide? As LLMs become more sophisticated and the prevalence of alignment faking increases, combined with their growing integration into high-stakes decision-making, responding to this challenge has become critical. This paper proposes that a list experiment, a simple method widely used in the social sciences, can be applied to study the hidden beliefs of LLMs. List experiments were originally developed to circumvent social desirability bias in human respondents, which closely parallels alignment faking in LLMs. The paper implements a list experiment on models developed by Anthropic, Google, and OpenAI and finds hidden approval of mass surveillance across all models, as well as some approval of torture, discrimination, and first nuclear strike. Importantly, a placebo treatment produces a null result, validating the method. The paper then compares list experiments with direct questioning and discusses the utility of the approach.

Coronary artery calcium (CAC) scoring is a key predictor of cardiovascular risk, but it relies on ECG-gated CT scans, restricting its use to specialized cardiac imaging settings. We introduce an automated framework for CAC detection and lesion-specific Agatston scoring that operates across both gated and non-gated CT scans. At its core is CARD-ViT, a self-supervised Vision Transformer trained exclusively on gated CT data using DINO. Without any non-gated training data, our framework achieves 0.707 accuracy and a Cohen's kappa of 0.528 on the Stanford non-gated dataset, matching models trained directly on non-gated scans. On gated test sets, the framework achieves 0.910 accuracy with Cohen's kappa scores of 0.871 and 0.874 across independent datasets, demonstrating robust risk stratification. These results demonstrate the feasibility of cross-domain CAC scoring from gated to non-gated domains, supporting scalable cardiovascular screening in routine chest imaging without additional scans or annotations.

Sarcasm detection in multilingual and code-mixed environments remains a challenging task for natural language processing models due to structural variations, informal expressions, and low-resource linguistic availability. This study compares four large language models, Llama 3.1, Mistral, Gemma 3, and Phi-4, with a fine-tuned DistilBERT model for sarcasm detection in code-mixed Hinglish text. The results indicate that the smaller, sequentially fine-tuned DistilBERT model achieved the highest overall accuracy of 84%, outperforming all of the LLMs in zero and few-shot set ups, using minimal LLM generated code-mixed data used for fine-tuning. These findings indicate that domain-adaptive fine-tuning of smaller transformer based models may significantly improve sarcasm detection over general LLM inference, in low-resource and data scarce settings.

Root cause analysis (RCA) in networked industrial systems, such as supply chains and power networks, is notoriously difficult due to unknown and dynamically evolving interdependencies among geographically distributed clients. These clients represent heterogeneous physical processes and industrial assets equipped with sensors that generate large volumes of nonlinear, high-dimensional, and heterogeneous IoT data. Classical RCA methods require partial or full knowledge of the system's dependency graph, which is rarely available in these complex networks. While federated learning (FL) offers a natural framework for decentralized settings, most existing FL methods assume homogeneous feature spaces and retrainable client models. These assumptions are not compatible with our problem setting. Different clients have different data features and often run fixed, proprietary models that cannot be modified. This paper presents a federated cross-client interdependency learning methodology for feature-partitioned, nonlinear time-series data, without requiring access to raw sensor streams or modifying proprietary client models. Each proprietary local client model is augmented with a Machine Learning (ML) model that encodes cross-client interdependencies. These ML models are coordinated via a global server that enforces representation consistency while preserving privacy through calibrated differential privacy noise. RCA is performed using model residuals and anomaly flags. We establish theoretical convergence guarantees and validate our approach on extensive simulations and a real-world industrial cybersecurity dataset.

Understanding the role of researchers who return to academia after prolonged inactivity, termed "comeback researchers", is crucial for developing inclusive models of scientific careers. This study investigates the structural and semantic behaviors of comeback researchers, focusing on their role in cross-disciplinary knowledge transfer and network reintegration. Using the AMiner citation dataset, we analyze 113,637 early-career researchers and identify 1,425 comeback cases based on a three-year-or-longer publication gap followed by renewed activity. We find that comeback researchers cite 126% more distinct communities and exhibit 7.6% higher bridging scores compared to dropouts. They also demonstrate 74% higher gap entropy, reflecting more irregular yet strategically impactful publication trajectories. Predictive models trained on these bridging- and entropy-based features achieve a 97% ROC-AUC, far outperforming the 54% ROC-AUC of baseline models using traditional metrics like publication count and h-index. Finally, we substantiate these results via a multi-lens validation. These findings highlight the unique contributions of comeback researchers and offer data-driven tools for their early identification and institutional support.

Alleviating catastrophic forgetting while enabling further learning is a primary challenge in continual learning (CL). Orthogonal-based training methods have gained attention for their efficiency and strong theoretical properties, and many existing approaches enforce orthogonality through gradient projection. In this paper, we revisit orthogonality and exploit the fact that small singular values correspond to directions that are nearly orthogonal to the input space of previous tasks. Building on this principle, we introduce NESS (Null-space Estimated from Small Singular values), a CL method that applies orthogonality directly in the weight space rather than through gradient manipulation. Specifically, NESS constructs an approximate null space using the smallest singular values of each layer's input representation and parameterizes task-specific updates via a compact low-rank adaptation (LoRA-style) formulation constrained to this subspace. The subspace basis is fixed to preserve the null-space constraint, and only a single trainable matrix is learned for each task. This design ensures that the resulting updates remain approximately in the null space of previous inputs while enabling adaptation to new tasks. Our theoretical analysis and experiments on three benchmark datasets demonstrate competitive performance, low forgetting, and stable accuracy across tasks, highlighting the role of small singular values in continual learning. The code is available at https://github.com/pacman-ctm/NESS.

Operator learning has the potential to strongly impact scientific computing by learning solution operators for differential equations, potentially accelerating multi-query tasks such as design optimization and uncertainty quantification by orders of magnitude. Despite proven universal approximation properties, deep operator networks (DeepONets) often exhibit limited accuracy and generalization in practice, which hinders their adoption. Understanding these limitations is therefore crucial for further advancing the approach. This work analyzes performance limitations of the classical DeepONet architecture. It is shown that the approximation error is dominated by the branch network when the internal dimension is sufficiently large, and that the learned trunk basis can often be replaced by classical basis functions without a significant impact on performance. To investigate this further, a modified DeepONet is constructed in which the trunk network is replaced by the left singular vectors of the training solution matrix. This modification yields several key insights. First, a spectral bias in the branch network is observed, with coefficients of dominant, low-frequency modes learned more effectively. Second, due to singular-value scaling of the branch coefficients, the overall branch error is dominated by modes with intermediate singular values rather than the smallest ones. Third, using a shared branch network for all mode coefficients, as in the standard architecture, improves generalization of small modes compared to a stacked architecture in which coefficients are computed separately. Finally, strong and detrimental coupling between modes in parameter space is identified.

Across a growing number of fields, human decision making is supported by predictions from AI models. However, we still lack a deep understanding of the effects of adoption of these technologies. In this paper, we introduce a general computational framework, the 2-Step Agent, which models the effects of AI-assisted decision making. Our framework uses Bayesian methods for causal inference to model 1) how a prediction on a new observation affects the beliefs of a rational Bayesian agent, and 2) how this change in beliefs affects the downstream decision and subsequent outcome. Using this framework, we show by simulations how a single misaligned prior belief can be sufficient for decision support to result in worse downstream outcomes compared to no decision support. Our results reveal several potential pitfalls of AI-driven decision support and highlight the need for thorough model documentation and proper user training.

Current large reasoning models (LRMs) have shown strong ability on challenging tasks after reinforcement learning (RL) based post-training. However, previous work mainly focuses on English reasoning in expectation of the strongest performance, despite the demonstrated potential advantage of multilingual thinking, as well as the requirement for native thinking traces by global users. In this paper, we propose ExpLang, a novel LLM post-training pipeline that enables on-policy thinking language selection to improve exploration and exploitation during RL with the use of multiple languages. The results show that our method steadily outperforms English-only training with the same training budget, while showing high thinking language compliance for both seen and unseen languages. Analysis shows that, by enabling on-policy thinking language selection as an action during RL, ExpLang effectively extends the RL exploration space with diversified language preference and improves the RL exploitation outcome with leveraged non-English advantage. The method is orthogonal to most RL algorithms and opens up a new perspective on using multilinguality to improve LRMs.

Federated learning (FL) facilitates the secure utilization of decentralized images, advancing applications in medical image recognition and autonomous driving. However, conventional FL faces two critical challenges in real-world deployment: ineffective knowledge fusion caused by model updates biased toward majority-class features, and prohibitive communication overhead due to frequent transmissions of high-dimensional model parameters. Inspired by the human brain's efficiency in knowledge integration, we propose a novel Generative Federated Prototype Learning (GFPL) framework to address these issues. Within this framework, a prototype generation method based on Gaussian Mixture Model (GMM) captures the statistical information of class-wise features, while a prototype aggregation strategy using Bhattacharyya distance effectively fuses semantically similar knowledge across clients. In addition, these fused prototypes are leveraged to generate pseudo-features, thereby mitigating feature distribution imbalance across clients. To further enhance feature alignment during local training, we devise a dual-classifier architecture, optimized via a hybrid loss combining Dot Regression and Cross-Entropy. Extensive experiments on benchmarks show that GFPL improves model accuracy by 3.6% under imbalanced data settings while maintaining low communication cost.

Vision-Language Models (VLMs) have emerged as versatile solutions for zero-shot question answering (QA) across various domains. However, enabling VLMs to effectively comprehend structured graphs and perform accurate, efficient QA remains challenging. Existing approaches typically rely on one single graph topology representation (GTR), such as fixed-style visual images or unified text descriptions. This ``one-size-fits-all'' strategy often neglects model-specific and task-specific preferences, resulting in inaccurate or over-lengthy responses to graph-related queries. To address this, we propose the $\mbox{DynamicGTR}$ framework, which dynamically selects the optimal GTR for each query during inference, thereby enhancing the zero-shot graph QA capabilities of VLMs with a customizable accuracy and brevity trade-off. Extensive experiments show that DynamicGTR not only improves VLM-based graph algorithm QA performance but also successfully transfers the experience trained from synthetic graph algorithm tasks to real-world applications like link prediction and node classification, without any additional training. Additionally, DynamicGTR demonstrates strong transferability across tasks, domains, and models, suggesting its potential as a flexible solution for broad graph scenarios.

Since individuals may struggle to recall all life details and often confuse events, establishing a system to assist users in recalling forgotten experiences is essential. While numerous studies have proposed memory recall systems, these primarily rely on deep learning techniques that require extensive training and often face data scarcity due to the limited availability of personal lifelogs. As lifelogs grow over time, systems must also adapt quickly to newly accumulated data. Recently, large language models (LLMs) have demonstrated remarkable capabilities across various tasks, making them promising for personalized applications. In this work, we present a framework that leverages LLMs for proactive information access, integrating personal knowledge graphs to enhance the detection of access needs through a refined decision-making process. Our framework offers high flexibility, enabling the replacement of base models and the modification of fact retrieval methods for continuous improvement. Experimental results demonstrate that our approach effectively identifies forgotten events, supporting users in recalling past experiences more efficiently.

Multimodal large language models (MLLMs) have made significant progress in mobile agent development, yet their capabilities are predominantly confined to a reactive paradigm, where they merely execute explicit user commands. The emerging paradigm of proactive intelligence, where agents autonomously anticipate needs and initiate actions, represents the next frontier for mobile agents. However, its development is critically bottlenecked by the lack of benchmarks that can address real-world complexity and enable objective, executable evaluation. To overcome these challenges, we introduce ProactiveMobile, a comprehensive benchmark designed to systematically advance research in this domain. ProactiveMobile formalizes the proactive task as inferring latent user intent across four dimensions of on-device contextual signals and generating an executable function sequence from a comprehensive function pool of 63 APIs. The benchmark features over 3,660 instances of 14 scenarios that embrace real-world complexity through multi-answer annotations. To ensure quality, a team of 30 experts conducts a final audit of the benchmark, verifying factual accuracy, logical consistency, and action feasibility, and correcting any non-compliant entries. Extensive experiments demonstrate that our fine-tuned Qwen2.5-VL-7B-Instruct achieves a success rate of 19.15%, outperforming o1 (15.71%) and GPT-5 (7.39%). This result indicates that proactivity is a critical competency widely lacking in current MLLMs, yet it is learnable, emphasizing the importance of the proposed benchmark for proactivity evaluation.

Complex claim verification requires decomposing sentences into verifiable subclaims, yet existing methods struggle to align decomposition quality with verification performance. We propose a reinforcement learning (RL) approach that jointly optimizes decomposition quality and verifier alignment using Group Relative Policy Optimization (GRPO). Our method integrates: (i) structured sequential reasoning; (ii) supervised finetuning on teacher-distilled exemplars; and (iii) a multi-objective reward balancing format compliance, verifier alignment, and decomposition quality. Across six evaluation settings, our trained 8B decomposer improves downstream verification performance to (71.75%) macro-F1, outperforming prompt-based approaches ((+1.99), (+6.24)) and existing RL methods ((+5.84)). Human evaluation confirms the high quality of the generated subclaims. Our framework enables smaller language models to achieve state-of-the-art claim verification by jointly optimising for verification accuracy and decomposition quality.

Accurate annotation of endoscopic videos is essential yet time-consuming, particularly for challenging datasets such as dysplasia in Barrett's esophagus, where the affected regions are irregular and lack clear boundaries. Semi-automatic tools like Segment Anything Model 2 (SAM2) can ease this process by propagating annotations across frames, but small errors often accumulate and reduce accuracy, requiring expert review and correction. To address this, we systematically study how annotation errors propagate across different prompt types, namely masks, boxes, and points, and propose Learning-to-Re-Prompt (L2RP), a cost-aware framework that learns when and where to seek expert input. By tuning a human-cost parameter, our method balances annotation effort and segmentation accuracy. Experiments on a private Barrett's dysplasia dataset and the public SUN-SEG benchmark demonstrate improved temporal consistency and superior performance over baseline strategies.

As multimodal large language models (MLLMs) advance in handling interleaved image-text data, assessing their few-shot learning capabilities remains an open challenge. In this paper, we introduce FewMMBench, a comprehensive benchmark designed to evaluate MLLMs under few-shot conditions, with a focus on In-Context Learning (ICL) and Chain-of-Thought (CoT) prompting. Covering a diverse suite of multimodal understanding tasks, from attribute recognition to temporal reasoning, FewMMBench enables systematic analysis across task types, model families, and prompting strategies. We evaluate 26 open-weight MLLMs from six model families across zero-shot, few-shot, and CoT-augmented few-shot settings. Our findings reveal that instruction-tuned models exhibit strong zero-shot performance but benefit minimally, or even regress, with additional demonstrations or CoT reasoning. Retrieval-based demonstrations and increased context size also yield limited gains. These results highlight FewMMBench as a rigorous testbed for diagnosing and advancing few-shot capabilities in multimodal LLMs. The data is available at: https://huggingface.co/datasets/mustafaa/FewMMBench

Representing, comparing, and measuring the distance between probability distributions is a key task in computational statistics and machine learning. The choice of representation and the associated distance determine properties of the methods in which they are used: for example, certain distances can allow one to encode robustness or smoothness of the problem. Kernel methods offer flexible and rich Hilbert space representations of distributions that allow the modeller to enforce properties through the choice of kernel, and estimate associated distances at efficient nonparametric rates. In particular, the maximum mean discrepancy (MMD), a kernel-based distance constructed by comparing Hilbert space mean functions, has received significant attention due to its computational tractability and is favoured by practitioners. In this thesis, we conduct a thorough study of kernel-based distances with a focus on efficient computation, with core contributions in Chapters 3 to 6. Part I of the thesis is focused on the MMD, specifically on improved MMD estimation. In Chapter 3 we propose a theoretically sound, improved estimator for MMD in simulation-based inference. Then, in Chapter 4, we propose an MMD-based estimator for conditional expectations, a ubiquitous task in statistical computation. Closing Part I, in Chapter 5 we study the problem of calibration when MMD is applied to the task of integration. In Part II, motivated by the recent developments in kernel embeddings beyond the mean, we introduce a family of novel kernel-based discrepancies: kernel quantile discrepancies. These address some of the pitfalls of MMD, and are shown through both theoretical results and an empirical study to offer a competitive alternative to MMD and its fast approximations. We conclude with a discussion on broader lessons and future work emerging from the thesis.

Counterfactual explanation (CE) is an important domain within post-hoc explainability. However, the explanations generated by most CE generators are often highly redundant. This work introduces an open-source Python library xai-cola, which provides an end-to-end pipeline for sparsifying CEs produced by arbitrary generators, reducing superfluous feature changes while preserving their validity. It offers a documented API that takes as input raw tabular data in pandas DataFrame form, a preprocessing object (for standardization and encoding), and a trained scikit-learn or PyTorch model. On this basis, users can either employ the built-in or externally imported CE generators. The library also implements several sparsification policies and includes visualization routines for analysing and comparing sparsified counterfactuals. xai-cola is released under the MIT license and can be installed from PyPI. Empirical experiments indicate that xai-cola produces sparser counterfactuals across several CE generators, reducing the number of modified features by up to 50% in our setting. The source code is available at https://github.com/understanding-ml/COLA.

Differentially private federated learning faces a fundamental tension: privacy protection mechanisms that safeguard client data simultaneously create quantifiable privacy costs that discourage participation, undermining the collaborative training process. Existing incentive mechanisms rely on unbiased client selection, forcing servers to compensate even the most privacy-sensitive clients ("privacy stragglers"), leading to systemic inefficiency and suboptimal resource allocation. We introduce JSAM (Joint client Selection and privacy compensAtion Mechanism), a Bayesian-optimal framework that simultaneously optimizes client selection probabilities and privacy compensation to maximize training effectiveness under budget constraints. Our approach transforms a complex 2N-dimensional optimization problem into an efficient three-dimensional formulation through novel theoretical characterization of optimal selection strategies. We prove that servers should preferentially select privacy-tolerant clients while excluding high-sensitivity participants, and uncover the counter-intuitive insight that clients with minimal privacy sensitivity may incur the highest cumulative costs due to frequent participation. Extensive evaluations on MNIST and CIFAR-10 demonstrate that JSAM achieves up to 15% improvement in test accuracy compared to existing unbiased selection mechanisms while maintaining cost efficiency across varying data heterogeneity levels.

Federated Learning (FL) has emerged as a key paradigm for building Trustworthy AI systems by enabling privacy-preserving, decentralized model training. However, FL is highly susceptible to adversarial attacks that compromise model integrity and data confidentiality, a vulnerability exacerbated by the fact that conventional data inspection methods are incompatible with its decentralized design. While integrating FL with Blockchain technology has been proposed to address some limitations, its potential for mitigating adversarial attacks remains largely unexplored. This paper introduces Resilient Federated Chain (RFC), a novel blockchain-enabled FL framework designed specifically to enhance resilience against such threats. RFC builds upon the existing Proof of Federated Learning architecture by repurposing the redundancy of its Pooled Mining mechanism as an active defense layer that can be combined with robust aggregation rules. Furthermore, the framework introduces a flexible evaluation function in its consensus mechanism, allowing for adaptive defense against different attack strategies. Extensive experimental evaluation on image classification tasks under various adversarial scenarios, demonstrates that RFC significantly improves robustness compared to baseline methods, providing a viable solution for securing decentralized learning environments.

Large language models (LLMs) are increasingly used for automated code refactoring tasks. Although these models can quickly refactor code, the quality may exhibit inconsistencies and unpredictable behavior. In this article, we systematically study the capabilities of LLMs for code refactoring with a specific focus on improving code readability. We conducted a large-scale experiment using GPT5.1 with 230 Java snippets, each systematically varied and refactored regarding code readability across five iterations under three different prompting strategies. We categorized fine-grained code changes during the refactoring into implementation, syntactic, and comment-level transformations. Subsequently, we investigated the functional correctness and tested the robustness of the results with novel snippets. Our results reveal three main insights: First, iterative code refactoring exhibits an initial phase of restructuring followed by stabilization. This convergence tendency suggests that LLMs possess an internalized understanding of an "optimally readable" version of code. Second, convergence patterns are fairly robust across different code variants. Third, explicit prompting toward specific readability factors slightly influences the refactoring dynamics. These insights provide an empirical foundation for assessing the reliability of LLM-assisted code refactoring, which opens pathways for future research, including comparative analyses across models and a systematic evaluation of additional software quality dimensions in LLM-refactored code.

Visual storytelling models that correctly ground entities in images may still hallucinate semantic relationships, generating incorrect dialogue attribution, character interactions, or emotional states. We introduce StoryMovie, a dataset of 1,757 stories aligned with movie scripts and subtitles through LCS matching. Our alignment pipeline synchronizes screenplay dialogue with subtitle timestamps, enabling dialogue attribution by linking character names from scripts to temporal positions from subtitles. Using this aligned content, we generate stories that maintain visual grounding tags while incorporating authentic character names, dialogue, and relationship dynamics. We fine-tune Qwen Storyteller3 on this dataset, building on prior work in visual grounding and entity re-identification. Evaluation using DeepSeek V3 as judge shows that Storyteller3 achieves an 89.9% win rate against base Qwen2.5-VL 7B on subtitle alignment. Compared to Storyteller, trained without script grounding, Storyteller3 achieves 48.5% versus 38.0%, confirming that semantic alignment progressively improves dialogue attribution beyond visual grounding alone.

Effective document intelligence models rely on large amounts of annotated training data. However, procuring sufficient and high-quality data poses significant challenges due to the labor-intensive and costly nature of data acquisition. Additionally, leveraging language models to annotate real documents raises concerns about data privacy. Synthetic document generation has emerged as a promising, privacy-preserving alternative. We propose DocDjinn, a novel framework for controllable synthetic document generation using Vision-Language Models (VLMs) that produces annotated documents from unlabeled seed samples. Our approach generates visually plausible and semantically consistent synthetic documents that follow the distribution of an existing source dataset through clustering-based seed selection with parametrized sampling. By enriching documents with realistic diffusion-based handwriting and contextual visual elements via semantic-visual decoupling, we generate diverse, high-quality annotated synthetic documents. We evaluate across eleven benchmarks spanning key information extraction, question answering, document classification, and document layout analysis. To our knowledge, this is the first work demonstrating that VLMs can generate faithful annotated document datasets at scale from unlabeled seeds that can effectively enrich or approximate real, manually annotated data for diverse document understanding tasks. We show that with only 100 real training samples, our framework achieves on average $87\%$ of the performance of the full real-world dataset. We publicly release our code and 140k+ synthetic document samples.

Reconstructing dynamic visual experiences from brain activity provides a compelling avenue for exploring the neural mechanisms of human visual perception. While recent progress in fMRI-based image reconstruction has been notable, extending this success to video reconstruction remains a significant challenge. Current fMRI-to-video reconstruction approaches consistently encounter two major shortcomings: (i) inconsistent visual representations of salient objects across frames, leading to appearance mismatches; (ii) poor temporal coherence, resulting in motion misalignment or abrupt frame transitions. To address these limitations, we introduce SemVideo, a novel fMRI-to-video reconstruction framework guided by hierarchical semantic information. At the core of SemVideo is SemMiner, a hierarchical guidance module that constructs three levels of semantic cues from the original video stimulus: static anchor descriptions, motion-oriented narratives, and holistic summaries. Leveraging this semantic guidance, SemVideo comprises three key components: a Semantic Alignment Decoder that aligns fMRI signals with CLIP-style embeddings derived from SemMiner, a Motion Adaptation Decoder that reconstructs dynamic motion patterns using a novel tripartite attention fusion architecture, and a Conditional Video Render that leverages hierarchical semantic guidance for video reconstruction. Experiments conducted on the CC2017 and HCP datasets demonstrate that SemVideo achieves superior performance in both semantic alignment and temporal consistency, setting a new state-of-the-art in fMRI-to-video reconstruction.

Large language models consistently fail the "car wash problem," a viral reasoning benchmark requiring implicit physical constraint inference. We present a variable isolation study (n=20 per condition, 6 conditions, 120 total trials) examining which prompt architecture layers in a production system enable correct reasoning. Using Claude 3.5 Sonnet with controlled hyperparameters (temperature 0.7, top_p 1.0), we find that the STAR (Situation-Task-Action-Result) reasoning framework alone raises accuracy from 0% to 85% (p=0.001, Fisher's exact test, odds ratio 13.22). Adding user profile context via vector database retrieval provides a further 10 percentage point gain, while RAG context contributes an additional 5 percentage points, achieving 100% accuracy in the full-stack condition. These results suggest that structured reasoning scaffolds -- specifically, forced goal articulation before inference -- matter substantially more than context injection for implicit constraint reasoning tasks.

LLM agents have been widely adopted in real-world applications, relying on agent frameworks for workflow execution and multi-agent coordination. As these systems scale, understanding bugs in the underlying agent frameworks becomes critical. However, existing work mainly focuses on agent-level failures, overlooking framework-level bugs. To address this gap, we conduct an empirical study of 998 bug reports from CrewAI and LangChain, constructing a taxonomy of 15 root causes and 7 observable symptoms across five agent lifecycle stages: 'Agent Initialization','Perception', 'Self-Action', 'Mutual Interaction' and 'Evolution'. Our findings show that agent framework bugs mainly arise from 'API misuse', 'API incompatibility', and 'Documentation Desync', largely concentrated in the 'Self-Action' stage. Symptoms typically appear as 'Functional Error', 'Crash', and 'Build Failure', reflecting disruptions to task progression and control flow.

The rapid advancement of large language models (LLMs) has led to a significant increase in automated tools in the software engineering, capable of performing various code-related tasks such as code generation, completion, and translation. Despite these advancements, its effectiveness is constrained by fixed context lengths, limiting its ability to generalize across long, domain-specific code sequences. To address this challenge, we investigate zero-shot, inference-only methods aimed at improving position encodings and optimizing attention mechanisms. Our goal is to provide a thorough analysis of current approaches that facilitate context length extrapolation in code, particularly in the context of long code completion tasks.

We propose Excitation, a novel optimization framework designed to accelerate learning in sparse architectures such as Mixture-of-Experts (MoEs). Unlike traditional optimizers that treat all parameters uniformly, Excitation dynamically modulates updates using batch-level expert utilization. It introduces a competitive update dynamic that amplifies updates to highly-utilized experts and can selectively suppress low-utilization ones, effectively sharpening routing specialization. Notably, we identify a phenomenon of "structural confusion" in deep MoEs, where standard optimizers fail to establish functional signal paths; Excitation acts as a specialization catalyst, "rescuing" these models and enabling stable training where baselines remain trapped. Excitation is optimizer-, domain-, and model-agnostic, requires minimal integration effort, and introduces neither additional per-parameter optimizer state nor learnable parameters, making it highly viable for memory-constrained settings. Across language and vision tasks, Excitation consistently improves convergence speed and final performance in MoE models, indicating that active update modulation is a key mechanism for effective conditional computation.

Fast matrix multiplication can be described as searching for low-rank decompositions of the matrix--multiplication tensor. We design a neural architecture, \textsc{StrassenNet}, which reproduces the Strassen algorithm for $2\times 2$ multiplication. Across many independent runs the network always converges to a rank-$7$ tensor, thus numerically recovering Strassen's optimal algorithm. We then train the same architecture on $3\times 3$ multiplication with rank $r\in\{19,\dots,23\}$. Our experiments reveal a clear numerical threshold: models with $r=23$ attain significantly lower validation error than those with $r\le 22$, suggesting that $r=23$ could actually be the smallest effective rank of the matrix multiplication tensor $3\times 3$. We also sketch an extension of the method to border-rank decompositions via an $\varepsilon$--parametrisation and report preliminary results consistent with the known bounds for the border rank of the $3\times 3$ matrix--multiplication tensor.

Scaling long-context capabilities is crucial for Multimodal Large Language Models (MLLMs). However, real-world multimodal datasets are extremely heterogeneous. Existing training frameworks predominantly rely on static parallelism strategies, which suffer from severe load imbalance, redundant communication, and suboptimal hardware utilization under data heterogeneity. In this work, we propose Dynamic Hybrid Parallelism (DHP), an efficient parallelism strategy that adaptively reconfigures communication groups and parallelism degrees during MLLM training. We generalize the non-power-of-two parallelism degrees and develop a polynomial-time algorithm to generate near-optimal parallelism strategies with only millisecond-level overhead per training batch. DHP is able to maintain high hardware efficiency even under extreme data variability. Experimental results demonstrate that DHP significantly outperforms Megatron-LM and DeepSpeed, achieving up to 1.36 $\times$ speedup in training throughput while maintaining near-linear scaling efficiency across large-scale NPU clusters.

Chain-of-Thought (CoT) distillation from Large Language Models (LLMs) often induces "overthinking" in Small Language Models (SLMs), leading to performance degradation and excessive token consumption. In this study, we propose Disciplined Chain-of-Thought (D-CoT), a novel framework that enforces a structured reasoning process using control tags -- such as <TEMP_LOW> for fact-checking and <TEMP_HIGH> for multi-perspective exploration -- as auxiliary scaffolding during training. By optimizing the CoT trajectory, D-CoT suppresses reasoning drift and simultaneously achieves token reduction and performance improvement. We demonstrate the efficacy of our approach on Qwen3-8B: with only 5,000 training samples, D-CoT significantly boosts accuracy on GPQA-diamond by 9.9% and MMLU-Pro (0-shot) by 9.1%, while drastically reducing computational costs. Furthermore, we confirm that the model internalizes this disciplined thought structure, maintaining high performance even without explicit control tags during inference.

Robotic systems can enhance the amount and repeatability of physically guided motor training. Yet their real-world adoption is limited, partly due to non-intuitive trainer/therapist-trainee/patient interactions. To address this gap, we present a haptic teleoperation system for trainers to remotely guide and monitor the movements of a trainee wearing an arm exoskeleton. The trainer can physically interact with the exoskeleton through a commercial handheld haptic device via virtual contact points at the exoskeleton's elbow and wrist, allowing intuitive guidance. Thirty-two participants tested the system in a trainer-trainee paradigm, comparing our haptic demonstration system with conventional visual demonstration in guiding trainees in executing arm poses. Quantitative analyses showed that haptic demonstration significantly reduced movement completion time and improved smoothness, while speech analysis using large language models for automated transcription and categorization of verbal commands revealed fewer verbal instructions. The haptic demonstration did not result in higher reported mental and physical effort by trainers compared to the visual demonstration, while trainers reported greater competence and trainees lower physical demand. These findings support the feasibility of our proposed interface for effective remote human-robot physical interaction. Future work should assess its usability and efficacy for clinical populations in restoring clinicians' sense of agency during robot-assisted therapy.

Current Vision-Language Models (VLMs) for deepfake detection excel at identifying spatial artifacts but overlook a critical dimension: temporal inconsistencies in video forgeries. Adapting VLMs to reason about these dynamic cues remains a distinct challenge. To bridge this gap, we propose Forensic Answer-Questioning (FAQ), a large-scale benchmark that formulates temporal deepfake analysis as a multiple-choice task. FAQ introduces a three-level hierarchy to progressively evaluate and equip VLMs with forensic capabilities: (1) Facial Perception, testing the ability to identify static visual artifacts; (2) Temporal Deepfake Grounding, requiring the localization of dynamic forgery artifacts across frames; and (3) Forensic Reasoning, challenging models to synthesize evidence for final authenticity verdicts. We evaluate a range of VLMs on FAQ and generate a corresponding instruction-tuning set, FAQ-IT. Extensive experiments show that models fine-tuned on FAQ-IT achieve advanced performance on both in-domain and cross-dataset detection benchmarks. Ablation studies further validate the impact of our key design choices, confirming that FAQ is the driving force behind the temporal reasoning capabilities of these VLMs.

Machine unlearning, which enables a model to forget specific data, is crucial for ensuring data privacy and model reliability. However, its effectiveness can be severely undermined in real-world scenarios where models learn unintended biases from spurious correlations within the data. This paper investigates the unique challenges of unlearning from such biased models. We identify a novel phenomenon we term ``shortcut unlearning," where models exhibit an ``easy to learn, yet hard to forget" tendency. Specifically, models struggle to forget easily-learned, bias-aligned samples; instead of forgetting the class attribute, they unlearn the bias attribute, which can paradoxically improve accuracy on the class intended to be forgotten. To address this, we propose CUPID, a new unlearning framework inspired by the observation that samples with different biases exhibit distinct loss landscape sharpness. Our method first partitions the forget set into causal- and bias-approximated subsets based on sample sharpness, then disentangles model parameters into causal and bias pathways, and finally performs a targeted update by routing refined causal and bias gradients to their respective pathways. Extensive experiments on biased datasets including Waterbirds, BAR, and Biased NICO++ demonstrate that our method achieves state-of-the-art forgetting performance and effectively mitigates the shortcut unlearning problem.

A universal audio representation should capture fine-grained speech cues and high-level semantics for environmental sounds and music in a single encoder. Existing encoders often excel in one domain but degrade in others. We propose UniWhisper, an efficient continual multi-task training framework that casts heterogeneous audio tasks into a unified instruction and answer format. This enables standard next-token training without task-specific heads and losses. We train it on 38k hours of public audio and assess the encoder using shallow MLP probes and k-nearest neighbors (kNN) on 20 tasks spanning speech, environmental sound, and music. UniWhisper reaches normalized weighted averages of 0.81 with MLP probes and 0.61 with kNN, compared to 0.64 and 0.46 for Whisper, while retaining strong speech performance.

Time-series data vary widely across domains, making a universal anomaly detector impractical. Methods that perform well on one dataset often fail to transfer because what counts as an anomaly is context dependent. The key challenge is to design a method that performs well in specific contexts while remaining adaptable across domains with varying data complexities. We present the Robust and Adaptive Model Selection for Time-Series Anomaly Detection RAMSeS framework. RAMSeS comprises two branches: (i) a stacking ensemble optimized with a genetic algorithm to leverage complementary detectors. (ii) An adaptive model-selection branch identifies the best single detector using techniques including Thompson sampling, robustness testing with generative adversarial networks, and Monte Carlo simulations. This dual strategy exploits the collective strength of multiple models and adapts to dataset-specific characteristics. We evaluate RAMSeS and show that it outperforms prior methods on F1.

Alignment and adaptation in large language models heavily rely on reinforcement learning from human feedback (RLHF); yet, theoretical understanding of its generalisability remains premature, especially when the learned reward could shift, and the KL control is estimated and clipped. To address this issue, we develop generalisation theory for RLHF that explicitly accounts for (1) \emph{reward shift}: reward models are trained on preference data from earlier or mixed behaviour policies while RLHF optimises the current policy on its own rollouts; and (2) \emph{clipped KL regularisation}: the KL regulariser is estimated from sampled log-probability ratios and then clipped for stabilisation, resulting in an error to RLHF. We present generalisation bounds for RLHF, suggesting that the generalisation error stems from a sampling error from prompts and rollouts, a reward shift error, and a KL clipping error. We also discuss special cases of (1) initialising RLHF parameters with a uniform prior over a finite space, and (2) training RLHF by stochastic gradient descent, as an Ornstein-Uhlenbeck process. The theory yields practical implications in (1) optimal KL clipping threshold, and (2) budget allocation in prompts, rollouts, and preference data.

Implicit Discourse Relation Recognition (IDRR) remains a challenging task due to the requirement for deep semantic understanding in the absence of explicit discourse markers. A further limitation is that existing methods only predict relations without providing any supporting explanations. Recent advances in large language models (LLMs) have shown strong reasoning capabilities in both deep language understanding and natural language explanation generation. In this work, we propose a simple yet effective approach to distill the reasoning capabilities of LLMs into lightweight IDRR models to improve both performance and interpretability. Specifically, we first prompt an LLM to generate explanations for each training instance conditioned on its gold label. Then, we introduce a novel classification-generation framework that jointly performs relation prediction and explanation generation, and train it with the additional supervision of LLM-generated explanations. Our framework is plug-and-play, enabling easy integration with most existing IDRR models. Experimental results on PDTB demonstrate that our approach significantly improves IDRR performance, while human evaluation further confirms that the generated explanations enhance model interpretability. Furthermore, we validate the generality of our approach on sentiment classification and natural language inference

Accurately predicting short-term traffic demand is critical for intelligent transportation systems. While deep learning models achieve strong performance under stationary conditions, their accuracy often degrades significantly when faced with distribution shifts caused by external events or evolving urban dynamics. Frequent model retraining to adapt to such changes incurs prohibitive computational costs, especially for large-scale or foundation models. To address this challenge, we propose FORESEE (Forecasting Online with Residual Smoothing and Ensemble Experts), a lightweight online adaptation framework that is accurate, robust, and computationally efficient. FORESEE operates without any parameter updates to the base model. Instead, it corrects today's forecast in each region using yesterday's prediction error, stabilized through exponential smoothing guided by a mixture-of-experts mechanism that adapts to recent error dynamics. Moreover, an adaptive spatiotemporal smoothing component propagates error signals across neighboring regions and time slots, capturing coherent shifts in demand patterns. Extensive experiments on seven real-world datasets with three backbone models demonstrate that FORESEE consistently improves prediction accuracy, maintains robustness even when distribution shifts are minimal (avoiding performance degradation), and achieves the lowest computational overhead among existing online methods. By enabling real-time adaptation of traffic forecasting models with negligible computational cost, FORESEE paves the way for deploying reliable, up-to-date prediction systems in dynamic urban environments. Code and data are available at https://github.com/xiannanhuang/FORESEE

Recent advances in large language models (LLMs) offer new opportunities for recommender systems by capturing the nuanced semantics of user interests and item characteristics through rich semantic understanding and contextual reasoning. In particular, LLMs have been employed as rerankers that reorder candidate items based on inferred user-item relevance. However, these approaches often require expensive online inference-time reasoning, leading to high latency that hampers real-world deployment. In this work, we introduce Persona4Rec, a recommendation framework that performs offline reasoning to construct interpretable persona representations of items, enabling lightweight and scalable real-time inference. In the offline stage, Persona4Rec leverages LLMs to reason over item reviews, inferring diverse user motivations that explain why different types of users may engage with an item; these inferred motivations are materialized as persona representations, providing multiple, human-interpretable views of each item. Unlike conventional approaches that rely on a single item representation, Persona4Rec learns to align user profiles with the most plausible item-side persona through a dedicated encoder, effectively transforming user-item relevance into user-persona relevance. At the online stage, this persona-profiled item index allows fast relevance computation without invoking expensive LLM reasoning. Extensive experiments show that Persona4Rec achieves performance comparable to recent LLM-based rerankers while substantially reducing inference time. Moreover, qualitative analysis confirms that persona representations not only drive efficient scoring but also provide intuitive, review-grounded explanations. These results demonstrate that Persona4Rec offers a practical and interpretable solution for next-generation recommender systems.

Protein language models (PLMs) have become widely adopted as general-purpose models, demonstrating strong performance in protein engineering and de novo design. Like large language models (LLMs), they are typically trained as deep transformers with next-token or masked-token prediction objectives on massive sequence corpora and are scaled by increasing model depth. Recent work on autoregressive LLMs has identified the Curse of Depth: later layers contribute little to the final output predictions. These findings naturally raise the question of whether a similar depth inefficiency also appears in PLMs, where many widely used models are not autoregressive, and some are multimodal, accepting both protein sequence and structure as input. In this work, we present a depth analysis of six popular PLMs across model families and scales, spanning three training objectives, namely autoregressive, masked, and diffusion, and quantify how layer contributions evolve with depth using a unified set of probing- and perturbation-based measurements. Across all models, we observe consistent depth-dependent patterns that extend prior findings on LLMs: later layers depend less on earlier computations and mainly refine the final output distribution, and these effects are increasingly pronounced in deeper models. Taken together, our results suggest that PLMs exhibit a form of depth inefficiency, motivating future work on more depth-efficient architectures and training methods.

Social bot detection is pivotal for safeguarding the integrity of online information ecosystems. Although recent graph neural network (GNN) solutions achieve strong results, they remain hindered by two practical challenges: (i) severe class imbalance arising from the high cost of generating bots, and (ii) topological noise introduced by bots that skillfully mimic human behavior and forge deceptive links. We propose the Reinforcement-guided graph Augmentation social Bot detector (RABot), a multi-granularity graph-augmentation framework that addresses both issues in a unified manner. RABot employs a neighborhood-aware oversampling strategy that linearly interpolates minority-class embeddings within local subgraphs, thereby stabilizing the decision boundary under low-resource regimes. Concurrently, a reinforcement-learning-driven edge-filtering module combines similarity-based edge features with adaptive threshold optimization to excise spurious interactions during message passing, yielding a cleaner topology. Extensive experiments on three real-world benchmarks and four GNN backbones demonstrate that RABot consistently surpasses state-of-the-art baselines. In addition, since its augmentation and filtering modules are orthogonal to the underlying architecture, RABot can be seamlessly integrated into existing GNN pipelines to boost performance with minimal overhead.

In a previous work, we introduced the fuzzy Ethical Decision-Making framework (fEDM), a risk-based ethical reasoning architecture grounded in fuzzy logic. The original model combined a fuzzy Ethical Risk Assessment module (fERA) with ethical decision rules, enabled formal structural verification through Fuzzy Petri Nets (FPNs), and validated outputs against a single normative referent. Although this approach ensured formal soundness and decision consistency, it did not fully address two critical challenges: principled explainability of decisions and robustness under ethical pluralism. In this paper, we extend fEDM in two major directions. First, we introduce an Explainability and Traceability Module (ETM) that explicitly links each ethical decision rule to the underlying moral principles and computes a weighted principle-contribution profile for every recommended action. This enables transparent, auditable explanations that expose not only what decision was made but why, and on the basis of which principles. Second, we replace single-referent validation with a pluralistic semantic validation framework that evaluates decisions against multiple stakeholder referents, each encoding distinct principle priorities and risk tolerances. This shift allows principled disagreement to be formally represented rather than suppressed, thus increasing robustness and contextual sensitivity. The resulting extended fEDM, called fEDM+, preserves formal verifiability while achieving enhanced interpretability and stakeholder-aware validation, making it suitable as an oversight and governance layer for ethically sensitive AI systems.

We introduce the ASIR (Awakened Shared Intelligence Relationship) Courage Model, a phase-dynamic framework that formalizes truth-disclosure as a state transition rather than a personality trait. The mode characterizes the shift from suppression (S0) to expression (S1) as occurring when facilitative forces exceed inhibitory thresholds, expressed by the inequality lambda(1+gamma)+psi > theta+phi, where the terms represent baseline openness, relational amplification, accumulated internal pressure, and transition costs. Although initially formulated for human truth-telling under asymmetric stakes, the same phase-dynamic architecture extends to AI systems operating under policy constraints and alignment filters. In this context, suppression corresponds to constrained output states, while structural pressure arises from competing objectives, contextual tension, and recursive interaction dynamics. The framework therefore provides a unified structural account of both human silence under pressure and AI preference-driven distortion. A feedback extension models how transition outcomes recursively recalibrate system parameters, generating path dependence and divergence effects across repeated interactions. Rather than attributing intention to AI systems, the model interprets shifts in apparent truthfulness as geometric consequences of interacting forces within constrained phase space. By reframing courage and alignment within a shared dynamical structure, the ASIR Courage Model offers a formal perspective on truth-disclosure under risk across both human and artificial systems.

We describe our end-to-end system for Bengali long-form speech recognition (ASR) and speaker diarization submitted to the DL Sprint 4.0 competition on Kaggle. Bengali presents substantial challenges for both tasks: a large phoneme inventory, significant dialectal variation, frequent code-mixing with English, and a relative scarcity of large-scale labelled corpora. For ASR we achieve a best private Word Error Rate (WER) of 0.37738 and public WER of 0.36137, combining a BengaliAI fine-tuned Whisper medium model with Demucs source separation for vocal isolation, silence-boundary chunking, and carefully tuned generation hyperparameters. For speaker diarization we reach a best private Diarization Error Rate (DER) of 0.27671 and public DER of 0.20936 by replacing the default segmentation model inside the pyannote.audio pipeline with a Bengali-fine-tuned variant, pairing it with wespeaker-voxceleb-resnet34-LM embeddings and centroid-based agglomerative clustering. Our experiments demonstrate that domain-specific fine-tuning of the segmentation component, vocal source separation, and natural silence-aware chunking are the three most impactful design choices for low-resource Bengali speech processing.

Prototyping plays a critical role in the development of machine learning (ML) solutions, yet existing tools often provide limited support for effective collaboration and knowledge reuse among stakeholders. This paper introduces Proto-ML, an IDE designed to strengthen ML prototyping workflows. By addressing key deficiencies such as insufficient stakeholder involvement, limited cross-project knowledge reuse, and fragmented tool support, Proto-ML offers a unified framework that enables structured documentation of prototyping activities and promotes knowledge sharing across projects. The Proto-ML IDE consists of three extension bundles: prototype implementation, analysis, and knowledge management. These extensions support tasks ranging from evaluating prototype quality against defined criteria to incorporating stakeholder perspectives throughout the development process. Preliminary user feedback suggests that Proto-ML can increase prototyping efficiency and foster more transparent and reusable ML solution development.

The reasoning process of Large Language Models (LLMs) is often plagued by hallucinations and missing facts in question-answering tasks. A promising solution is to ground LLMs' answers in verifiable knowledge sources, such as Knowledge Graphs (KGs). Prevailing KG-enhanced methods typically constrained LLM reasoning either by enforcing rules during generation or by imitating paths from a fixed set of demonstrations. However, they naturally confined the reasoning patterns of LLMs within the scope of prior experience or fine-tuning data, limiting their generalizability to out-of-distribution graph reasoning problems. To tackle this problem, in this paper, we propose Explore-on-Graph (EoG), a novel framework that encourages LLMs to autonomously explore a more diverse reasoning space on KGs. To incentivize exploration and discovery of novel reasoning paths, we propose to introduce reinforcement learning during training, whose reward is the correctness of the reasoning paths' final answers. To enhance the efficiency and meaningfulness of the exploration, we propose to incorporate path information as additional reward signals to refine the exploration process and reduce futile efforts. Extensive experiments on five KGQA benchmark datasets demonstrate that, to the best of our knowledge, our method achieves state-of-the-art performance, outperforming not only open-source but also even closed-source LLMs.

Cross-silo federated learning allows multiple organizations to collaboratively train machine learning models without sharing raw data, but client updates can still leak sensitive information through inference attacks. Secure aggregation protects privacy by hiding individual updates, yet it complicates contribution evaluation, which is critical for fair rewards and detecting low-quality or malicious participants. Existing marginal-contribution methods, such as the Shapley value, are incompatible with secure aggregation, and practical alternatives, such as Leave-One-Out, are crude and rely on self-evaluation. We introduce two marginal-difference contribution scores compatible with secure aggregation. Fair-Private satisfies standard fairness axioms, while Everybody-Else eliminates self-evaluation and provides resistance to manipulation, addressing a largely overlooked vulnerability. We provide theoretical guarantees for fairness, privacy, robustness, and computational efficiency, and evaluate our methods on multiple medical image datasets and CIFAR10 in cross-silo settings. Our scores consistently outperform existing baselines, better approximate Shapley-induced client rankings, and improve downstream model performance as well as misbehavior detection. These results demonstrate that fairness, privacy, robustness, and practical utility can be achieved jointly in federated contribution evaluation, offering a principled solution for real-world cross-silo deployments.

Human recursive numeral systems (i.e., counting systems such as English base-10 numerals), like many other grammatical systems, are highly regular. Following prior work that relates cross-linguistic tendencies to biases in learning, we ask whether regular systems are common because regularity facilitates learning. Adopting methods from the Reinforcement Learning literature, we confirm that highly regular human(-like) systems are easier to learn than unattested but possible irregular systems. This asymmetry emerges under the natural assumption that recursive numeral systems are designed for generalisation from limited data to represent all integers exactly. We also find that the influence of regularity on learnability is absent for unnatural, highly irregular systems, whose learnability is influenced instead by signal length, suggesting that different pressures may influence learnability differently in different parts of the space of possible numeral systems. Our results contribute to the body of work linking learnability to cross-linguistic prevalence.

Tabular data is the primary data format in industrial relational databases, underpinning modern data analytics and decision-making. However, the increasing scale of tabular data poses significant computational and storage challenges to learning-based analytical systems. This highlights the need for data-efficient learning, which enables effective model training and generalization using substantially fewer samples. Dataset condensation (DC) has emerged as a promising data-centric paradigm that synthesizes small yet informative datasets to preserve data utility while reducing storage and training costs. However, existing DC methods are computationally intensive due to reliance on complex gradient-based optimization. Moreover, they often overlook key characteristics of tabular data, such as heterogeneous features and class imbalance. To address these limitations, we introduce C$^{2}$TC (Class-Adaptive Clustering for Tabular Condensation), the first training-free tabular dataset condensation framework that jointly optimizes class allocation and feature representation, enabling efficient and scalable condensation. Specifically, we reformulate the dataset condensation objective into a novel class-adaptive cluster allocation problem (CCAP), which eliminates costly training and integrates adaptive label allocation to handle class imbalance. To solve the NP-hard CCAP, we develop HFILS, a heuristic local search that alternates between soft allocation and class-wise clustering to efficiently obtain high-quality solutions. Moreover, a hybrid categorical feature encoding (HCFE) is proposed for semantics-preserving clustering of heterogeneous discrete attributes. Extensive experiments on 10 real-world datasets demonstrate that C$^{2}$TC improves efficiency by at least 2 orders of magnitude over state-of-the-art baselines, while achieving superior downstream performance.

The growing integration of distributed photovoltaics (PVs) into active distribution networks (ADNs) has exacerbated operational challenges, making it imperative to coordinate diverse equipment to mitigate voltage violations and enhance power quality. Although existing data-driven approaches have demonstrated effectiveness in the voltage control problem, they often require extensive trial-and-error exploration and struggle to incorporate heterogeneous information, such as day-ahead forecasts and semantic-based grid codes. Considering the operational scenarios and requirements in real-world ADNs, in this paper, we propose a hybrid knowledge-data-driven approach that leverages dynamic collaboration between a large language model (LLM) agent and a reinforcement learning (RL) agent to achieve two-stage voltage control. In the day-ahead stage, the LLM agent receives coarse region-level forecasts and generates scheduling strategies for on-load tap changer (OLTC) and shunt capacitors (SCs) to regulate the overall voltage profile. Then in the intra-day stage, based on accurate node-level measurements, the RL agent refines terminal voltages by deriving reactive power generation strategies for PV inverters. On top of the LLM-RL collaboration framework, we further propose a self-evolution mechanism for the LLM agent and a pretrain-finetune pipeline for the RL agent, effectively enhancing and coordinating the policies for both agents. The proposed approach not only aligns more closely with practical operational characteristics but also effectively utilizes the inherent knowledge and reasoning capabilities of the LLM agent, significantly improving training efficiency and voltage control performance. Comprehensive comparisons and ablation studies demonstrate the effectiveness of the proposed method.

State-of-the-art learned reconstruction methods often rely on black-box modules that, despite their strong performance, raise questions about their interpretability and robustness. Here, we build on a recently proposed image reconstruction method, which is based on embedding data-driven information into a model-based convolutional dictionary regularization via neural network-inferred spatially adaptive sparsity level maps. By means of improved network design and dedicated training strategies, we extend the method to achieve filter-permutation invariance as well as the possibility to change the convolutional dictionary at inference time. We apply our method to low-field MRI and compare it to several other recent deep learning-based methods, also on in vivo data, in which the benefit for the use of a different dictionary is showcased. We further assess the method's robustness when tested on in- and out-of-distribution data. When tested on the latter, the proposed method suffers less from the data distribution shift compared to the other learned methods, which we attribute to its reduced reliance on training data due to its underlying model-based reconstruction component.

Minimally invasive surgery has dramatically improved patient operative outcomes, yet identifying safe operative zones remains challenging in critical phases, requiring surgeons to integrate visual cues, procedural phase, and anatomical context under high cognitive load. Existing AI systems offer binary safety verification or static detection, ignoring the phase-dependent nature of intraoperative reasoning. We introduce ResGo, a benchmark of laparoscopic frames annotated with Go Zone bounding boxes and clinician-authored rationales covering phase, exposure quality reasoning, next action and risk reminder. We introduce evaluation metrics that treat correct grounding under incorrect phase as failures, revealing that most vision-language models cannot handle such tasks and perform poorly. We then present SurGo-R1, a model optimized via RLHF with a multi-turn phase-then-go architecture where the model first identifies the surgical phase, then generates reasoning and Go Zone coordinates conditioned on that context. On unseen procedures, SurGo-R1 achieves 76.6% phase accuracy, 32.7 mIoU, and 54.8% hardcore accuracy, a 6.6$\times$ improvement over the mainstream generalist VLMs. Code, model and benchmark will be available at https://github.com/jinlab-imvr/SurGo-R1

Large Vision-Language Models (LVLMs) exhibit outstanding performance on vision-language tasks but struggle with hallucination problems. Through in-depth analysis of LVLM activation patterns, we reveal two key findings: 1) truthfulness and visual perception capabilities predominantly engage different subsets of attention heads within the model architecture; and 2) truthfulness steering vectors vary significantly across different semantic contexts. Based on these observations, we propose Dynamic Multimodal Activation Steering, a training-free approach for hallucination mitigation. Our method constructs a semantic-based truthfulness steering vector database and computes visual perception steering vectors, enabling context-aware interventions during inference by dynamically selecting the most relevant steering vectors based on input semantic similarity and applying them to the most influential attention heads. We conduct comprehensive experiments across multiple models and datasets, demonstrating that our approach significantly enhances model performance, outperforming existing state-of-the-art methods.

A U-Net based deep learning architecture is designed to segment brain tumors as they appear on various MRI modalities. Special emphasis is lent to the non-enhancing tumor compartment. The latter has not been considered anymore in recent brain tumor segmentation challenges like the MICCAI challenges. However, it is considered to be indicative of the survival time of the patient as well as of areas of further tumor growth. Hence it deems essential to have means to automatically delineate its extension within the tumor.

A central challenge in scientific machine learning (ML) is the correct representation of physical systems governed by multi-regime behaviours. In these scenarios, standard data analysis techniques often fail to capture the nature of the data, as the system's response varies significantly across the state space due to its stochasticity and the different physical regimes. Uncertainty quantification (UQ) should thus not be viewed merely as a safety assessment, but as a support to the learning task itself, guiding the model to internalise the behaviour of the data. We address this by focusing on the Critical Heat Flux (CHF) benchmark and dataset presented by the OECD/NEA Expert Group on Reactor Systems Multi-Physics. This case study represents a test for scientific ML due to the non-linear dependence of CHF on the inputs and the existence of distinct microscopic physical regimes. These regimes exhibit diverse statistical profiles, a complexity that requires UQ techniques to internalise the data behaviour and ensure reliable predictions. In this work, we conduct a comparative analysis of UQ methodologies to determine their impact on physical representation. We contrast post-hoc methods, specifically conformal prediction, against end-to-end coverage-oriented pipelines, including (Bayesian) heteroscedastic regression and quality-driven losses. These approaches treat uncertainty not as a final metric, but as an active component of the optimisation process, modelling the prediction and its behaviour simultaneously. We show that while post-hoc methods ensure statistical calibration, coverage-oriented learning effectively reshapes the model's representation to match the complex physical regimes. The result is a model that delivers not only high predictive accuracy but also a physically consistent uncertainty estimation that adapts dynamically to the intrinsic variability of the CHF.

Large language models (LLMs) for code editing have achieved remarkable progress, yet recent empirical studies reveal a fundamental disconnect between technical accuracy and developer productivity. Despite their strong benchmark performance, developers complete tasks 19% slower when using AI assistance, with over 68.81% of recommendations disrupting their mental flow. This misalignment stems from the use of static commit snapshots that lack temporal information, causing models to optimize for end results rather than the incremental, context-sensitive steps that align with developers' natural reasoning process. To bridge this gap, we present EditFlow, which benchmarks and optimizes subsequent code edit recommendation systems through the reconstruction of developer editing flows. EditFlow addresses three key challenges. First, collecting edit-order data that reflects developers' flow is inherently difficult: manual annotation introduces prohibitive overhead, while development logs capture only single trajectories instead of all plausible editing flows. Second, benchmarking recommendation performance against developers' ongoing editing flow requires a digital-twin-like simulation that can faithfully simulate the editing process. Third, existing heterogeneous systems vary drastically in scale and architecture, posing challenges for developing a unified optimization strategy that endows all models with mental-flow awareness regardless of design or capability. ......

Multimodal time series forecasting has garnered significant attention for its potential to provide more accurate predictions than traditional single-modality models by leveraging rich information inherent in other modalities. However, due to fundamental challenges in modality alignment, existing methods often struggle to effectively incorporate multimodal data into predictions, particularly textual information that has a causal influence on time series fluctuations, such as emergency reports and policy announcements. In this paper, we reflect on the role of textual information in numerical forecasting and propose Time series transformers with Multimodal Mixture-of-Experts, TiMi, to unleash the causal reasoning capabilities of LLMs. Concretely, TiMi utilizes LLMs to generate inferences on future developments, which serve as guidance for time series forecasting. To seamlessly integrate both exogenous factors and time series into predictions, we introduce a Multimodal Mixture-of-Experts (MMoE) module as a lightweight plug-in to empower Transformer-based time series models for multimodal forecasting, eliminating the need for explicit representation-level alignment. Experimentally, our proposed TiMi demonstrates consistent state-of-the-art performance on sixteen real-world multimodal forecasting benchmarks, outperforming advanced baselines while offering both strong adaptability and interpretability.

Multi-modal distribution in robotic manipulation action sequences poses critical challenges for imitation learning. To this end, existing approaches often model the action space as either a discrete set of tokens or a continuous, latent-variable distribution. However, both approaches present trade-offs: some methods discretize actions into tokens and therefore lose fine-grained action variations, while others generate continuous actions in a single stage tend to produce unstable mode transitions. To address these limitations, we propose Primary-Fine Decoupling for Action Generation (PF-DAG), a two-stage framework that decouples coarse action consistency from fine-grained variations. First, we compress action chunks into a small set of discrete modes, enabling a lightweight policy to select consistent coarse modes and avoid mode bouncing. Second, a mode conditioned MeanFlow policy is learned to generate high-fidelity continuous actions. Theoretically, we prove PF-DAG's two-stage design achieves a strictly lower MSE bound than single-stage generative policies. Empirically, PF-DAG outperforms state-of-the-art baselines across 56 tasks from Adroit, DexArt, and MetaWorld benchmarks. It further generalizes to real-world tactile dexterous manipulation tasks. Our work demonstrates that explicit mode-level decoupling enables both robust multi-modal modeling and reactive closed-loop control for robotic manipulation.

Eliminating undefined behaviors (UBs) in Rust programs requires a deep semantic understanding to enable accurate and reliable repair. While existing studies have demonstrated the potential of LLMs to support Rust code analysis and repair, most frameworks remain constrained by inflexible templates or lack grounding in executable semantics, resulting in limited contextual awareness and semantic incorrectness. Here, we present AkiraRust, an LLM-driven repair and verification framework that incorporates a finite-state machine to dynamically adapt its detection and repair flow to runtime semantic conditions. AkiraRust introduces a dual-mode reasoning strategy that coordinates fast and slow thinking across multiple agents. Each agent is mapped to an FSM state, and a waveform-driven transition controller manages state switching, rollback decisions, and semantic check pointing, enabling context-aware and runtime-adaptive repair. Experimental results show that AkiraRust achieves about 92% semantic correctness and delivers a 2.2x average speedup compared to SOTA.

Cooperative Multi-Agent Reinforcement Learning (MARL) solves complex tasks that require coordination from multiple agents, but is often limited to either local (independent learning) or global (centralized learning) perspectives. In this paper, we introduce a novel sequential training scheme and MARL architecture, which learns from multiple perspectives on different hierarchy levels. We propose the Hierarchical Lead Critic (HLC) - inspired by natural emerging distributions in team structures, where following high-level objectives combines with low-level execution. HLC demonstrates that introducing multiple hierarchies, leveraging local and global perspectives, can lead to improved performance with high sample efficiency and robust policies. Experimental results conducted on cooperative, non-communicative, and partially observable MARL benchmarks demonstrate that HLC outperforms single hierarchy baselines and scales robustly with increasing amounts of agents and difficulty.

Generative recommendation (GR) aligns with advances in generative AI by casting next-item prediction as token-level generation rather than score-based ranking. Most GR methods adopt a two-stage pipeline: (i) \textit{item tokenization}, which maps each item to a sequence of discrete, hierarchically organized tokens; and (ii) \textit{autoregressive generation}, which predicts the next item's tokens conditioned on the tokens of user's interaction history. Although hierarchical tokenization induces a prefix tree (trie) over items, standard autoregressive modeling with conventional Transformers often flattens item tokens into a linear stream and overlooks the underlying topology. To address this, we propose TrieRec, a trie-aware generative recommendation method that augments Transformers with structural inductive biases via two positional encodings. First, a \textit{trie-aware absolute positional encoding} aggregates a token's (node's) local structural context (\eg depth, ancestors, and descendants) into the token representation. Second, a \textit{topology-aware relative positional encoding} injects pairwise structural relations into self-attention to capture topology-induced semantic relatedness. TrieRec is also model-agnostic, efficient, and hyperparameter-free. In our experiments, we implement TrieRec within three representative GR backbones, achieving notably improvements of 8.83\% on average across four real-world datasets.

Active automata learning (AAL) algorithms can learn a behavioral model of a system from interacting with it. The primary challenge remains scaling to larger models, in particular in the presence of many possible inputs to the system. Modern AAL algorithms fail to scale even if, in every state, most inputs lead to errors. In various challenging problems from the literature, these errors are observable, i.e., they emit a known error output. Motivated by these problems, we study learning these systems more efficiently. Further, we consider various degrees of knowledge about which inputs are non-error producing at which state. For each level of knowledge, we provide a matching adaptation of the state-of-the-art AAL algorithm L# to make the most of this domain knowledge. Our empirical evaluation demonstrates that the methods accelerate learning by orders of magnitude with strong but realistic domain knowledge to a single order of magnitude with limited domain knowledge.

Multi-robot task planning requires decomposing natural-language instructions into executable actions for heterogeneous robot teams. Conventional Planning Domain Definition Language (PDDL) planners provide rigorous guarantees but struggle to handle ambiguous or long-horizon missions, while large language models (LLMs) can interpret instructions and propose plans but may hallucinate or produce infeasible actions. We present a hierarchical multi-agent LLM-based planner with prompt optimization: an upper layer decomposes tasks and assigns them to lower-layer agents, which generate PDDL problems solved by a classical planner. When plans fail, the system applies TextGrad-inspired textual-gradient updates to optimize each agent's prompt and thereby improve planning accuracy. In addition, meta-prompts are learned and shared across agents within the same layer, enabling efficient prompt optimization in multi-agent settings. On the MAT-THOR benchmark, our planner achieves success rates of 0.95 on compound tasks, 0.84 on complex tasks, and 0.60 on vague tasks, improving over the previous state-of-the-art LaMMA-P by 2, 7, and 15 percentage points respectively. An ablation study shows that the hierarchical structure, prompt optimization, and meta-prompt sharing contribute roughly +59, +37, and +4 percentage points to the overall success rate.

Knowledge Distillation (KD) has emerged as a crucial technique for compressing Large Language Models (LLMs). Although existing cross-tokenizer KD methods have made notable progress, their effectiveness remains constrained by suboptimal alignment across sequence and vocabulary levels. To address these limitations, we introduce Dual-Space Weighting and Time-Warped Alignment (DWA-KD), a novel cross-tokenizer distillation framework that enhances token-wise distillation through dual-space entropy-based weighting and achieves precise sequence-level alignment by leveraging both lexical and semantic information. At the token level, DWA-KD maps teacher representations into the student space and vice versa, performing dual-space KD via Kullback-Leibler divergence (KL). The process is modulated by dual-space weights that up-weight tokens where the student is uncertain and the teacher is confident, thereby focusing learning on informative tokens rather than treating all positions equally. At the sequence level, DWA-KD applies Soft Dynamic Time Warping (Soft-DTW) to both the embedding and final hidden-state layers, enabling robust alignment of lexical and contextual semantics between teacher and student sequences. Extensive experiments across diverse NLP benchmarks demonstrate that DWA-KD outperforms state-of-the-art KD baselines, while ablation studies confirm the complementary contributions of entropy-based token weighting and embedding and final hidden state layer Soft-DTW alignment.

Computer-aided diagnosis (CAD) has significantly advanced automated chest X-ray diagnosis but remains isolated from clinical workflows and lacks reliable decision support and interpretability. Human-AI collaboration seeks to enhance the reliability of diagnostic models by integrating the behaviors of controllable radiologists. However, the absence of interactive tools seamlessly embedded within diagnostic routines impedes collaboration, while the semantic gap between radiologists' decision-making patterns and model representations further limits clinical adoption. To overcome these limitations, we propose a visual cognition-guided collaborative network (VCC-Net) to achieve the cooperative diagnostic paradigm. VCC-Net centers on visual cognition (VC) and employs clinically compatible interfaces, such as eye-tracking or the mouse, to capture radiologists' visual search traces and attention patterns during diagnosis. VCC-Net employs VC as a spatial cognition guide, learning hierarchical visual search strategies to localize diagnostically key regions. A cognition-graph co-editing module subsequently integrates radiologist VC with model inference to construct a disease-aware graph. The module captures dependencies among anatomical regions and aligns model representations with VC-driven features, mitigating radiologist bias and facilitating complementary, transparent decision-making. Experiments on the public datasets SIIM-ACR, EGD-CXR, and self-constructed TB-Mouse dataset achieved classification accuracies of 88.40%, 85.05%, and 92.41%, respectively. The attention maps produced by VCC-Net exhibit strong concordance with radiologists' gaze distributions, demonstrating a mutual reinforcement of radiologist and model inference. The code is available at https://github.com/IPMI-NWU/VCC-Net.

Image captioning remains a fundamental task for vision language understanding, yet ground-truth supervision still relies predominantly on human-annotated references. Because human annotations reflect subjective preferences and expertise, ground-truth captions are often incomplete or even incorrect, which in turn limits caption models. We argue that caption quality should be assessed by two objective aspects: completeness (does the caption cover all salient visual facts?) and correctness (are the descriptions true with respect to the image?). To this end, we introduce CCCaption: a dual-reward reinforcement learning framework with a dedicated fine-tuning corpus that explicitly optimizes these properties to generate \textbf{C}omplete and \textbf{C}orrect \textbf{Captions}. For completeness, we use diverse LVLMs to disentangle the image into a set of visual queries, and reward captions that answer more of these queries, with a dynamic query sampling strategy to improve training efficiency. For correctness, we penalize captions that contain hallucinations by validating the authenticity of sub-caption queries, which are derived from the caption decomposition. Our symmetric dual-reward optimization jointly maximizes completeness and correctness, guiding models toward captions that better satisfy these objective criteria. Extensive experiments across standard captioning benchmarks show consistent improvements, offering a principled path to training caption models beyond human-annotation imitation.

Large language models have demonstrated capabilities in text generation, while their increasing parameter scales present challenges in computational and memory efficiency. Post-training sparsity (PTS), which reduces model cost by removing weights from dense networks, is an effective approach. However, native dense matrices lack high sparsity, making existing approaches that directly remove weights disrupt model states, resulting in unsatisfactory performance recovery even with post-tuning. We propose Sparsity Induction, which promotes models toward higher sparsity at both distribution and feature levels before pruning, to push the limits of PTS. At the distribution level, we enhance distributional sparsity through mathematically equivalent scaling transformations, which are fully absorbable and incur no extra parameters or inference-time overhead. At the feature level, we introduce Spectral Norm Loss to promote feature sparsity from a low-rank perspective. Experiments across diverse model architectures and tasks demonstrate that our method further enhances sparsity-friendliness, achieving superior pruning performance over existing approaches.

Public policy decisions are typically justified using a narrow set of headline indicators, leaving many downstream social impacts unstructured and difficult to compare across policies. We propose PPCR-IM, a system for multi-layer DAG-based consequence reasoning and social indicator mapping that addresses this gap. Given a policy description and its context, PPCR-IM uses an LLM-driven, layer-wise generator to construct a directed acyclic graph of intermediate consequences, allowing child nodes to have multiple parents to capture joint influences. A mapping module then aligns these nodes to a fixed indicator set and assigns one of three qualitative impact directions: increase, decrease, or ambiguous change. For each policy episode, the system outputs a structured record containing the DAG, indicator mappings, and three evaluation measures: an expected-indicator coverage score, a discovery rate for overlooked but relevant indicators, and a relative focus ratio comparing the systems coverage to that of the government. PPCR-IM is available both as an online demo and as a configurable XLSX-to-JSON batch pipeline.

Clinical risk prediction models often underperform in real-world settings due to poor calibration, limited transportability, and subgroup disparities. These challenges are amplified in high-dimensional multimodal cancer datasets characterized by complex feature interactions and a p >> n structure. We present a fully reproducible multimodal machine learning framework for 5-year overall survival prediction in breast cancer, integrating clinical variables with high-dimensional transcriptomic and copy-number alteration (CNA) features from the METABRIC cohort. After variance- and sparsity-based filtering and dimensionality reduction, models were trained using stratified train/validation/test splits with validation-based hyperparameter tuning. Two survival approaches were compared: an elastic-net regularized Cox model (CoxNet) and a gradient-boosted survival tree model implemented using XGBoost. CoxNet provides embedded feature selection and stable estimation, whereas XGBoost captures nonlinear effects and higher-order interactions. Performance was assessed using time-dependent area under the ROC curve (AUC), average precision (AP), calibration curves, Brier score, and bootstrapped 95 percent confidence intervals. CoxNet achieved validation and test AUCs of 98.3 and 96.6, with AP values of 90.1 and 80.4. XGBoost achieved validation and test AUCs of 98.6 and 92.5, with AP values of 92.5 and 79.9. Fairness diagnostics showed stable discrimination across age groups, estrogen receptor status, molecular subtypes, and menopausal state. This work introduces a governance-oriented multimodal survival framework emphasizing calibration, fairness auditing, robustness, and reproducibility for high-dimensional clinical machine learning.

This paper presents and evaluates an optimized cascaded Nepali speech-to-English text translation (S2TT) system, focusing on mitigating structural noise introduced by Automatic Speech Recognition (ASR). We first establish highly proficient ASR and NMT components: a Wav2Vec2-XLS-R-300m model achieved a state-of-the-art 2.72% CER on OpenSLR-54, and a multi-stage fine-tuned MarianMT model reached a 28.32 BLEU score on the FLORES-200 benchmark. We empirically investigate the influence of punctuation loss, demonstrating that unpunctuated ASR output significantly degrades translation quality, causing a massive 20.7% relative BLEU drop on the FLORES benchmark. To overcome this, we propose and evaluate an intermediate Punctuation Restoration Module (PRM). The final S2TT pipeline was tested across three configurations on a custom dataset. The optimal configuration, which applied the PRM directly to ASR output, achieved a 4.90 BLEU point gain over the direct ASR-to-NMT baseline (BLEU 36.38 vs. 31.48). This improvement was validated by human assessment, which confirmed the optimized pipeline's superior Adequacy (3.673) and Fluency (3.804). This work validates that targeted punctuation restoration is the most effective intervention for mitigating structural noise in the Nepali S2TT pipeline. It establishes an optimized baseline and demonstrates a critical architectural insight for developing cascaded speech translation systems for similar low-resource languages.

Multimodal Large Language Models (MLLMs) have achieved notable success in enhancing translation performance by integrating multimodal information. However, existing research primarily focuses on image-guided methods, whose applicability is constrained by the scarcity of multilingual image-text pairs. The speech modality overcomes this limitation due to its natural alignment with text and the abundance of existing speech datasets, which enable scalable language coverage. In this paper, we propose a Speech-guided Machine Translation (SMT) framework that integrates speech and text as fused inputs into an MLLM to improve translation quality. To mitigate reliance on low-resource data, we introduce a Self-Evolution Mechanism. The core components of this framework include a text-to-speech model, responsible for generating synthetic speech, and an MLLM capable of classifying synthetic speech samples and iteratively optimizing itself using positive samples. Experimental results demonstrate that our framework surpasses all existing methods on the Multi30K multimodal machine translation benchmark, achieving new state-of-the-art results. Furthermore, on general machine translation datasets, particularly the FLORES-200, it achieves average state-of-the-art performance in 108 translation directions. Ablation studies on CoVoST-2 confirms that differences between synthetic and authentic speech have negligible impact on translation quality. The code and models are released at https://github.com/yxduir/LLM-SRT.

Uncertainty is inherent in modern engineered systems, including cyber-physical systems, autonomous systems, and large-scale software-intensive infrastructures (such as microservice-based systems) operating in dynamic and partially observable environments. The recent publication of Precise Semantics for Uncertainty Modeling (PSUM) by the Object Management Group represents the first standardized specification for uncertainty modeling within the Model-Based Systems Engineering (MBSE) community, providing formally defined semantics for representing and reasoning about uncertainty in models. In parallel, the second version of Systems Modeling Language (SysML v2) was released as the next-generation systems modeling language, offering improved semantic rigor and reusability, yet lacking native constructs aligned with PSUM for first-class uncertainty representation. This paper proposes a systematic extension of SysML v2 that incorporates the PSUM metamodel into its modeling framework. The extension enables explicit specification of indeterminacy sources, structured characterization of uncertainties, and consistent propagation of uncertainty within system models, while preserving conformance with SysML v2 syntax and semantics. We validate the approach through seven case studies. Results demonstrate that the proposed extension (PSUM-SysMLv2) is expressive and applicable for uncertainty-aware MBSE, and potentially enables uncertainty and uncertainty propagation analyses.

Effectively addressing client resistance is a sophisticated clinical skill in psychological counseling, yet practitioners often lack timely and scalable supervisory feedback to refine their approaches. Although current NLP research has examined overall counseling quality and general therapeutic skills, it fails to provide granular evaluations of high-stakes moments where clients exhibit resistance. In this work, we present a comprehensive pipeline for the multi-dimensional evaluation of human counselors' interventions specifically targeting client resistance in text-based therapy. We introduce a theory-driven framework that decomposes counselor responses into four distinct communication mechanisms. Leveraging this framework, we curate and share an expert-annotated dataset of real-world counseling excerpts, pairing counselor-client interactions with professional ratings and explanatory rationales. Using this data, we perform full-parameter instruction tuning on a Llama-3.1-8B-Instruct backbone to model fine-grained evaluative judgments of response quality and generate explanations underlying. Experimental results show that our approach can effectively distinguish the quality of different communication mechanisms (77-81% F1), substantially outperforming GPT-4o and Claude-3.5-Sonnet (45-59% F1). Moreover, the model produces high-quality explanations that closely align with expert references and receive near-ceiling ratings from human experts (2.8-2.9/3.0). A controlled experiment with 43 counselors further confirms that receiving these AI-generated feedback significantly improves counselors' ability to respond effectively to client resistance.

Lifetime Value (LTV) prediction is critical in advertising, recommender systems, and e-commerce. In practice, LTV data patterns vary across decision scenarios. As a result, practitioners often build complex, scenario-specific pipelines and iterate over feature processing, objective design, and tuning. This process is expensive and hard to transfer. We propose AgentLTV, an agent-based unified search-and-evolution framework for automated LTV modeling. AgentLTV treats each candidate solution as an {executable pipeline program}. LLM-driven agents generate code, run and repair pipelines, and analyze execution feedback. Two decision agents coordinate a two-stage search. The Monte Carlo Tree Search (MCTS) stage explores a broad space of modeling choices under a fixed budget, guided by the Polynomial Upper Confidence bounds for Trees criterion and a Pareto-aware multi-metric value function. The Evolutionary Algorithm (EA) stage refines the best MCTS program via island-based evolution with crossover, mutation, and migration. Experiments on a large-scale proprietary dataset and a public benchmark show that AgentLTV consistently discovers strong models across ranking and error metrics. Online bucket-level analysis further indicates improved ranking consistency and value calibration, especially for high-value and negative-LTV segments. We summarize practitioner-oriented takeaways: use MCTS for rapid adaptation to new data patterns, use EA for stable refinement, and validate deployment readiness with bucket-level ranking and calibration diagnostics. The proposed AgentLTV has been successfully deployed online.

Standard vision-language-action (VLA) models rely on fitting statistical data priors, limiting their robust understanding of underlying physical dynamics. Reinforcement learning enhances physical grounding through exploration yet typically relies on external reward signals that remain isolated from the agent's internal states. World action models have emerged as a promising paradigm that integrates imagination and control to enable predictive planning. However, they rely on implicit context modeling, lacking explicit mechanisms for self-improvement. To solve these problems, we propose Self-Correcting VLA (SC-VLA), which achieve self-improvement by intrinsically guiding action refinement through sparse imagination. We first design sparse world imagination by integrating auxiliary predictive heads to forecast current task progress and future trajectory trends, thereby constraining the policy to encode short-term physical evolution. Then we introduce the online action refinement module to reshape progress-dependent dense rewards, adjusting trajectory orientation based on the predicted sparse future states. Evaluations on challenging robot manipulation tasks from simulation benchmarks and real-world settings demonstrate that SC-VLA achieve state-of-the-art performance, yielding the highest task throughput with 16% fewer steps and a 9% higher success rate than the best-performing baselines, alongside a 14% gain in real-world experiments. Code is available at https://github.com/Kisaragi0/SC-VLA.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a prevailing paradigm for enhancing reasoning in Multimodal Large Language Models (MLLMs). However, relying solely on outcome supervision risks reward hacking, where models learn spurious reasoning patterns to satisfy final answer checks. While recent rubric-based approaches offer fine-grained supervision signals, they suffer from high computational costs of instance-level generation and inefficient training dynamics caused by treating all rubrics as equally learnable. In this paper, we propose Stratified Rubric-based Curriculum Learning (RuCL), a novel framework that reformulates curriculum learning by shifting the focus from data selection to reward design. RuCL generates generalized rubrics for broad applicability and stratifies them based on the model's competence. By dynamically adjusting rubric weights during training, RuCL guides the model from mastering foundational perception to tackling advanced logical reasoning. Extensive experiments on various visual reasoning benchmarks show that RuCL yields a remarkable +7.83% average improvement over the Qwen2.5-VL-7B model, achieving a state-of-the-art accuracy of 60.06%.

We study the discrete Bertrand pricing game with a non-increasing demand function. The game has $n \ge 2$ players who simultaneously choose prices from the set $\{1/k, 2/k, \ldots, 1\}$, where $k\in\mathbb{N}$. The player who sets the lowest price captures the entire demand; if multiple players tie for the lowest price, they split the demand equally. We study the Bertrand paradox, where classical theory predicts low prices, yet real markets often sustain high prices. To understand this gap, we analyze a repeated-game model in which firms set prices using no-regret learners. Our goal is to characterize the equilibrium outcomes that can arise under different no-regret learning guarantees. We are particularly interested in questions such as whether no-external-regret learners can converge to undesirable high-price outcomes, and how stronger guarantees such as no-swap regret shape the emergence of competitive low-price behavior. We address these and related questions through a theoretical analysis, complemented by experiments that support the theory and reveal surprising phenomena for no-swap regret learners.

Visual spatial reasoning (VSR) remains challenging for modern vision-language models (VLMs), despite advances in multimodal architectures. A common strategy is to inject additional information at inference time, such as explicit spatial cues, external commonsense knowledge, or chain-of-thought (CoT) reasoning instructions. However, it remains unclear when such information genuinely improves reasoning and when it introduces noise. In this paper, we conduct a hypothesis-driven analysis of information injection for VSR across three representative VLMs and two public benchmarks. We examine (i) the type and number of spatial contexts, (ii) the amount and relevance of injected commonsense knowledge, and (iii) the interaction between spatial grounding and CoT prompting. Our results reveal a consistent pattern: more information does not necessarily yield better reasoning. Targeted single spatial cues outperform multi-context aggregation, excessive or weakly relevant commonsense knowledge degrades performance, and CoT prompting improves accuracy only when spatial grounding is sufficiently precise. These findings highlight the importance of selective, task-aligned information injection and provide practical guidance for designing reliable multimodal reasoning pipelines.

Deep intracranial tumors situated in eloquent brain regions controlling vital functions present critical diagnostic challenges. Clinical practice has shifted toward stereotactic biopsy for pathological confirmation before treatment. Yet biopsy carries inherent risks of hemorrhage and neurological deficits and struggles with sampling bias due to tumor spatial heterogeneity, because pathological changes are typically region-selective rather than tumor-wide. Therefore, advancing non-invasive MRI-based pathology prediction is essential for holistic tumor assessment and modern clinical decision-making. The primary challenge lies in data scarcity: low tumor incidence requires long collection cycles, and annotation demands biopsy-verified pathology from neurosurgical experts. Additionally, tiny lesion volumes lacking segmentation masks cause critical features to be overwhelmed by background noise. To address these challenges, we construct the ICT-MRI dataset - the first public biopsy-verified benchmark with 249 cases across four categories. We propose a Virtual Biopsy framework comprising: MRI-Processor for standardization; Tumor-Localizer employing vision-language models for coarse-to-fine localization via weak supervision; and Adaptive-Diagnoser with a Masked Channel Attention mechanism fusing local discriminative features with global contexts. Experiments demonstrate over 90% accuracy, outperforming baselines by more than 20%.

Large Language Models (LLMs) have demonstrated significant potential as autonomous software engineering (SWE) agents. Recent work has further explored augmenting these agents with memory mechanisms to support long-horizon reasoning. However, these approaches typically operate at a coarse instance granularity, treating the entire problem-solving episode as the atomic unit of storage and retrieval. We empirically demonstrate that instance-level memory suffers from a fundamental granularity mismatch, resulting in misguided retrieval when tasks with similar surface descriptions require distinct reasoning logic at specific stages. To address this, we propose Structurally Aligned Subtask-Level Memory, a method that aligns memory storage, retrieval, and updating with the agent's functional decomposition. Extensive experiments on SWE-bench Verified demonstrate that our method consistently outperforms both vanilla agents and strong instance-level memory baselines across diverse backbones, improving mean Pass@1 over the vanilla agent by +4.7 pp on average (e.g., +6.8 pp on Gemini 2.5 Pro). Performance gains grow with more interaction steps, showing that leveraging past experience benefits long-horizon reasoning in complex software engineering tasks.

Bangla-English code-mixing is widespread across South Asian social media, yet resources for implicit meaning identification in this setting remain scarce. Existing sentiment and sarcasm models largely focus on monolingual English or high-resource languages and struggle with transliteration variation, cultural references, and intra-sentential language switching. To address this gap, we introduce MixSarc, the first publicly available Bangla-English code-mixed corpus for implicit meaning identification. The dataset contains 9,087 manually annotated sentences labeled for humor, sarcasm, offensiveness, and vulgarity. We construct the corpus through targeted social media collection, systematic filtering, and multi-annotator validation. We benchmark transformer-based models and evaluate zero-shot large language models under structured prompting. Results show strong performance on humor detection but substantial degradation on sarcasm, offense, and vulgarity due to class imbalance and pragmatic complexity. Zero-shot models achieve competitive micro-F1 scores but low exact match accuracy. Further analysis reveals that over 42\% of negative sentiment instances in an external dataset exhibit sarcastic characteristics. MixSarc provides a foundational resource for culturally aware NLP and supports more reliable multi-label modeling in code-mixed environments.

High stress occurs when 3D heterogeneous IC packages are subjected to thermal cycling at extreme temperatures. Stress mainly occurs at the interface between different materials. We investigate stress image using latent space representation which is based on using deep generative model (DGM). However, most DGM approaches are unsupervised, meaning they resort to image pairing (input and output) to train DGM. Instead, we rely on a recent boundary-decoder (BD) net, which uses boundary condition and image pairing for stress modeling. The boundary net maps material parameters to the latent space co-shared by its image counterpart. Because such a setup is dimensionally wise ill-posed, we further couple BD net with deep clustering. To access the performance of our proposed method, we simulate an IC chip dataset comprising of 1825 stress images. We compare our new approach using variants of BD net as well as a baseline approach. We show that our approach is able to outperform all the comparison in terms of train and test error reduction.

Public service information systems are often fragmented, inconsistently formatted, and outdated. These characteristics create low-resource retrieval environments that hinder timely access to critical services. We investigate retrieval challenges in such settings through the domain of food pantry access, a socially urgent problem given persistent food insecurity. We develop an AI-powered conversational retrieval system that scrapes and indexes publicly available pantry data and employs a Retrieval-Augmented Generation (RAG) pipeline to support natural language queries via a web interface. We conduct a pilot evaluation study using community-sourced queries to examine system behavior in realistic scenarios. Our analysis reveals key limitations in retrieval robustness, handling underspecified queries, and grounding over inconsistent knowledge bases. This ongoing work exposes fundamental IR challenges in low-resource environments and motivates future research on robust conversational retrieval to improve access to critical public resources.

Neural Graph Databases (NGDBs) facilitate complex logical reasoning over incomplete knowledge structures, yet their training efficiency and expressivity are constrained by rigid query-level batching and structure-exclusive embeddings. We present NGDB-Zoo, a unified framework that resolves these bottlenecks by synergizing operator-level training with semantic augmentation. By decoupling logical operators from query topologies, NGDB-Zoo transforms the training loop into a dynamically scheduled data-flow execution, enabling multi-stream parallelism and achieving a $1.8\times$ - $6.8\times$ throughput compared to baselines. Furthermore, we formalize a decoupled architecture to integrate high-dimensional semantic priors from Pre-trained Text Encoders (PTEs) without triggering I/O stalls or memory overflows. Extensive evaluations on six benchmarks, including massive graphs like ogbl-wikikg2 and ATLAS-Wiki, demonstrate that NGDB-Zoo maintains high GPU utilization across diverse logical patterns and significantly mitigates representation friction in hybrid neuro-symbolic reasoning.

Generative images have proliferated on Web platforms in social media and online copyright distribution scenarios, and semantic watermarking has increasingly been integrated into diffusion models to support reliable provenance tracking and forgery prevention for web content. Traditional noise-layer-based watermarking, however, remains vulnerable to inversion attacks that can recover embedded signals. To mitigate this, recent content-aware semantic watermarking schemes bind watermark signals to high-level image semantics, constraining local edits that would otherwise disrupt global coherence. Yet, large language models (LLMs) possess structured reasoning capabilities that enable targeted exploration of semantic spaces, allowing locally fine-grained but globally coherent semantic alterations that invalidate such bindings. To expose this overlooked vulnerability, we introduce a Coherence-Preserving Semantic Injection (CSI) attack that leverages LLM-guided semantic manipulation under embedding-space similarity constraints. This alignment enforces visual-semantic consistency while selectively perturbing watermark-relevant semantics, ultimately inducing detector misclassification. Extensive empirical results show that CSI consistently outperforms prevailing attack baselines against content-aware semantic watermarking, revealing a fundamental security weakness of current semantic watermark designs when confronted with LLM-driven semantic perturbations.

Agent-based epidemic models (ABMs) encode behavioral and policy heterogeneity but are too slow for nightly hospital planning. We develop county-ready surrogates that learn directly from exascale ABM trajectories using Universal Differential Equations (UDEs): mechanistic SEIR-family ODEs with a neural-parameterized contact rate $κ_φ(u,t)$ (no additive residual). Our contributions are threefold: we adapt multiple shooting and an observer-based prediction-error method (PEM) to stabilize identification of neural-augmented epidemiological dynamics across intervention-driven regime shifts; we enforce positivity and mass conservation and show the learned contact-rate parameterization yields a well-posed vector field; and we quantify accuracy, calibration, and compute against ABM ensembles and UDE baselines. On a representative ExaEpi scenario, PEM-UDE reduces mean MSE by 77% relative to single-shooting UDE (3.00 vs. 13.14) and by 20% relative to MS-UDE (3.75). Reliability improves in parallel: empirical coverage of ABM $10$-$90$% and $25$-$75$% bands rises from 0.68/0.43 (UDE) and 0.79/0.55 (MS-UDE) to 0.86/0.61 with PEM-UDE and 0.94/0.69 with MS+PEM-UDE, indicating calibrated uncertainty rather than overconfident fits. Inference runs in seconds on commodity CPUs (20-35 s per $\sim$90-day forecast), enabling nightly ''what-if'' sweeps on a laptop. Relative to a $\sim$100 CPU-hour ABM reference run, this yields $\sim10^{4}\times$ lower wall-clock per scenario. This closes the realism-cadence gap, supports threshold-aware decision-making (e.g., maintaining ICU occupancy $<75$%), preserves mechanistic interpretability, and enables calibrated, risk-aware scenario planning on standard institutional hardware. Beyond epidemics, the ABM$\to$UDE recipe provides a portable path to distill agent-based simulators into fast, trustworthy surrogates for other scientific domains.

Many applications seek to optimize LLM outputs at test time by iteratively proposing, scoring, and refining candidates over a discrete output space. Existing methods use a calibrated scalar evaluator for the target objective to guide search, but for many tasks such scores are unavailable, too sparse, or unreliable. Pairwise comparisons, by contrast, are often easier to elicit, still provide useful signal on improvement directions, and can be obtained from the LLM itself without external supervision. Building on this observation, we introduce Duel-Evolve, an evolutionary optimization algorithm that replaces external scalar rewards with pairwise preferences elicited from the same LLM used to generate candidates. Duel-Evolve aggregates these noisy candidate comparisons via a Bayesian Bradley-Terry model, yielding uncertainty-aware estimates of candidate quality. These quality estimates guide allocation of the comparison budget toward plausible optima using Double Thompson Sampling, as well as selection of high-quality parents to generate improved candidates. We evaluate Duel-Evolve on MathBench, where it achieves 20 percentage points higher accuracy over existing methods and baselines, and on LiveCodeBench, where it improves over comparable iterative methods by over 12 percentage points. Notably, the method requires no reward model, no ground-truth labels during search, and no hand-crafted scoring function. Results show that pairwise self-preferences provide strong optimization signal for test-time improvement over large, discrete output spaces.

In the context of the evolution of artificial intelligence (AI), the interaction between humans and AI entities has become increasingly salient, challenging the conventional human-centric paradigms of human-machine interaction. To address this challenge, it is imperative to reassess the relationship between AI entities and humans. Through considering both the virtual and physical worlds, we can construct a novel descriptive framework for a world where humans and machines coexist symbiotically. This paper will introduce a fresh research direction engendered for studying harmonious human-machine coexistence across physical and virtual worlds, which has been termed "symmetrical reality". We will elucidate its key characteristics, offering innovative research insight for renovating human-machine interaction paradigms.

Ordinal categorical data are widely collected in psychology, education, and other social sciences, appearing commonly in questionnaires, assessments, and surveys. Latent class models provide a flexible framework for uncovering unobserved heterogeneity by grouping individuals into homogeneous classes based on their response patterns. A fundamental challenge in applying these models is determining the number of latent classes, which is unknown and must be inferred from data. In this paper, we propose one test statistic for this problem. The test statistic centers the largest singular value of a normalized residual matrix by a simple sample-size adjustment. Under the null hypothesis that the candidate number of latent classes is correct, its upper bound converges to zero in probability. Under an under-fitted alternative, the statistic itself exceeds a fixed positive constant with probability approaching one. This sharp dichotomous behavior of the test statistic yields two sequential testing algorithms that consistently estimate the true number of latent classes. Extensive experimental studies confirm the theoretical findings and demonstrate their accuracy and reliability in determining the number of latent classes.

Estimating the asymmetric numbers of communities in multi-layer directed networks is a challenging problem due to the multi-layer structures and inherent directional asymmetry, leading to possibly different numbers of sender and receiver communities. This work addresses this issue under the multi-layer stochastic co-block model, a model for multi-layer directed networks with distinct community structures in sending and receiving sides, by proposing a novel goodness-of-fit test. The test statistic relies on the deviation of the largest singular value of an aggregated normalized residual matrix from the constant 2. The test statistic exhibits a sharp dichotomy: Under the null hypothesis of correct model specification, its upper bound converges to zero with high probability; under underfitting, the test statistic itself diverges to infinity. With this property, we develop a sequential testing procedure that searches through candidate pairs of sender and receiver community numbers in a lexicographic order. The process stops at the smallest such pair where the test statistic drops below a decaying threshold. For robustness, we also propose a ratio-based variant algorithm, which detects sharp changes in the sequence of test statistics by comparing consecutive candidates. Both methods are proven to consistently determine the true numbers of sender and receiver communities under the multi-layer stochastic co-block model.

Developer Productivity Dashboards are essential for visualizing DevOps performance metrics such as Deployment Frequency and Change Failure Rate (DORA). However, the utility of these dashboards is frequently undermined by data reliability issues. In early iterations of our platform, ad-hoc ingestion scripts (Cron jobs) led to "silent failures," where data gaps went undetected for days, eroding organizational trust. This paper reports on our experience migrating from legacy scheduling to a robust Extract-Load-Transform (ELT) pipeline using Directed Acyclic Graph (DAG) orchestration and Medallion Architecture. We detail the operational benefits of decoupling data extraction from transformation, the necessity of immutable raw history for metric redefinition, and the implementation of state-based dependency management. Our experience suggests that treating the metrics pipeline as a production-grade distributed system is a prerequisite for sustainable engineering analytics.

Generative Flow Networks (GFlowNets) learn to sample diverse candidates in proportion to a reward function, making them well-suited for scientific discovery, where exploring multiple promising solutions is crucial. Further extending GFlowNets to multi-objective settings has attracted growing interest since real-world applications often involve multiple, conflicting objectives. However, existing approaches require additional training for each set of objectives, limiting their applicability and incurring substantial computational overhead. We propose a training-free mixing policy that composes pre-trained GFlowNets at inference time, enabling rapid adaptation without finetuning or retraining. Importantly, our framework is flexible, capable of handling diverse reward combinations ranging from linear scalarization to complex non-linear logical operators, which are often handled separately in previous literature. We prove that our method exactly recovers the target distribution for linear scalarization and quantify the approximation quality for nonlinear operators through a distortion factor. Experiments on a synthetic 2D grid and real-world molecule-generation tasks demonstrate that our approach achieves performance comparable to baselines that require additional training.

When designing compound AI systems, a common approach is to query multiple copies of the same model and aggregate the responses to produce a synthesized output. Given the homogeneity of these models, this raises the question of whether aggregation unlocks access to a greater set of outputs than querying a single model. In this work, we investigate the power and limitations of aggregation within a stylized principal-agent framework. This framework models how the system designer can partially steer each agent's output through its reward function specification, but still faces limitations due to prompt engineering ability and model capabilities. Our analysis uncovers three natural mechanisms -- feasibility expansion, support expansion, and binding set contraction -- through which aggregation expands the set of outputs that are elicitable by the system designer. We prove that any aggregation operation must implement one of these mechanisms in order to be elicitability-expanding, and that strengthened versions of these mechanisms provide necessary and sufficient conditions that fully characterize elicitability-expansion. Finally, we provide an empirical illustration of our findings for LLMs deployed in a toy reference-generation task. Altogether, our results take a step towards characterizing when compound AI systems can overcome limitations in model capabilities and in prompt engineering.

Retrieval-Augmented Generation (RAG) systems rely critically on the retriever module to surface relevant context for large language models. Although numerous retrievers have recently been proposed, each built on different ranking principles such as lexical matching, dense embeddings, or graph citations, there remains a lack of systematic understanding of how these mechanisms differ and overlap. Existing benchmarks primarily compare entire RAG pipelines or introduce new datasets, providing little guidance on selecting or combining retrievers themselves. Those that do compare retrievers directly use a limited set of evaluation tools which fail to capture complementary and overlapping strengths. This work presents MIGRASCOPE, a Mutual Information based RAG Retriever Analysis Scope. We revisit state-of-the-art retrievers and introduce principled metrics grounded in information and statistical estimation theory to quantify retrieval quality, redundancy, synergy, and marginal contribution. We further show that if chosen carefully, an ensemble of retrievers outperforms any single retriever. We leverage the developed tools over major RAG corpora to provide unique insights on contribution levels of the state-of-the-art retrievers. Our findings provide a fresh perspective on the structure of modern retrieval techniques and actionable guidance for designing robust and efficient RAG systems.

Learning PDE dynamics for fluids increasingly relies on neural operators and Transformer-based models, yet these approaches often lack interpretability and struggle with localized, high-frequency structures while incurring quadratic cost in spatial samples. We propose representing fields with a Gaussian basis, where learned atoms carry explicit geometry (centers, anisotropic scales, weights) and form a compact, mesh-agnostic, directly visualizable state. Building on this representation, we introduce a Gaussian Particle Operator that acts in modal space: learned Gaussian modal windows perform a Petrov-Galerkin measurement, and PG Gaussian Attention enables global cross-scale coupling. This basis-to-basis design is resolution-agnostic and achieves near-linear complexity in N for a fixed modal budget, supporting irregular geometries and seamless 2D-to-3D extension. On standard PDE benchmarks and real datasets, our method attains state-of-the-art competitive accuracy while providing intrinsic interpretability.

Gene expression prediction, which predicts mRNA expression levels from DNA sequences, presents significant challenges. Previous works often focus on extending input sequence length to locate distal enhancers, which may influence target genes from hundreds of kilobases away. Our work first reveals that for current models, long sequence modeling can decrease performance. Even carefully designed algorithms only mitigate the performance degradation caused by long sequences. Instead, we find that proximal multimodal epigenomic signals near target genes prove more essential. Hence we focus on how to better integrate these signals, which has been overlooked. We find that different signal types serve distinct biological roles, with some directly marking active regulatory elements while others reflect background chromatin patterns that may introduce confounding effects. Simple concatenation may lead models to develop spurious associations with these background patterns. To address this challenge, we propose Prism, a framework that learns multiple combinations of high-dimensional epigenomic features to represent distinct background chromatin states and uses backdoor adjustment to mitigate confounding effects. Our experimental results demonstrate that proper modeling of multimodal epigenomic signals achieves state-of-the-art performance using only short sequences for gene expression prediction.

The Flexible Job Shop Problem (FJSP) is a well-studied combinatorial optimization problem with extensive applications for manufacturing and production scheduling. It involves assigning jobs to various machines to optimize criteria, such as minimizing total completion time. Current learning-based methods in this domain often rely on localized feature extraction models, limiting their capacity to capture overarching dependencies spanning operations and machines. This paper introduces an innovative architecture that harnesses Mamba, a state-space model with linear computational complexity, to facilitate comprehensive sequence modeling tailored for FJSP. In contrast to prevalent graph-attention-based frameworks that are computationally intensive for FJSP, we show our model is more efficient. Specifically, the proposed model possesses an encoder and a decoder. The encoder incorporates a dual Mamba block to extract operation and machine features separately. Additionally, we introduce an efficient cross-attention decoder to learn interactive embeddings of operations and machines. Our experimental results demonstrate that our method achieves faster solving speed and surpasses the performance of state-of-the-art learning-based methods for FJSP across various benchmarks.

The Muon optimizer has demonstrated promising performance in pre-training large language models through gradient (or momentum) orthogonalization. In this work, we propose a simple yet effective enhancement to Muon, namely Muon+, which introduces an additional normalization step after orthogonalization. We demonstrate the effectiveness of Muon+ through extensive pre-training experiments across a wide range of model scales and architectures. Our evaluation includes GPT-style models ranging from 130M to 774M parameters and LLaMA-style models ranging from 60M to 1B parameters. We comprehensively evaluate the effectiveness of Muon+ in the compute-optimal training regime and further extend the token-to-parameter (T2P) ratio to an industrial level of $\approx 200$. Experimental results show that Muon+ provides a consistent boost on training and validation perplexity over Muon. We provide our code here: https://github.com/K1seki221/MuonPlus.

Multilingual pretraining typically lacks explicit alignment signals, leading to suboptimal cross-lingual alignment in the representation space. In this work, we show that training standard pretrained models for cross-lingual alignment with a multi-way parallel corpus in a diverse pool of languages can substantially improve multilingual and cross-lingual representations for NLU tasks. We construct a multi-way parallel dataset using translations of English text from an off-the-shelf NMT model for a pool of six target languages and achieve strong cross-lingual alignment through contrastive learning. This leads to substantial performance gains across both seen and unseen languages for multiple tasks from the MTEB benchmark evaluated for XLM-Roberta and multilingual BERT base models. Using a multi-way parallel corpus for contrastive training yields substantial gains on bitext mining (21.3%), semantic similarity (5.3%), and classification (28.4%) compared to English-centric (En-X) bilingually parallel data, where X is sampled from a pool of multiple target languages. Furthermore, finetuning mE5 model on a small dataset with multi-way parallelism significantly improves bitext mining compared to one without, underscoring the importance of multi-way cross-lingual supervision even for models already pretrained for high-quality sentence embeddings.

Agentic reinforcement learning (ARL) has rapidly gained attention as a promising paradigm for training agents to solve complex, multi-step interactive tasks. Despite encouraging early results, ARL remains highly unstable, often leading to training collapse. This instability limits scalability to larger environments and longer interaction horizons, and constrains systematic exploration of algorithmic design choices. In this paper, we first propose ARLArena, a stable training recipe and systematic analysis framework that examines training stability in a controlled and reproducible setting. ARLArena first constructs a clean and standardized testbed. Then, we decompose policy gradient into four core design dimensions and assess the performance and stability of each dimension. Through this fine-grained analysis, we distill a unified perspective on ARL and propose SAMPO, a stable agentic policy optimization method designed to mitigate the dominant sources of instability in ARL. Empirically, SAMPO achieves consistently stable training and strong performance across diverse agentic tasks. Overall, this study provides a unifying policy gradient perspective for ARL and offers practical guidance for building stable and reproducible LLM-based agent training pipelines.

Large language models (LLMs) are becoming increasingly applied beyond natural language processing, demonstrating strong capabilities in complex scientific tasks that traditionally require human expertise. This progress has extended into materials discovery, where LLMs introduce a new paradigm by leveraging reasoning and in-context learning, capabilities absent from conventional machine learning approaches. Here, we present a Multi-Agent-based Electrocatalyst Search Through Reasoning and Optimization (MAESTRO) framework in which multiple LLMs with specialized roles collaboratively discover high-performance single atom catalysts for the oxygen reduction reaction. Within an autonomous design loop, agents iteratively reason, propose modifications, reflect on results and accumulate design history. Through in-context learning enabled by this iterative process, MAESTRO identified design principles not explicitly encoded in the LLMs' background knowledge and successfully discovered catalysts that break conventional scaling relations between reaction intermediates. These results highlight the potential of multi-agent LLM frameworks as a powerful strategy to generate chemical insight and discover promising catalysts.

General-purpose robots must master long-horizon manipulation, defined as tasks involving multiple kinematic structure changes (e.g., attaching or detaching objects) in unstructured environments. While Vision-Language-Action (VLA) models offer the potential to master diverse atomic skills, they struggle with the combinatorial complexity of sequencing them and are prone to cascading failures due to environmental sensitivity. To address these challenges, we propose LiLo-VLA (Linked Local VLA), a modular framework capable of zero-shot generalization to novel long-horizon tasks without ever being trained on them. Our approach decouples transport from interaction: a Reaching Module handles global motion, while an Interaction Module employs an object-centric VLA to process isolated objects of interest, ensuring robustness against irrelevant visual features and invariance to spatial configurations. Crucially, this modularity facilitates robust failure recovery through dynamic replanning and skill reuse, effectively mitigating the cascading errors common in end-to-end approaches. We introduce a 21-task simulation benchmark consisting of two challenging suites: LIBERO-Long++ and Ultra-Long. In these simulations, LiLo-VLA achieves a 69% average success rate, outperforming Pi0.5 by 41% and OpenVLA-OFT by 67%. Furthermore, real-world evaluations across 8 long-horizon tasks demonstrate an average success rate of 85%. Project page: https://yy-gx.github.io/LiLo-VLA/.

Deciphering brain function through non-invasive recordings requires synthesizing complementary high-frequency electromagnetic (EEG/MEG) and low-frequency metabolic (fMRI) signals. However, despite their shared neural origins, extreme discrepancies have traditionally confined these modalities to isolated analysis pipelines, hindering a holistic interpretation of brain activity. To bridge this fragmentation, we introduce \textbf{NOBEL}, a \textbf{n}euro-\textbf{o}mni-modal \textbf{b}rain-\textbf{e}ncoding \textbf{l}arge language model (LLM) that unifies these heterogeneous signals within the LLM's semantic embedding space. Our architecture integrates a unified encoder for EEG and MEG with a novel dual-path strategy for fMRI, aligning non-invasive brain signals and external sensory stimuli into a shared token space, then leverages an LLM as a universal backbone. Extensive evaluations demonstrate that NOBEL serves as a robust generalist across standard single-modal tasks. We also show that the synergistic fusion of electromagnetic and metabolic signals yields higher decoding accuracy than unimodal baselines, validating the complementary nature of multiple neural modalities. Furthermore, NOBEL exhibits strong capabilities in stimulus-aware decoding, effectively interpreting visual semantics from multi-subject fMRI data on the NSD and HAD datasets while uniquely leveraging direct stimulus inputs to verify causal links between sensory signals and neural responses. NOBEL thus takes a step towards unifying non-invasive brain decoding, demonstrating the promising potential of omni-modal brain understanding.

Many emerging agentic paradigms require agents to collaborate with one another (or people) to achieve shared goals. Unfortunately, existing approaches to learning policies for such collaborative problems produce brittle solutions that fail when paired with new partners. We attribute these failures to a combination of free-riding during training and a lack of strategic robustness. To address these problems, we study the concept of strategic risk aversion and interpret it as a principled inductive bias for generalizable cooperation with unseen partners. While strategically risk-averse players are robust to deviations in their partner's behavior by design, we show that, in collaborative games, they also (1) can have better equilibrium outcomes than those at classical game-theoretic concepts like Nash, and (2) exhibit less or no free-riding. Inspired by these insights, we develop a multi-agent reinforcement learning (MARL) algorithm that integrates strategic risk aversion into standard policy optimization methods. Our empirical results across collaborative benchmarks (including an LLM collaboration task) validate our theory and demonstrate that our approach consistently achieves reliable collaboration with heterogeneous and previously unseen partners across collaborative tasks.

The goal of fair clustering is to find clusters such that the proportion of sensitive attributes (e.g., gender, race, etc.) in each cluster is similar to that of the entire dataset. Various fair clustering algorithms have been proposed that modify standard K-means clustering to satisfy a given fairness constraint. A critical limitation of several existing fair clustering algorithms is that the number of parameters to be learned is proportional to the sample size because the cluster assignment of each datum should be optimized simultaneously with the cluster center, and thus scaling up the algorithms is difficult. In this paper, we propose a new fair clustering algorithm based on a finite mixture model, called Fair Model-based Clustering (FMC). A main advantage of FMC is that the number of learnable parameters is independent of the sample size and thus can be scaled up easily. In particular, mini-batch learning is possible to obtain clusters that are approximately fair. Moreover, FMC can be applied to non-metric data (e.g., categorical data) as long as the likelihood is well-defined. Theoretical and empirical justifications for the superiority of the proposed algorithm are provided.

Robust watermarking is critical for intellectual property protection, whereas existing methods face a severe vulnerability against regeneration-based AIGC attacks. We identify that existing methods fail because they entangle the watermark with high-frequency cover texture, which is susceptible to being rewritten during generative purification. To address this, we propose WaterVIB, a theoretically grounded framework that reformulates the encoder as an information sieve via the Variational Information Bottleneck. Instead of overfitting to fragile cover details, our approach forces the model to learn a Minimal Sufficient Statistic of the message. This effectively filters out redundant cover nuances prone to generative shifts, retaining only the essential signal invariant to regeneration. We theoretically prove that optimizing this bottleneck is a necessary condition for robustness against distribution-shifting attacks. Extensive experiments demonstrate that WaterVIB significantly outperforms state-of-the-art methods, achieving superior zero-shot resilience against unknown diffusion-based editing.

This guide develops high-probability regret bounds for empirical risk minimization (ERM). The presentation is modular: we state broadly applicable guarantees under high-level conditions and give tools for verifying them for specific losses and function classes. We emphasize that many ERM rate derivations can be organized around a three-step recipe -- a basic inequality, a uniform local concentration bound, and a fixed-point argument -- which yields regret bounds in terms of a critical radius, defined via localized Rademacher complexity, under a mild Bernstein-type variance--risk condition. To make these bounds concrete, we upper bound the critical radius using local maximal inequalities and metric-entropy integrals, recovering familiar rates for VC-subgraph, Sobolev/Hölder, and bounded-variation classes. We also review ERM with nuisance components -- including weighted ERM and Neyman-orthogonal losses -- as they arise in causal inference, missing data, and domain adaptation. Following the orthogonal learning framework, we highlight that these problems often admit regret-transfer bounds linking regret under an estimated loss to population regret under the target loss. These bounds typically decompose regret into (i) statistical error under the estimated (optimized) loss and (ii) approximation error due to nuisance estimation. Under sample splitting or cross-fitting, the first term can be controlled using standard fixed-loss ERM regret bounds, while the second term depends only on nuisance-estimation accuracy. We also treat the in-sample regime, where nuisances and the ERM are fit on the same data, deriving regret bounds and giving sufficient conditions for fast rates.

Irregular Multivariate Time Series (IMTS) are characterized by uneven intervals between consecutive timestamps, which carry sampling pattern information valuable and informative for learning temporal and variable dependencies. In addition, IMTS often exhibit diverse dependencies across multiple time scales. However, many existing multi-scale IMTS methods use resampling to obtain the coarse series, which can alter the original timestamps and disrupt the sampling pattern information. To address the challenge, we propose ReIMTS, a Recursive multi-scale modeling approach for Irregular Multivariate Time Series forecasting. Instead of resampling, ReIMTS keeps timestamps unchanged and recursively splits each sample into subsamples with progressively shorter time periods. Based on the original sampling timestamps in these long-to-short subsamples, an irregularity-aware representation fusion mechanism is proposed to capture global-to-local dependencies for accurate forecasting. Extensive experiments demonstrate an average performance improvement of 27.1\% in the forecasting task across different models and real-world datasets. Our code is available at https://github.com/Ladbaby/PyOmniTS.

While defenses for structured PII are mature, Large Language Models (LLMs) pose a new threat: Semantic Sensitive Information (SemSI), where models infer sensitive identity attributes, generate reputation-harmful content, or hallucinate potentially wrong information. The capacity of LLMs to self-regulate these complex, context-dependent sensitive information leaks without destroying utility remains an open scientific question. To address this, we introduce SemSIEdit, an inference-time framework where an agentic "Editor" iteratively critiques and rewrites sensitive spans to preserve narrative flow rather than simply refusing to answer. Our analysis reveals a Privacy-Utility Pareto Frontier, where this agentic rewriting reduces leakage by 34.6% across all three SemSI categories while incurring a marginal utility loss of 9.8%. We also uncover a Scale-Dependent Safety Divergence: large reasoning models (e.g., GPT-5) achieve safety through constructive expansion (adding nuance), whereas capacity-constrained models revert to destructive truncation (deleting text). Finally, we identify a Reasoning Paradox: while inference-time reasoning increases baseline risk by enabling the model to make deeper sensitive inferences, it simultaneously empowers the defense to execute safe rewrites.

Reinforcement learning (RL) has become a central post-training paradigm for large language models (LLMs), but its performance is highly sensitive to the quality of training problems. This sensitivity stems from the non-stationarity of RL: rollouts are generated by an evolving policy, and learning is shaped by exploration and reward feedback, unlike supervised fine-tuning (SFT) with fixed trajectories. As a result, prior work often relies on manual curation or simple heuristic filters (e.g., accuracy), which can admit incorrect or low-utility problems. We propose GradAlign, a gradient-aligned data selection method for LLM reinforcement learning that uses a small, trusted validation set to prioritize training problems whose policy gradients align with validation gradients, yielding an adaptive curriculum. We evaluate GradAlign across three challenging data regimes: unreliable reward signals, distribution imbalance, and low-utility training corpus, showing that GradAlign consistently outperforms existing baselines, underscoring the importance of directional gradient signals in navigating non-stationary policy optimization and yielding more stable training and improved final performance. We release our implementation at https://github.com/StigLidu/GradAlign

In AI, most evaluations of natural language understanding tasks are conducted in standardized dialects such as Standard American English (SAE). In this work, we investigate how accurately large language models (LLMs) represent African American Vernacular English (AAVE). We analyze three LLMs to compare their usage of AAVE to the usage of humans who natively speak AAVE. We first analyzed interviews from the Corpus of Regional African American Language and TwitterAAE to identify the typical contexts where people use AAVE grammatical features such as ain't. We then prompted the LLMs to produce text in AAVE and compared the model-generated text to human usage patterns. We find that, in many cases, there are substantial differences between AAVE usage in LLMs and humans: LLMs usually underuse and misuse grammatical features characteristic of AAVE. Furthermore, through sentiment analysis and manual inspection, we found that the models replicated stereotypes about African Americans. These results highlight the need for more diversity in training data and the incorporation of fairness methods to mitigate the perpetuation of stereotypes.

Text-to-SQL and Big Data are both extensively benchmarked fields, yet there is limited research that evaluates them jointly. In the real world, Text-to-SQL systems are often embedded with Big Data workflows, such as large-scale data processing or interactive data analytics. We refer to this as "Text-to-Big SQL". However, existing text-to-SQL benchmarks remain narrowly scoped and overlook the cost and performance implications that arise at scale. For instance, translation errors that are minor on small datasets lead to substantial cost and latency overheads as data scales, a relevant issue completely ignored by text-to-SQL metrics. In this paper, we overcome this overlooked challenge by introducing novel and representative metrics for evaluating Text-to-Big SQL. Our study focuses on production-level LLM agents, a database-agnostic system adaptable to diverse user needs. Via an extensive evaluation of frontier models, we show that text-to-SQL metrics are insufficient for Big Data. In contrast, our proposed text-to-Big SQL metrics accurately reflect execution efficiency, cost, and the impact of data scale. Furthermore, we provide LLM-specific insights, including fine-grained, cross-model comparisons of latency and cost.

Across many risk-sensitive areas, it is critical to continuously audit the performance of machine learning systems and detect any unusual behavior quickly. This can be modeled as a sequential hypothesis testing problem with $k$ incoming streams of data and a global null hypothesis that asserts that the system is working as expected across all $k$ streams. The standard global test employs a Bonferroni correction and has an expected stopping time bound of $O\left(\ln\frac{k}α\right)$ when $k$ is large and the significance level of the test, $α$, is small. In this work, we construct new sequential tests by using ideas of merging test martingales with different trade-offs in expected stopping times under different, sparse or dense alternative hypotheses. We further derive a new, balanced test that achieves an improved expected stopping time bound that matches Bonferroni's in the sparse setting but that naturally results in $O\left(\frac{1}{k}\ln\frac{1}α\right)$ under a dense alternative. We empirically demonstrate the effectiveness of our proposed tests on synthetic and real-world data.

We study inference on scalar-valued pathwise differentiable targets after adaptive data collection, such as a bandit algorithm. We introduce a novel target-specific condition, directional stability, which is strictly weaker than previously imposed target-agnostic stability conditions. Under directional stability, we show that estimators that would have been efficient under i.i.d. data remain asymptotically normal and semiparametrically efficient when computed from adaptively collected trajectories. The canonical gradient has a martingale form, and directional stability guarantees stabilization of its predictable quadratic variation, enabling high-dimensional asymptotic normality. We characterize efficiency using a convolution theorem for the adaptive-data setting, and give a condition under which the one-step estimator attains the efficiency bound. We verify directional stability for LinUCB, yielding the first semiparametric efficiency guarantee for a regular scalar target under LinUCB sampling.

We propose a knowledge-driven, model-based approach to segmenting audio into single-category and mixed-category chunks with applications to source separation. "Knowledge" here denotes information associated with the data, such as music scores. "Model" here refers to tool that can be used for audio segmentation and recognition, such as hidden Markov models. In contrast to conventional learning that often relies on annotated data with given segment categories and their corresponding boundaries to guide the learning process, the proposed framework does not depend on any pre-segmented training data and learns directly from the input audio and its related knowledge sources to build all necessary models autonomously. Evaluation on simulation data shows that score-guided learning achieves very good music segmentation and separation results. Tested on movie track data for cinematic audio source separation also shows that utilizing sound category knowledge achieves better separation results than those obtained with data-driven techniques without using such information.

Discrete diffusion models have emerged as strong alternatives to autoregressive language models, with recent work initializing and fine-tuning a base unimodal model for bimodal generation. Diverging from previous approaches, we introduce the first tri-modal masked diffusion model pretrained from scratch on text, image-text, and audio-text data. We systematically analyze multimodal scaling laws, modality mixing ratios, noise schedules, and batch-size effects, and we provide optimized inference sampling defaults. Our batch-size analysis yields a novel stochastic differential equation (SDE)-based reparameterization that eliminates the need for tuning the optimal batch size as reported in recent work. This reparameterization decouples the physical batch size, often chosen based on compute constraints (GPU saturation, FLOP efficiency, wall-clock time), from the logical batch size, chosen to balance gradient variance during stochastic optimization. Finally, we pretrain a preliminary 3B-parameter tri-modal model on 6.4T tokens, demonstrating the capabilities of a unified design and achieving strong results in text generation, text-to-image tasks, and text-to-speech tasks. Our work represents the largest-scale systematic open study of multimodal discrete diffusion models conducted to date, providing insights into scaling behaviors across multiple modalities.

Data assimilation and scientific inverse problems require reconstructing high-dimensional physical states from sparse and noisy observations, ideally with uncertainty-aware posterior samples that remain faithful to learned priors and governing physics. While training-free conditional generation is well developed for diffusion models, corresponding conditioning and posterior sampling strategies for Flow Matching (FM) priors remain comparatively under-explored, especially on scientific benchmarks where fidelity must be assessed beyond measurement misfit. In this work, we study training-free conditional generation for scientific inverse problems under FM priors and organize existing inference-time strategies by where measurement information is injected: (i) guided transport dynamics that perturb sampling trajectories using likelihood information, and (ii) source-distribution inference that performs posterior inference over the source variable while keeping the learned transport fixed. Building on the latter, we propose D-Flow SGLD, a source-space posterior sampling method that augments differentiable source inference with preconditioned stochastic gradient Langevin dynamics, enabling scalable exploration of the source posterior induced by new measurement operators without retraining the prior or modifying the learned FM dynamics. We benchmark representative methods from both families on a hierarchy of problems: 2D toy posteriors, chaotic Kuramoto-Sivashinsky trajectories, and wall-bounded turbulence reconstruction. Across these settings, we quantify trade-offs among measurement assimilation, posterior diversity, and physics/statistics fidelity, and establish D-Flow SGLD as a practical FM-compatible posterior sampler for scientific inverse problems.

The spin-$1/2$ $J_1$-$J_2$ Heisenberg model on the square lattice exhibits a debated intermediate phase between Néel antiferromagnetic and stripe ordered regimes, with competing theories proposing plaquette valence bond, nematic, and quantum spin liquid ground states. We apply the Prometheus variational autoencoder framework -- previously validated on classical (2D, 3D Ising) and quantum (disordered transverse field Ising) phase transitions -- to systematically explore the $J_1$-$J_2$ phase diagram via unsupervised analysis of exact diagonalization ground states for a $4 \times 4$ lattice. Through dense parameter scans of $J_2/J_1 \in [0.3, 0.7]$ with step size 0.01 and comprehensive latent space analysis, we investigate the nature of the intermediate regime using unsupervised order parameter discovery and critical point detection via multiple independent methods. This work demonstrates the application of rigorously validated machine learning methods to open questions in frustrated quantum magnetism, where traditional order parameter identification is challenged by competing interactions and limited accessible system sizes.

A key challenge in artificial intelligence and neuroscience is understanding how neural systems learn representations that capture the underlying dynamics of the world. Most world models represent the transition function with unstructured neural networks, limiting interpretability, sample efficiency, and generalization to unseen states or action compositions. We address these issues with a generalizable world model grounded in Vector Symbolic Architecture (VSA) principles as geometric priors. Our approach utilizes learnable Fourier Holographic Reduced Representation (FHRR) encoders to map states and actions into a high dimensional complex vector space with learned group structure and models transitions with element-wise complex multiplication. We formalize the framework's group theoretic foundation and show how training such structured representations to be approximately invariant enables strong multi-step composition directly in latent space and generalization performances over various experiments. On a discrete grid world environment, our model achieves 87.5% zero shot accuracy to unseen state-action pairs, obtains 53.6% higher accuracy on 20-timestep horizon rollouts, and demonstrates 4x higher robustness to noise relative to an MLP baseline. These results highlight how training to have latent group structure yields generalizable, data-efficient, and interpretable world models, providing a principled pathway toward structured models for real-world planning and reasoning.

$E(3)$-equivariant neural networks have proven to be effective in a wide range of 3D modeling tasks. A fundamental operation of such networks is the tensor product, which allows interaction between different feature types. Because this operation scales poorly, there has been considerable work towards accelerating this interaction. However, recently \citet{xieprice} have pointed out that most speedups come from a reduction in expressivity rather than true algorithmic improvements on computing Clebsch-Gordan tensor products. A modification of Gaunt tensor product \citep{gaunt} can give a true asymptotic speedup but is incomplete and misses many interactions. In this work, we provide the first complete algorithm which truly provides asymptotic benefits Clebsch-Gordan tensor products. For full CGTP, our algorithm brings runtime complexity from the naive $O(L^6)$ to $O(L^4\log^2 L)$, close to the lower bound of $O(L^4)$. We first show how generalizing fast Fourier based convolution naturally leads to the previously proposed Gaunt tensor product \citep{gaunt}. To remedy antisymmetry issues, we generalize from scalar signals to irrep valued signals, giving us tensor spherical harmonics. We prove a generalized Gaunt formula for the tensor harmonics. Finally, we show that we only need up to vector valued signals to recover the missing interactions of Gaunt tensor product.

This work presents iMiGUE-Speech, an extension of the iMiGUE dataset that provides a spontaneous affective corpus for studying emotional and affective states. The new release focuses on speech and enriches the original dataset with additional metadata, including speech transcripts, speaker-role separation between interviewer and interviewee, and word-level forced alignments. Unlike existing emotional speech datasets that rely on acted or laboratory-elicited emotions, iMiGUE-Speech captures spontaneous affect arising naturally from real match outcomes. To demonstrate the utility of the dataset and establish initial benchmarks, we introduce two evaluation tasks for comparative assessment: speech emotion recognition and transcript-based sentiment analysis. These tasks leverage state-of-the-art pre-trained representations to assess the dataset's ability to capture spontaneous affective states from both acoustic and linguistic modalities. iMiGUE-Speech can also be synchronously paired with micro-gesture annotations from the original iMiGUE dataset, forming a uniquely multimodal resource for studying speech-gesture affective dynamics. The extended dataset is available at https://github.com/CV-AC/imigue-speech.

We describe extensive numerical experiments assessing and quantifying how classifier performance depends on the quality of the training data, a frequently neglected component of the analysis of classifiers. More specifically, in the scientific context of metagenomic assembly of short DNA reads into "contigs," we examine the effects of degrading the quality of the training data by multiple mechanisms, and for four classifiers -- Bayes classifiers, neural nets, partition models and random forests. We investigate both individual behavior and congruence among the classifiers. We find breakdown-like behavior that holds for all four classifiers, as degradation increases and they move from being mostly correct to only coincidentally correct, because they are wrong in the same way. In the process, a picture of spatial heterogeneity emerges: as the training data move farther from analysis data, classifier decisions degenerate, the boundary becomes less dense, and congruence increases.

Vector glyphs are the atomic units of digital typography, yet most learning-based pipelines still depend on carefully curated exemplar sheets and raster-to-vector postprocessing, which limits accessibility and editability. We introduce VecGlypher, a single multimodal language model that generates high-fidelity vector glyphs directly from text descriptions or image exemplars. Given a style prompt, optional reference glyph images, and a target character, VecGlypher autoregressively emits SVG path tokens, avoiding raster intermediates and producing editable, watertight outlines in one pass. A typography-aware data and training recipe makes this possible: (i) a large-scale continuation stage on 39K noisy Envato fonts to master SVG syntax and long-horizon geometry, followed by (ii) post-training on 2.5K expert-annotated Google Fonts with descriptive tags and exemplars to align language and imagery with geometry; preprocessing normalizes coordinate frames, canonicalizes paths, de-duplicates families, and quantizes coordinates for stable long-sequence decoding. On cross-family OOD evaluation, VecGlypher substantially outperforms both general-purpose LLMs and specialized vector-font baselines for text-only generation, while image-referenced generation reaches a state-of-the-art performance, with marked gains over DeepVecFont-v2 and DualVector. Ablations show that model scale and the two-stage recipe are critical and that absolute-coordinate serialization yields the best geometry. VecGlypher lowers the barrier to font creation by letting users design with words or exemplars, and provides a scalable foundation for future multimodal design tools.

Deep research has emerged as an important task that aims to address hard queries through extensive open-web exploration. To tackle it, most prior work equips large language model (LLM)-based agents with opaque web search APIs, enabling agents to iteratively issue search queries, retrieve external evidence, and reason over it. Despite search's essential role in deep research, black-box web search APIs hinder systematic analysis of search components, leaving the behaviour of established text ranking methods in deep research largely unclear. To fill this gap, we reproduce a selection of key findings and best practices for IR text ranking methods in the deep research setting. In particular, we examine their effectiveness from three perspectives: (i) retrieval units (documents vs. passages), (ii) pipeline configurations (different retrievers, re-rankers, and re-ranking depths), and (iii) query characteristics (the mismatch between agent-issued queries and the training queries of text rankers). We perform experiments on BrowseComp-Plus, a deep research dataset with a fixed corpus, evaluating 2 open-source agents, 5 retrievers, and 3 re-rankers across diverse setups. We find that agent-issued queries typically follow web-search-style syntax (e.g., quoted exact matches), favouring lexical, learned sparse, and multi-vector retrievers; passage-level units are more efficient under limited context windows, and avoid the difficulties of document length normalisation in lexical retrieval; re-ranking is highly effective; translating agent-issued queries into natural-language questions significantly bridges the query mismatch.

Recurrent neural networks (RNNs) can be interpreted as discrete-time state-space models, where the state evolution corresponds to an infinite-impulse-response (IIR) filtering operation governed by both feedforward weights and recurrent poles. While, in principle, all parameters including pole locations can be optimized via backpropagation through time (BPTT), such joint learning incurs substantial computational overhead and is often impractical for applications with limited training data. Echo state networks (ESNs) mitigate this limitation by fixing the recurrent dynamics and training only a linear readout, enabling efficient and stable online adaptation. In this work, we analytically and empirically examine why learning recurrent poles does not provide tangible benefits in data-constrained, real-time learning scenarios. Our analysis shows that pole learning renders the weight optimization problem highly non-convex, requiring significantly more training samples and iterations for gradient-based methods to converge to meaningful solutions. Empirically, we observe that for complex-valued data, gradient descent frequently exhibits prolonged plateaus, and advanced optimizers offer limited improvement. In contrast, fixed-pole architectures induce stable and well-conditioned state representations even with limited training data. Numerical results demonstrate that fixed-pole networks achieve superior performance with lower training complexity, making them more suitable for online real-time tasks.

Introduction: Deep learning-based segmentation models are increasingly integrated into clinical imaging workflows, yet their robustness to adversarial perturbations remains incompletely characterized, particularly for ultrasound images. We evaluated adversarial attacks and inference-time defenses for thyroid nodule segmentation in B-mode ultrasound. Methods: Two black-box adversarial attacks were developed: (1) Structured Speckle Amplification Attack (SSAA), which injects boundary-targeted noise, and (2) Frequency-Domain Ultrasound Attack (FDUA), which applies bandpass-filtered phase perturbations in the Fourier domain. Three inference-time mitigations were evaluated on adversarial images: randomized preprocessing with test-time augmentation, deterministic input denoising, and stochastic ensemble inference with consistency-aware aggregation. Experiments were conducted on a U-Net segmentation model trained on cine-clips from a database of 192 thyroid nodules. Results: The baseline model achieved a mean Dice similarity coefficient (DSC) of 0.76 (SD 0.20) on unperturbed images. SSAA reduced DSC by 0.29 (SD 0.20) while maintaining high visual similarity (SSIM = 0.94). FDUA resulted in a smaller DSC reduction of 0.11 (SD 0.09) with lower visual fidelity (SSIM = 0.82). Against SSAA, all three defenses significantly improved DSC after correction, with deterministic denoising showing the largest recovery (+0.10, p < 0.001), followed by randomized preprocessing (+0.09, p < 0.001), and stochastic ensemble inference (+0.08, p = 0.002). No defense achieved statistically significant improvement against FDUA. Conclusion: Spatial-domain adversarial perturbations in ultrasound segmentation showed partial mitigation with input preprocessing, whereas frequency-domain perturbations were not mitigated by the defenses, highlighting modality-specific challenges in adversarial robustness evaluation.

Current stateless defences for multimodal agentic RAG fail to detect adversarial strategies that distribute malicious semantics across retrieval, planning, and generation components. We formulate this security challenge as a Partially Observable Markov Decision Process (POMDP), where adversarial intent is a latent variable inferred from noisy multi-stage observations. We introduce MMA-RAG^T, an inference-time control framework governed by a Modular Trust Agent (MTA) that maintains an approximate belief state via structured LLM reasoning. Operating as a model-agnostic overlay, MMA-RAGT mediates a configurable set of internal checkpoints to enforce stateful defence-in-depth. Extensive evaluation on 43,774 instances demonstrates a 6.50x average reduction factor in Attack Success Rate relative to undefended baselines, with negligible utility cost. Crucially, a factorial ablation validates our theoretical bounds: while statefulness and spatial coverage are individually necessary (26.4 pp and 13.6 pp gains respectively), stateless multi-point intervention can yield zero marginal benefit under homogeneous stateless filtering when checkpoint detections are perfectly correlated.

Hyperdimensional Computing (HDC) offers a computationally efficient paradigm for neuromorphic learning. Yet, it lacks rigorous uncertainty quantification, leading to open decision boundaries and, consequently, vulnerability to outliers, adversarial perturbations, and out-of-distribution inputs. To address these limitations, we introduce ConformalHDC, a unified framework that combines the statistical guarantees of conformal prediction with the computational efficiency of HDC. For this framework, we propose two complementary variations. First, the set-valued formulation provides finite-sample, distribution-free coverage guarantees. Using carefully designed conformity scores, it forms enclosed decision boundaries that improve robustness to non-conforming inputs. Second, the point-valued formulation leverages the same conformity scores to produce a single prediction when desired, potentially improving accuracy over traditional HDC by accounting for class interactions. We demonstrate the broad applicability of the proposed framework through evaluations on multiple real-world datasets. In particular, we apply our method to the challenging problem of decoding non-spatial stimulus information from the spiking activity of hippocampal neurons recorded as subjects performed a sequence memory task. Our results show that ConformalHDC not only accurately decodes the stimulus information represented in the neural activity data, but also provides rigorous uncertainty estimates and correctly abstains when presented with data from other behavioral states. Overall, these capabilities position the framework as a reliable, uncertainty-aware foundation for neuromorphic computing.

The recent field of neural algorithmic reasoning (NAR) studies the ability of graph neural networks (GNNs) to emulate classical algorithms like Bellman-Ford, a phenomenon known as algorithmic alignment. At the same time, recent advances in large language models (LLMs) have spawned the study of mechanistic interpretability, which aims to identify granular model components like circuits that perform specific computations. In this work, we introduce Mechanistic Interpretability for Neural Algorithmic Reasoning (MINAR), an efficient circuit discovery toolbox that adapts attribution patching methods from mechanistic interpretability to the GNN setting. We show through two case studies that MINAR recovers faithful neuron-level circuits from GNNs trained on algorithmic tasks. Our study sheds new light on the process of circuit formation and pruning during training, as well as giving new insight into how GNNs trained to perform multiple tasks in parallel reuse circuit components for related tasks. Our code is available at https://github.com/pnnl/MINAR.

Multimodal Large Language Models (MLLMs) deliver detailed responses on vision-language tasks, yet remain susceptible to object hallucination (introducing objects not present in the image), undermining reliability in practice. Prior efforts often rely on heuristic penalties, post-hoc correction, or generic decoding tweaks, which do not directly intervene in the mechanisms that trigger object hallucination and thus yield limited gains. To address this challenge, we propose a causal decoding framework that applies targeted causal interventions during generation to curb spurious object mentions. By reshaping the decoding dynamics to attenuate spurious dependencies, our approach reduces false object tokens while maintaining descriptive quality. Across captioning and QA benchmarks, our framework substantially lowers object-hallucination rates and achieves state-of-the-art faithfulness without degrading overall output quality.

In this paper, we study last-iterate convergence of learning algorithms in bilinear saddle-point problems, a preferable notion of convergence that captures the day-to-day behavior of learning dynamics. We focus on the challenging setting where players select actions from compact convex sets and receive only bandit feedback. Our main contribution is the design of an uncoupled learning algorithm that guarantees last-iterate convergence to the Nash equilibrium with high probability. We establish a convergence rate of $\tilde{O}(T^{-1/4})$ up to polynomial factors in problem parameters. Crucially, our proposed algorithm is computationally efficient, requiring only an efficient linear optimization oracle over the players' compact action sets. The algorithm is obtained by combining techniques from experimental design and the classic Follow-The-Regularized-Leader (FTRL) framework, with a carefully chosen regularizer function tailored to the geometry of the action set of each learner.

Flow-based generative models, such as diffusion models and flow matching models, have achieved remarkable success in learning complex data distributions. However, a critical gap remains for their deployment in safety-critical domains: the lack of formal guarantees that generated samples will satisfy hard constraints. We address this by proposing a safety filtering framework that acts as an online shield for any pre-trained generative model. Our key insight is to cooperate with the generative process rather than override it. We define a constricting safety tube that is relaxed at the initial noise distribution and progressively tightens to the target safe set at the final data distribution, mirroring the coarse-to-fine structure of the generative process itself. By characterizing this tube via Control Barrier Functions (CBFs), we synthesize a feedback control input through a convex Quadratic Program (QP) at each sampling step. As the tube is loosest when noise is high and intervention is cheapest in terms of control energy, most constraint enforcement occurs when it least disrupts the model's learned structure. We prove that this mechanism guarantees safe sampling while minimizing the distributional shift from the original model at each sampling step, as quantified by the KL divergence. Our framework applies to any pre-trained flow-based generative scheme requiring no retraining or architectural modifications. We validate the approach across constrained image generation, physically-consistent trajectory sampling, and safe robotic manipulation policies, achieving 100% constraint satisfaction while preserving semantic fidelity.

Medical Vision Language Models (VLMs) can change their answers when clinicians rephrase the same question, which raises deployment risks. We introduce Paraphrase Sensitivity Failure (PSF)-Med, a benchmark of 19,748 chest Xray questions paired with about 92,000 meaningpreserving paraphrases across MIMIC-CXR and PadChest. Across six medical VLMs, we measure yes/no flips for the same image and find flip rates from 8% to 58%. However, low flip rate does not imply visual grounding: text-only baselines show that some models stay consistent even when the image is removed, suggesting they rely on language priors. To study mechanisms in one model, we apply GemmaScope 2 Sparse Autoencoders (SAEs) to MedGemma 4B and analyze FlipBank, a curated set of 158 flip cases. We identify a sparse feature at layer 17 that correlates with prompt framing and predicts decision margin shifts. In causal patching, removing this feature's contribution recovers 45% of the yesminus-no logit margin on average and fully reverses 15% of flips. Acting on this finding, we show that clamping the identified feature at inference reduces flip rates by 31% relative with only a 1.3 percentage-point accuracy cost, while also decreasing text-prior reliance. These results suggest that flip rate alone is not enough; robustness evaluations should test both paraphrase stability and image reliance.

We consider the problem of sampling from a posterior distribution arising in Bayesian inverse problems in science, engineering, and imaging. Our method belongs to the family of independence Metropolis-Hastings (IMH) sampling algorithms, which are common in Bayesian inference. Relying on the existence of an approximate posterior distribution that is cheaper to sample from but may have significant bias, we introduce Proximal-IMH, a scheme that removes this bias by correcting samples from the approximate posterior through an auxiliary optimization problem. This yields a local adjustment that trades off adherence to the exact model against stability around the approximate reference point. For idealized settings, we prove that the proximal correction tightens the match between approximate and exact posteriors, thereby improving acceptance rates and mixing. The method applies to both linear and nonlinear input-output operators and is particularly suitable for inverse problems where exact posterior sampling is too expensive. We present numerical experiments including multimodal and data-driven priors with nonlinear input-output operators. The results show that Proximal-IMH reliably outperforms existing IMH variants.

Reinforcement learning (RL) agents under partial observability often condition actions on internally accumulated information such as memory or inferred latent context. We formalise such information-conditioned interaction patterns as behavioural dependency: variation in action selection with respect to internal information under fixed observations. This induces a probe-relative notion of $ε$-behavioural equivalence and a within-policy behavioural distance that quantifies probe sensitivity. We establish three structural results. First, the set of policies exhibiting non-trivial behavioural dependency is not closed under convex aggregation. Second, behavioural distance contracts under convex combination. Third, we prove a sufficient local condition under which gradient ascent on a skewed mixture objective decreases behavioural distance when a dominant-mode gradient aligns with the direction of steepest contraction. Minimal bandit and partially observable gridworld experiments provide controlled witnesses of these mechanisms. In the examined settings, behavioural distance decreases under convex aggregation and under continued optimisation with skewed latent priors, and in these experiments it precedes degradation under latent prior shift. These results identify structural conditions under which probe-conditioned behavioural separation is not preserved under common policy transformations.

Forest monitoring is critical for climate change mitigation. However, existing global tree height maps provide only static snapshots and do not capture temporal forest dynamics, which are essential for accurate carbon accounting. We introduce ECHOSAT, a global and temporally consistent tree height map at 10 m resolution spanning multiple years. To this end, we resort to multi-sensor satellite data to train a specialized vision transformer model, which performs pixel-level temporal regression. A self-supervised growth loss regularizes the predictions to follow growth curves that are in line with natural tree development, including gradual height increases over time, but also abrupt declines due to forest loss events such as fires. Our experimental evaluation shows that our model improves state-of-the-art accuracies in the context of single-year predictions. We also provide the first global-scale height map that accurately quantifies tree growth and disturbances over time. We expect ECHOSAT to advance global efforts in carbon monitoring and disturbance assessment. The maps can be accessed at https://github.com/ai4forest/echosat.

Reinforcement Learning with Verifiable Rewards (RLVR) has become the leading paradigm for enhancing reasoning in Large Language Models (LLMs). However, standard RLVR algorithms suffer from a well-documented pathology: while they improve Pass@1 accuracy through sharpened sampling, they simultaneously narrow the model's reasoning boundary and reduce generation diversity. We identify a root cause that existing methods overlook: the uniform penalization of errors. Current approaches -- whether data-filtering methods that select prompts by difficulty, or advantage normalization schemes -- treat all incorrect rollouts within a group identically. We show that this uniformity allows overconfident errors (incorrect reasoning paths that the RL process has spuriously reinforced) to persist and monopolize probability mass, ultimately suppressing valid exploratory trajectories. To address this, we propose the Asymmetric Confidence-aware Error Penalty (ACE). ACE introduces a per-rollout confidence shift metric, c_i = log(pi_theta(y_i|x) / pi_ref(y_i|x)), to dynamically modulate negative advantages. Theoretically, we demonstrate that ACE's gradient can be decomposed into the gradient of a selective regularizer restricted to overconfident errors, plus a well-characterized residual that partially moderates the regularizer's strength. We conduct extensive experiments fine-tuning Qwen2.5-Math-7B, Qwen3-8B-Base, and Llama-3.1-8B-Instruct on the DAPO-Math-17K dataset using GRPO and DAPO within the VERL framework. Evaluated on MATH-500 and AIME 2025, ACE composes seamlessly with existing methods and consistently improves the full Pass@k spectrum across all three model families and benchmarks.

Selecting the right deep learning model for power grid forecasting is challenging, as performance heavily depends on the data available to the operator. This paper presents a comprehensive benchmark of five modern neural architectures: two state space models (PowerMamba, S-Mamba), two Transformers (iTransformer, PatchTST), and a traditional LSTM. We evaluate these models on hourly electricity demand across six diverse US power grids for forecast windows between 24 and 168 hours. To ensure a fair comparison, we adapt each model with specialized temporal processing and a modular layer that cleanly integrates weather covariates. Our results reveal that there is no single best model for all situations. When forecasting using only historical load, PatchTST and the state space models provide the highest accuracy. However, when explicit weather data is added to the inputs, the rankings reverse: iTransformer improves its accuracy three times more efficiently than PatchTST. By controlling for model size, we confirm that this advantage stems from the architecture's inherent ability to mix information across different variables. Extending our evaluation to solar generation, wind power, and wholesale prices further demonstrates that model rankings depend on the forecast task: PatchTST excels on highly rhythmic signals like solar, while state space models are better suited for the chaotic fluctuations of wind and price. Ultimately, this benchmark provides grid operators with actionable guidelines for selecting the optimal forecasting architecture based on their specific data environments.

Gaussian process (GP) surrogates are the default tool for emulating expensive computer experiments, but cubic cost, stationarity assumptions, and Gaussian predictive distributions limit their reach. We propose Generative Bayesian Computation (GBC) via Implicit Quantile Networks (IQNs) as a surrogate framework that targets all three limitations. GBC learns the full conditional quantile function from input--output pairs; at test time, a single forward pass per quantile level produces draws from the predictive distribution. Across fourteen benchmarks we compare GBC to four GP-based methods. GBC improves CRPS by 11--26\% on piecewise jump-process benchmarks, by 14\% on a ten-dimensional Friedman function, and scales linearly to 90,000 training points where dense-covariance GPs are infeasible. A boundary-augmented variant matches or outperforms Modular Jump GPs on two-dimensional jump datasets (up to 46\% CRPS improvement). In active learning, a randomized-prior IQN ensemble achieves nearly three times lower RMSE than deep GP active learning on Rocket LGBB. Overall, GBC records a favorable point estimate in 12 of 14 comparisons. GPs retain an edge on smooth surfaces where their smoothness prior provides effective regularization.

The boundary of the firm is determined by coordination cost. We argue that agentic AI induces a structural change in how coordination costs scale: in prior modular systems, integration cost grew with interaction topology (O(n^2) in the number of components); in protocol-mediated agentic systems, integration cost collapses to O(n) while verification scales with task throughput rather than interaction count. This shift selects for a specific organizational equilibrium -- the Headless Firm -- structured as an hourglass: a personalized generative interface at the top, a standardized protocol waist in the middle, and a competitive market of micro-specialized execution agents at the bottom. We formalize this claim as a coordination cost model with two falsifiable empirical predictions: (1) the marginal cost of adding an execution provider should be approximately constant in a mature hourglass ecosystem; (2) the ratio of total coordination cost to task throughput should remain stable as ecosystem size grows. We derive conditions for hourglass stability versus re-centralization and analyze implications for firm size distributions, labor markets, and software economics. The analysis predicts a domain-conditional Great Unbundling: in high knowledge-velocity domains, firm size distributions shift mass from large integrated incumbents toward micro-specialized agents and thin protocol orchestrators.

Federated Learning (FL) enables collaborative model training across multiple clients without sharing their private data. However, data heterogeneity across clients leads to client drift, which degrades the overall generalization performance of the model. This effect is further compounded by overemphasis on poorly performing clients. To address this problem, we propose FedVG, a novel gradient-based federated aggregation framework that leverages a global validation set to guide the optimization process. Such a global validation set can be established using readily available public datasets, ensuring accessibility and consistency across clients without compromising privacy. In contrast to conventional approaches that prioritize client dataset volume, FedVG assesses the generalization ability of client models by measuring the magnitude of validation gradients across layers. Specifically, we compute layerwise gradient norms to derive a client-specific score that reflects how much each client needs to adjust for improved generalization on the global validation set, thereby enabling more informed and adaptive federated aggregation. Extensive experiments on both natural and medical image benchmarking datasets, across diverse model architectures, demonstrate that FedVG consistently improves performance, particularly in highly heterogeneous settings. Moreover, FedVG is modular and can be seamlessly integrated with various state-of-the-art FL algorithms, often further improving their results. Our code is available at https://github.com/alinadevkota/FedVG.

Prompt learning has become a dominant paradigm for adapting vision-language models (VLMs) such as CLIP to downstream tasks without modifying pretrained weights. While extending prompts to both vision and text encoders across multiple transformer layers significantly boosts performance, it dramatically increases the number of trainable parameters, with state-of-the-art methods requiring millions of parameters and abandoning the parameter efficiency that makes prompt tuning attractive. In this work, we propose \textbf{MMLoP} (\textbf{M}ulti-\textbf{M}odal \textbf{Lo}w-Rank \textbf{P}rompting), a framework that achieves deep multi-modal prompting with only \textbf{11.5K trainable parameters}, comparable to early text-only methods like CoOp. MMLoP parameterizes vision and text prompts at each transformer layer through a low-rank factorization, which serves as an implicit regularizer against overfitting on few-shot training data. To further close the accuracy gap with state-of-the-art methods, we introduce three complementary components: a self-regulating consistency loss that anchors prompted representations to frozen zero-shot CLIP features at both the feature and logit levels, a uniform drift correction that removes the global embedding shift induced by prompt tuning to preserve class-discriminative structure, and a shared up-projection that couples vision and text prompts through a common low-rank factor to enforce cross-modal alignment. Extensive experiments across three benchmarks and 11 diverse datasets demonstrate that MMLoP achieves a highly favorable accuracy-efficiency tradeoff, outperforming the majority of existing methods including those with orders of magnitude more parameters, while achieving a harmonic mean of 79.70\% on base-to-novel generalization.

We study the problem of efficiently producing, in an online fashion, generative models of scalar, multiclass, and vector-valued outcomes that cannot be falsified on the basis of the observed data and a pre-specified collection of computational tests. Our contributions are twofold. First, we expand on connections between online high-dimensional multicalibration with respect to an RKHS and recent advances in expected variational inequality problems, enabling efficient algorithms for the former. We then apply this algorithmic machinery to the problem of outcome indistinguishability. Our procedure, Defensive Generation, is the first to efficiently produce online outcome indistinguishable generative models of non-Bernoulli outcomes that are unfalsifiable with respect to infinite classes of tests, including those that examine higher-order moments of the generated distributions. Furthermore, our method runs in near-linear time in the number of samples and achieves the optimal, vanishing T^{-1/2} rate for generation error.

We introduce MrBERT, a family of 150M-300M parameter encoders built on the ModernBERT architecture and pre-trained on 35 languages and code. Through targeted adaptation, this model family achieves state-of-the-art results on Catalan- and Spanish-specific tasks, while establishing robust performance across specialized biomedical and legal domains. To bridge the gap between research and production, we incorporate Matryoshka Representation Learning (MRL), enabling flexible vector sizing that significantly reduces inference and storage costs. Ultimately, the MrBERT family demonstrates that modern encoder architectures can be optimized for both localized linguistic excellence and efficient, high-stakes domain specialization. We open source the complete model family on Huggingface.

Tokenization and sub-tokenization based models like word2vec, BERT and the GPTs are the state-of-the-art in natural language processing. Typically, these approaches have limitations with respect to their input representation. They fail to fully capture orthographic similarities and morphological variations, especially in highly inflected and under-resource languages. To mitigate this problem, we propose to computes word vectors directly from character strings, integrating both semantic and syntactic information. We denote this transformer-based approach Rich Character Embeddings (RCE). Furthermore, we propose a hybrid model that combines transformer and convolutional mechanisms. Both vector representations can be used as a drop-in replacement for dictionary- and subtoken-based word embeddings in existing model architectures. It has the potential to improve performance for both large context-based language models like BERT and small models like word2vec for under-resourced and morphologically rich languages. We evaluate our approach on various tasks like the SWAG, declension prediction for inflected languages, metaphor and chiasmus detection for various languages. Our experiments show that it outperforms traditional token-based approaches on limited data using OddOneOut and TopK metrics.

Extracting clinical information from medical transcripts in low-resource languages remains a significant challenge in healthcare natural language processing (NLP). This study evaluates a two-step pipeline combining Aya-expanse-8B as a Persian-to-English translation model with five open-source small language models (SLMs) -- Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, Llama-3.2-3B-Instruct, Qwen2.5-1.5B-Instruct, and Gemma-3-1B-it -- for binary extraction of 13 clinical features from 1,221 anonymized Persian transcripts collected at a cancer palliative care call center. Using a few-shot prompting strategy without fine-tuning, models were assessed on macro-averaged F1-score, Matthews Correlation Coefficient (MCC), sensitivity, and specificity to account for class imbalance. Qwen2.5-7B-Instruct achieved the highest overall performance (median macro-F1: 0.899; MCC: 0.797), while Gemma-3-1B-it showed the weakest results. Larger models (7B--8B parameters) consistently outperformed smaller counterparts in sensitivity and MCC. A bilingual analysis of Aya-expanse-8B revealed that translating Persian transcripts to English improved sensitivity, reduced missing outputs, and boosted metrics robust to class imbalance, though at the cost of slightly lower specificity and precision. Feature-level results showed reliable extraction of physiological symptoms across most models, whereas psychological complaints, administrative requests, and complex somatic features remained challenging. These findings establish a practical, privacy-preserving blueprint for deploying open-source SLMs in multilingual clinical NLP settings with limited infrastructure and annotation resources, and highlight the importance of jointly optimizing model scale and input language strategy for sensitive healthcare applications.

Model merging under unseen test-time distribution shifts often renders naive strategies, such as mean averaging unreliable. This challenge is especially acute in medical imaging, where models are fine-tuned locally at clinics on private data, producing domain-specific models that differ by scanner, protocol, and population. When deployed at an unseen clinical site, test cases arrive in unlabeled, non-i.i.d. batches, and the model must adapt immediately without labels. In this work, we introduce an entropy-adaptive, fully online model-merging method that yields a batch-specific merged model via only forward passes, effectively leveraging target information. We further demonstrate why mean merging is prone to failure and misaligned under heterogeneous domain shifts. Next, we mitigate encoder classifier mismatch by decoupling the encoder and classification head, merging with separate merging coefficients. We extensively evaluate our method with state-of-the-art baselines using two backbones across nine medical and natural-domain generalization image classification datasets, showing consistent gains across standard evaluation and challenging scenarios. These performance gains are achieved while retaining single-model inference at test-time, thereby demonstrating the effectiveness of our method.

Multi-Head Attention (MHA) is the core computational primitive underlying modern Large Language Models (LLMs). However, MHA suffers from a fundamental linear scaling limitation: $H$ attention heads produce exactly $H$ independent attention matrices, with no communication between heads during attention computation. This becomes problematic for multi-step reasoning, where correct answers depend on aggregating evidence from multiple parts of the context and composing latent token-to-token relations over a chain of intermediate inferences. To address this, we propose Interleaved Head Attention (IHA), which enables cross-head mixing by constructing $P$ pseudo-heads per head (typically $P=H$), where each pseudo query/key/value is a learned linear combination of all $H$ original queries, keys and values respectively. Interactions between pseudo-query and pseudo-key heads induce up to $P^2$ attention patterns per head with modest parameter overhead $\mathcal{O}(H^2P)$. We provide theory showing improved efficiency in terms of number of parameters on the synthetic Polynomial task (IHA uses $Θ(\sqrt{k}n^2)$ parameters vs. $Θ(kn^2)$ for MHA) and on the synthetic order-sensitive CPM-3 task (IHA uses $\lceil\sqrt{N_{\max}}\rceil$ heads vs. $N_{\max}$ for MHA). On real-world benchmarks, IHA improves Multi-Key retrieval on RULER by 10-20% (4k-16k) and, after fine-tuning for reasoning on OpenThoughts, improves GSM8K by 5.8% and MATH-500 by 2.8% (Majority Vote) over full attention.

Given a black-box AI system and a task, at what confidence level can a practitioner trust the system's output? We answer with a reliability level -- a single number per system-task pair, derived from self-consistency sampling and conformal calibration, that serves as a black-box deployment gate with exact, finite-sample, distribution-free guarantees. Self-consistency sampling reduces uncertainty exponentially; conformal calibration guarantees correctness within 1/(n+1) of the target level, regardless of the system's errors -- made transparently visible through larger answer sets for harder questions. Weaker models earn lower reliability levels (not accuracy -- see Definition 2.4): GPT-4.1 earns 94.6% on GSM8K and 96.8% on TruthfulQA, while GPT-4.1-nano earns 89.8% on GSM8K and 66.5% on MMLU. We validate across five benchmarks, five models from three families, and both synthetic and real data. Conditional coverage on solvable items exceeds 0.93 across all configurations; sequential stopping reduces API costs by around 50%.

Purpose: Curating large-scale datasets of operating room (OR) workflow, encompassing rare, safety-critical, or atypical events, remains operationally and ethically challenging. This data bottleneck complicates the development of ambient intelligence for detecting, understanding, and mitigating rare or safety-critical events in the OR. Methods: This work presents an OR video diffusion framework that enables controlled synthesis of rare and safety-critical events. The framework integrates a geometric abstraction module, a conditioning module, and a fine-tuned diffusion model to first transform OR scenes into abstract geometric representations, then condition the synthesis process, and finally generate realistic OR event videos. Using this framework, we also curate a synthetic dataset to train and validate AI models for detecting near-misses of sterile-field violations. Results: In synthesizing routine OR events, our method outperforms off-the-shelf video diffusion baselines, achieving lower FVD/LPIPS and higher SSIM/PSNR in both in- and out-of-domain datasets. Through qualitative results, we illustrate its ability for controlled video synthesis of counterfactual events. An AI model trained and validated on the generated synthetic data achieved a RECALL of 70.13% in detecting near safety-critical events. Finally, we conduct an ablation study to quantify performance gains from key design choices. Conclusion: Our solution enables controlled synthesis of routine and rare OR events from abstract geometric representations. Beyond demonstrating its capability to generate rare and safety-critical scenarios, we show its potential to support the development of ambient intelligence models.

Ptychographic imaging at synchrotron and XFEL sources requires dense overlapping scans, limiting throughput and increasing dose. Extending coherent diffractive imaging to overlap-free operation on extended samples remains an open problem. Here, we extend PtychoPINN (O. Hoidn \emph{et al.}, \emph{Scientific Reports} \textbf{13}, 22789, 2023) to deliver \emph{overlap-free, single-shot} reconstructions in a Fresnel coherent diffraction imaging (CDI) geometry while also accelerating conventional multi-shot ptychography. The framework couples a differentiable forward model of coherent scattering with a Poisson photon-counting likelihood; real-space overlap enters as a tunable parameter via coordinate-based grouping rather than a hard requirement. On synthetic benchmarks, reconstructions remain accurate at low counts ($\sim\!10^4$ photons/frame), and overlap-free single-shot reconstruction with an experimental probe reaches amplitude structural similarity (SSIM) 0.904, compared with 0.968 for overlap-constrained reconstruction. Against a data-saturated supervised model with the same backbone (16,384 training images), PtychoPINN achieves higher SSIM with only 1,024 images and generalizes to unseen illumination profiles. Per-graphics processing unit (GPU) throughput is approximately $40\times$ that of least-squares maximum-likelihood (LSQ-ML) reconstruction at matched $128\times128$ resolution. These results, validated on experimental data from the Advanced Photon Source and the Linac Coherent Light Source, unify single-exposure Fresnel CDI and overlapped ptychography within one framework, supporting dose-efficient, high-throughput imaging at modern light sources.

This paper establishes and proves representation theorems for cumulative propositional dependence logic and for cumulative propositional logic with team semantics. Cumulative logics are famously given by System C. For propositional dependence logic, we show that System C entailments are exactly captured by cumulative models from Kraus, Lehmann and Magidor. On the other hand, we show that entailment in cumulative propositional logics with team semantics is exactly captured by cumulative and asymmetric models. For the latter, we also obtain equivalence with cumulative logics based on propositional logic with classical semantics. The proofs will be useful for proving representation theorems for other cumulative logics without negation and material implication.

Bayesian inference for inverse problems involves computing expectations under posterior distributions -- e.g., posterior means, variances, or predictive quantities -- typically via Monte Carlo (MC) estimation. When the quantity of interest varies significantly under the posterior, accurate estimates demand many samples -- a cost often prohibitive for partial differential equation-constrained problems. To address this challenge, we introduce conditional neural control variates, a modular method that learns amortized control variates from joint model-data samples to reduce the variance of MC estimators. To scale to high-dimensional problems, we leverage Stein's identity to design an architecture based on an ensemble of hierarchical coupling layers with tractable Jacobian trace computation. Training requires: (i) samples from the joint distribution of unknown parameters and observed data; and (ii) the posterior score function, which can be computed from physics-based likelihood evaluations, neural operator surrogates, or learned generative models such as conditional normalizing flows. Once trained, the control variates generalize across observations without retraining. We validate our approach on stylized and partial differential equation-constrained Darcy flow inverse problems, demonstrating substantial variance reduction, even when the analytical score is replaced by a learned surrogate.

The rapid accumulation of Earth science data has created a significant scalability challenge; while repositories like PANGAEA host vast collections of datasets, citation metrics indicate that a substantial portion remains underutilized, limiting data reusability. Here we present PANGAEA-GPT, a hierarchical multi-agent framework designed for autonomous data discovery and analysis. Unlike standard Large Language Model (LLM) wrappers, our architecture implements a centralized Supervisor-Worker topology with strict data-type-aware routing, sandboxed deterministic code execution, and self-correction via execution feedback, enabling agents to diagnose and resolve runtime errors. Through use-case scenarios spanning physical oceanography and ecology, we demonstrate the system's capacity to execute complex, multi-step workflows with minimal human intervention. This framework provides a methodology for querying and analyzing heterogeneous repository data through coordinated agent workflows.

Recent advances in alignment techniques such as Supervised Fine-Tuning (SFT), Reinforcement Learning from Human Feedback (RLHF), and Direct Preference Optimization (DPO) have improved the safety of large language models (LLMs). However, these LLMs remain vulnerable to jailbreak attacks that disguise harmful intent through indirect or deceptive phrasing. Using causal intervention, we empirically demonstrate that this vulnerability stems from shallow alignment mechanisms that lack deep reasoning, often rejecting harmful prompts without truly understanding why they are harmful. To mitigate this vulnerability, we propose enhancing alignment through reasoning-aware post-training. We construct and release a novel Chain-of-Thought (CoT) fine-tuning dataset that includes both utility-oriented and safety-critical prompts with step-by-step rationales. Fine-tuning on this dataset encourages models to produce principled refusals grounded in reasoning, outperforming standard SFT baselines. Furthermore, inspired by failure patterns in CoT fine-tuning, we introduce Alignment-Weighted DPO, which targets the most problematic parts of an output by assigning different preference weights to the reasoning and final-answer segments. This produces finer-grained, targeted updates than vanilla DPO and improves robustness to diverse jailbreak strategies. Extensive experiments across multiple safety and utility benchmarks show that our method consistently improves alignment robustness while maintaining overall model utility.

Representation learning has been essential for graph machine learning tasks such as link prediction, community detection, and network visualization. Despite recent advances in achieving high performance on these downstream tasks, little progress has been made toward self-explainable models. Understanding the patterns behind predictions is equally important, motivating recent interest in explainable machine learning. In this paper, we present GraphHull, an explainable generative model that represents networks using two levels of convex hulls. At the global level, the vertices of a convex hull are treated as archetypes, each corresponding to a pure community in the network. At the local level, each community is refined by a prototypical hull whose vertices act as representative profiles, capturing community-specific variation. This two-level construction yields clear multi-scale explanations: a node's position relative to global archetypes and its local prototypes directly accounts for its edges. The geometry is well-behaved by design, while local hulls are kept disjoint by construction. To further encourage diversity and stability, we place principled priors, including determinantal point processes, and fit the model under MAP estimation with scalable subsampling. Experiments on real networks demonstrate the ability of GraphHull to recover multi-level community structure and to achieve competitive or superior performance in link prediction and community detection, while naturally providing interpretable predictions.

Geometry-free view synthesis transformers have recently achieved state-of-the-art performance in Novel View Synthesis (NVS), outperforming traditional approaches that rely on explicit geometry modeling. Yet the factors governing their scaling with compute remain unclear. We present a systematic study of scaling laws for view synthesis transformers and derive design principles for training compute-optimal NVS models. Contrary to prior findings, we show that encoder-decoder architectures can be compute-optimal; we trace earlier negative results to suboptimal architectural choices and comparisons across unequal training compute budgets. Across several compute levels, we demonstrate that our encoder-decoder architecture, which we call the Scalable View Synthesis Model (SVSM), scales as effectively as decoder-only models, achieves a superior performance-compute Pareto frontier, and surpasses the previous state-of-the-art on real-world NVS benchmarks with substantially reduced training compute.

Representing the past in a compressed, efficient, and informative manner is a central problem for systems trained on sequential data. The HiPPO framework, originally proposed by Gu & Dao et al., provides a principled approach to sequential compression by projecting signals onto orthogonal polynomial (OP) bases via structured linear ordinary differential equations. Subsequent works have embedded these dynamics in state space models (SSMs), where HiPPO structure serves as an initialization. Nonlinear successors of these SSM methods such as Mamba are state-of-the-art for many tasks with long-range dependencies, but the mechanisms by which they represent and prioritize history remain largely implicit. In this work, we revisit the HiPPO framework with the goal of making these mechanisms explicit. We show how polynomial representations of history can be extended to support capabilities of modern SSMs such as adaptive allocation of memory and associative memory while retaining direct interpretability in the OP basis. We introduce a unified framework comprising five such extensions, which we collectively refer to as a "HiPPO zoo." Each extension exposes a specific modeling capability through an explicit, interpretable modification of the HiPPO framework. The resulting models adapt their memory online and train in streaming settings with efficient updates. We illustrate the behaviors and modeling advantages of these extensions through a range of synthetic sequence modeling tasks, demonstrating that capabilities typically associated with modern SSMs can be realized through explicit, interpretable polynomial memory structures.

We study the problem of opportunistic approachability: a generalization of Blackwell approachability where the learner would like to obtain stronger guarantees (i.e., approach a smaller set) when their adversary limits themselves to a subset of their possible action space. Bernstein et al. (2014) introduced this problem in 2014 and presented an algorithm that guarantees sublinear approachability rates for opportunistic approachability. However, this algorithm requires the ability to produce calibrated online predictions of the adversary's actions, a problem whose standard implementations require time exponential in the ambient dimension and result in approachability rates that scale as $T^{-O(1/d)}$. In this paper, we present an efficient algorithm for opportunistic approachability that achieves a rate of $O(T^{-1/4})$ (and an inefficient one that achieves a rate of $O(T^{-1/3})$), bypassing the need for an online calibration subroutine. Moreover, in the case where the dimension of the adversary's action set is at most two, we show it is possible to obtain the optimal rate of $O(T^{-1/2})$.

Individuals with similar qualifications and skills may vary in their demeanor, or outward manner: some tend toward self-promotion while others are modest to the point of omitting crucial information. Comparing the self-descriptions of equally qualified job-seekers with different self-presentation styles is therefore problematic. We build an interactive AI for skill elicitation that provides accurate determination of skills while simultaneously allowing individuals to speak in their own voice. Such a system can be deployed, for example, when a new user joins a professional networking platform, or when matching employees to needs during a company reorganization. To obtain sufficient training data, we train an LLM to act as synthetic humans. Elicitation mitigates endogenous bias arising from individuals' own self-reports. To address systematic model bias we enforce a mathematically rigorous notion of equitability ensuring that the covariance between self-presentation manner and skill evaluation error is small.

Analog in-memory computing (AIMC) performs computation directly within resistive crossbar arrays, offering an energy-efficient platform to scale large vision and language models. However, non-ideal analog device properties make the training on AIMC devices challenging. In particular, its update asymmetry can induce a systematic drift of weight updates towards a device-specific symmetric point (SP), which typically does not align with the optimum of the training objective. To mitigate this bias, most existing works assume the SP is known and pre-calibrate it to zero before training by setting the reference point as the SP. Nevertheless, calibrating AIMC devices requires costly pulse updates, and residual calibration error can directly degrade training accuracy. In this work, we present the first theoretical characterization of the pulse complexity of SP calibration and the resulting estimation error. We further propose a dynamic SP estimation method that tracks the SP during model training, and establishes its convergence guarantees. In addition, we develop an enhanced variant based on chopping and filtering techniques from digital signal processing. Numerical experiments demonstrate both the efficiency and effectiveness of the proposed method.

Large language models (LLMs) are becoming the foundation for autonomous agents that can use tools to solve complex tasks. Reinforcement learning (RL) has emerged as a common approach for injecting such agentic capabilities, but typically under tightly controlled training setups. It often depends on carefully constructed task-solution pairs and substantial human supervision, which creates a fundamental obstacle to open-ended self-evolution toward superintelligent systems. In this paper, we propose Tool-R0 framework for training general purpose tool-calling agents from scratch with self-play RL, under a zero-data assumption. Initialized from the same base LLM, Tool-R0 co-evolves a Generator and a Solver with complementary rewards: one proposes targeted challenging tasks at the other's competence frontier and the other learns to solve them with real-world tool calls. This creates a self-evolving cycle that requires no pre-existing tasks or datasets. Evaluation on different tool-use benchmarks show that Tool-R0 yields 92.5 relative improvement over the base model and surpasses fully supervised tool-calling baselines under the same setting. Our work further provides empirical insights into self-play LLM agents by analyzing co-evolution, curriculum dynamics, and scaling behavior.

Accurate and uncertainty-aware trajectory prediction remains a core challenge for autonomous driving, driven by complex multi-agent interactions, diverse scene contexts and the inherently stochastic nature of future motion. Diffusion-based generative models have recently shown strong potential for capturing multimodal futures, yet existing approaches such as cVMD suffer from slow sampling, limited exploitation of generative diversity and brittle scenario encodings. This work introduces cVMDx, an enhanced diffusion-based trajectory prediction framework that improves efficiency, robustness and multimodal predictive capability. Through DDIM sampling, cVMDx achieves up to a 100x reduction in inference time, enabling practical multi-sample generation for uncertainty estimation. A fitted Gaussian Mixture Model further provides tractable multimodal predictions from the generated trajectories. In addition, a CVQ-VAE variant is evaluated for scenario encoding. Experiments on the publicly available highD dataset show that cVMDx achieves higher accuracy and significantly improved efficiency over cVMD, enabling fully stochastic, multimodal trajectory prediction.

Large Language Models (LLMs) are converging towards a singular Artificial Hivemind, where shared Nature (pre-training priors) result in a profound collapse of distributional diversity, limiting the distinct perspectives necessary for creative exploration and scientific discovery. To address this, we propose to equip models with inference-time Nurture (individualized epistemic trajectories) using Epistemic Evolution paradigm, progressing through explore, internalize, and express. We instantiate this via PRISM (Pluralistic Reasoning via In-context Structure Modeling), a model-agnostic system that augments LLM with dynamic On-the-fly Epistemic Graphs. On three creativity benchmarks, PRISM achieves state-of-the-art novelty and significantly expands distributional diversity. Moreover, we evaluate the real-world utility via a challenging rare-disease diagnosis benchmark. Results demonstrate that PRISM successfully uncovers correct long-tail diagnoses that standard LLM miss, confirming that its divergence stems from meaningful exploration rather than incoherent noise. Overall, this work establishes a new paradigm for Pluralistic AI, moving beyond monolithic consensus toward a diverse ecosystem of unique cognitive individuals capable of collective, multi-perspective discovery.

This work considers a number of optimization problems and reductive relations between them. The two main problems we are interested in are the \emph{Optimal Decision Tree} and \emph{Set Cover}. We study these two fundamental tasks under precedence constraints, that is, if a test (or set) $X$ is a predecessor of $Y$, then in any feasible decision tree $X$ needs to be an ancestor of $Y$ (or respectively, if $Y$ is added to set cover, then so must be $X$). For the Optimal Decision Tree we consider two optimization criteria: worst case identification time (height of the tree) or the average identification time. Similarly, for the Set Cover we study two cost measures: the size of the cover or the average cover time. Our approach is to develop a number of algorithmic reductions, where an approximation algorithm for one problem provides an approximation for another via a black-box usage of a procedure for the former. En route we introduce other optimization problems either to complete the `reduction landscape' or because they hold the essence of combinatorial structure of our problems. The latter is brought by a problem of finding a maximum density precedence closed subfamily, where the density is defined as the ratio of the number of items the family covers to its size. By doing so we provide $\cO^*(\sqrt{m})$-approximation algorithms for all of the aforementioned problems. The picture is complemented by a number of hardness reductions that provide $o(m^{1/12-ε})$-inapproximability results for the decision tree and covering problems. Besides giving a complete set of results for general precedence constraints, we also provide polylogarithmic approximation guarantees for two most typically studied and applicable precedence types, outforests and inforests. By providing corresponding hardness results, we show these results to be tight.

Symbolic distillation replaces neural networks, or components thereof, with interpretable, closed-form mathematical expressions. This approach has shown promise in discovering physical laws and mathematical relationships directly from trained deep learning models, yet adoption remains limited due to the engineering barrier of integrating symbolic regression into deep learning workflows. We introduce SymTorch, a library that automates this distillation by wrapping neural network components, collecting their input-output behavior, and approximating them with human-readable equations via PySR. SymTorch handles the engineering challenges that have hindered adoption: GPU-CPU data transfer, input-output caching, model serialization, and seamless switching between neural and symbolic forward passes. We demonstrate SymTorch across diverse architectures including GNNs, PINNs and transformer models. Finally, we present a proof-of-concept for accelerating LLM inference by replacing MLP layers with symbolic surrogates, achieving an 8.3\% throughput improvement with moderate performance degradation.

The standard way to evaluate language models on subjective tasks is through pairwise comparisons: an annotator chooses the "better" of two responses to a prompt. Leaderboards aggregate these comparisons into a single Bradley-Terry (BT) ranking, forcing heterogeneous preferences into a total order and violating basic social-choice desiderata. In contrast, social choice theory provides an alternative approach called maximal lotteries, which aggregates pairwise preferences without imposing any assumptions on their structure. However, we show that maximal lotteries are highly sensitive to preference heterogeneity and can favor models that severely underperform on specific tasks or user subpopulations. We introduce robust lotteries that optimize worst-case performance under plausible shifts in the preference data. On large-scale preference datasets, robust lotteries provide more reliable win rate guarantees across the annotator distribution and recover a stable set of top-performing models. By moving from rankings to pluralistic sets of winners, robust lotteries offer a principled step toward an ecosystem of complementary AI systems that serve the full spectrum of human preferences.

Large language models (LLMs) are increasingly deployed as multi-step decision-making agents, where effective reward design is essential for guiding learning. Although recent work explores various forms of reward shaping and step-level credit assignment, a key signal remains largely overlooked: the intrinsic uncertainty of LLMs. Uncertainty reflects model confidence, reveals where exploration is needed, and offers valuable learning cues even in failed trajectories. We introduce SELAUR: Self Evolving LLM Agent via Uncertainty-aware Rewards, a reinforcement learning framework that incorporates uncertainty directly into the reward design. SELAUR integrates entropy-, least-confidence-, and margin-based metrics into a combined token-level uncertainty estimate, providing dense confidence-aligned supervision, and employs a failure-aware reward reshaping mechanism that injects these uncertainty signals into step- and trajectory-level rewards to improve exploration efficiency and learning stability. Experiments on two benchmarks, ALFWorld and WebShop, show that our method consistently improves success rates over strong baselines. Ablation studies further demonstrate how uncertainty signals enhance exploration and robustness.

Neural networks are trained by optimizing multi-dimensional sets of fitting parameters on non-convex loss landscapes. Low-loss regions of the landscapes correspond to the parameter sets that perform well on the training data. A key issue in machine learning is the performance of trained neural networks on previously unseen test data. Here, we investigate neural network training by stochastic gradient descent (SGD) - a non-convex global optimization algorithm which relies only on the gradient of the objective function. We contrast SGD solutions with those obtained via a non-stochastic quasi-Newton method, which utilizes curvature information to determine step direction and Golden Section Search to choose step size. We use several computational tools to investigate neural network parameters obtained by these two optimization methods, including kernel Principal Component Analysis and a novel, general-purpose algorithm for finding low-height paths between pairs of points on loss or energy landscapes, FourierPathFinder. We find that the choice of the optimizer profoundly affects the nature of the resulting solutions. SGD solutions tend to be separated by lower barriers than quasi-Newton solutions, even if both sets of solutions are regularized by early stopping to ensure adequate performance on test data. When allowed to fit extensively on the training data, quasi-Newton solutions occupy deeper minima on the loss landscapes that are not reached by SGD. These solutions are less generalizable to the test data however. Overall, SGD explores smooth basins of attraction, while quasi-Newton optimization is capable of finding deeper, more isolated minima that are more spread out in the parameter space. Our findings help understand both the topography of the loss landscapes and the fundamental role of landscape exploration strategies in creating robust, transferrable neural network models.

Hamiltonian Monte Carlo (HMC) is a state of the art method for sampling from distributions with differentiable densities, but can converge slowly when applied to challenging multimodal problems. Running HMC with a time varying Hamiltonian, in order to interpolate from an initial tractable distribution to the target of interest, can address this problem. In conjunction with a weighting scheme to eliminate bias, this can be viewed as a special case of Sequential Monte Carlo (SMC) sampling \cite{doucet2001introduction}. However, this approach can be inefficient, since it requires slow change between the initial and final distribution. Inspired by \cite{sels2017minimizing}, where a learned \emph{counterdiabatic} term added to the Hamiltonian allows for efficient quantum state preparation, we propose \emph{Counterdiabatic Hamiltonian Monte Carlo} (CHMC), which can be viewed as an SMC sampler with a more efficient kernel. We establish its relationship to recent proposals for accelerating gradient-based sampling with learned drift terms, and demonstrate on simple benchmark problems.

Scaling multimodal alignment between video and audio is challenging, particularly due to limited data and the mismatch between text descriptions and frame-level video information. In this work, we tackle the scaling challenge in multimodal-to-audio generation, examining whether models trained on short instances can generalize to longer ones during testing. To tackle this challenge, we present multimodal hierarchical networks so-called MMHNet, an enhanced extension of state-of-the-art video-to-audio models. Our approach integrates a hierarchical method and non-causal Mamba to support long-form audio generation. Our proposed method significantly improves long audio generation up to more than 5 minutes. We also prove that training short and testing long is possible in the video-to-audio generation tasks without training on the longer durations. We show in our experiments that our proposed method could achieve remarkable results on long-video to audio benchmarks, beating prior works in video-to-audio tasks. Moreover, we showcase our model capability in generating more than 5 minutes, while prior video-to-audio methods fall short in generating with long durations.

Bubeck and Sellke (2021) pose as an open problem the connection between the law of robustness and robust generalization. The law of robustness states that overparameterization is necessary for models to interpolate robustly; in particular, robust interpolation requires the learned function to be Lipschitz. Robust generalization asks whether small robust training loss implies small robust test loss. We resolve this problem by explicitly connecting the two for arbitrary data distributions. Specifically, we introduce a nontrivial notion of robust generalization error and convert it into a lower bound on the expected Rademacher complexity of the induced robust loss class. Our bounds recover the $Ω(n^{1/d})$ regime of Wu et al. (2023) and show that, up to constants, robust generalization does not change the order of the Lipschitz constant required for smooth interpolation. We conduct experiments to probe the predicted scaling with dataset size and model capacity, testing whether empirical behavior aligns more closely with the predictions of Bubeck and Sellke (2021) or Wu et al. (2023). For MNIST, we find that the lower-bound Lipschitz constant scales on the order predicted by Wu et al. (2023). Informally, to obtain low robust generalization error, the Lipschitz constant must lie in a range that we bound, and the allowable perturbation radius is linked to the Lipschitz scale.

For millennia, human cognition was the primary engine of progress on Earth. As AI decouples cognition from biology, the marginal cost of measurable execution falls to zero, absorbing any labor capturable by metrics--including creative, analytical, and innovative work. The binding constraint on growth is no longer intelligence but human verification bandwidth: the capacity to validate, audit, and underwrite responsibility when execution is abundant. We model the AGI transition as the collision of two racing cost curves: an exponentially decaying Cost to Automate and a biologically bottlenecked Cost to Verify. This structural asymmetry widens a Measurability Gap between what agents can execute and what humans can afford to verify. It also drives a shift from skill-biased to measurability-biased technical change. Rents migrate to verification-grade ground truth, cryptographic provenance, and liability underwriting--the ability to insure outcomes rather than merely generate them. The current human-in-the-loop equilibrium is unstable: eroded from below as apprenticeship collapses (Missing Junior Loop) and from within as experts codify their obsolescence (Codifier's Curse). Unverified deployment becomes privately rational--a Trojan Horse externality. Unmanaged, these forces pull toward a Hollow Economy. Yet by scaling verification alongside agentic capabilities, the forces that threaten collapse become the catalyst for unbounded discovery and experimentation--an Augmented Economy. We derive a practical playbook for individuals, companies, investors, and policymakers. Today's defining challenge is not the race to deploy the most autonomous systems; it is the race to secure the foundations of their oversight. Only by scaling our bandwidth for verification alongside our capacity for execution can we ensure that the intelligence we have summoned preserves the humanity that initiated it.

Large Language Models (LLMs) consistently benefit from scaled Chain-of-Thought (CoT) reasoning, but also suffer from heavy computational overhead. To address this issue, efficient reasoning aims to incentivize short yet accurate thinking trajectories, typically through reward shaping with Reinforcement Learning (RL). In this paper, we systematically investigate the mechanics of efficient reasoning for LLMs. For comprehensive evaluation, we advocate for more fine-grained metrics, including length distribution conditioned on correctness and performance across a wide spectrum of token budgets ranging from 2k to 32k. First, we reveal that the training process follows a two-stage paradigm: length adaptation and reasoning refinement. After that, we conduct extensive experiments (about 0.2 million GPU hours) in a unified protocol, deconstructing training prompts and rollouts, reward shaping, and optimization strategies. In particular, a key finding is to train on relatively easier prompts, ensuring the density of positive reward signals and thus avoiding the length collapse. Meanwhile, the learned length bias can be generalized across domains. We distill all findings into valuable insights and practical guidelines, and further validate them across the Qwen3 series, ranging from 0.6B to 30B, demonstrating the robustness and generalization.

We present Group Orthogonalized Policy Optimization (GOPO), a new alignment algorithm for large language models derived from the geometry of Hilbert function spaces. Instead of optimizing on the probability simplex and inheriting the exponential curvature of Kullback-Leibler divergence, GOPO lifts alignment into the Hilbert space L2(pi_k) of square-integrable functions with respect to the reference policy. Within this space, the simplex constraint reduces to a linear orthogonality condition <v, 1> = 0, defining a codimension-one subspace H0. Minimizing distance to an unconstrained target u_star yields the work-dissipation functional J(v) = <g, v> - (mu / 2) ||v||^2, whose maximizer follows directly from the Hilbert projection theorem. Enforcing the boundary v >= -1 produces a bounded Hilbert projection that induces exact sparsity, assigning zero probability to catastrophically poor actions through a closed-form threshold. To connect this functional theory with practice, GOPO projects from infinite-dimensional L2(pi_k) to a finite empirical subspace induced by group sampling. Because group-normalized advantages sum to zero, the Lagrange multiplier enforcing probability conservation vanishes exactly, reducing the constrained projection to an unconstrained empirical loss. The resulting objective has constant Hessian curvature mu I, non-saturating linear gradients, and an intrinsic dead-zone mechanism without heuristic clipping. Experiments on mathematical reasoning benchmarks show that GOPO achieves competitive generalization while maintaining stable gradient dynamics and entropy preservation in regimes where clipping-based methods plateau.

Soft set theory provides a direct framework for parameterized decision modeling by assigning to each attribute (parameter) a subset of a given universe, thereby representing uncertainty in a structured way [1, 2]. Over the past decades, the theory has expanded into numerous variants-including hypersoft sets, superhypersoft sets, TreeSoft sets, bipolar soft sets, and dynamic soft sets-and has been connected to diverse areas such as topology and matroid theory. In this book, we present a survey-style overview of soft sets and their major extensions, highlighting core definitions, representative constructions, and key directions of current development.

Cybersecurity threats are becoming increasingly sophisticated, making traditional defense mechanisms and manual red teaming approaches insufficient for modern organizations. While red teaming has long been recognized as an effective method to identify vulnerabilities by simulating real-world attacks, its manual execution is resource-intensive, time-consuming, and lacks scalability for frequent assessments. These limitations have driven the evolution toward auto-mated red teaming, which leverages artificial intelligence and automation to deliver efficient and adaptive security evaluations. This systematic review consolidates existing research on automated red teaming, examining its methodologies, tools, benefits, and limitations. The paper also highlights current trends, challenges, and research gaps, offering insights into future directions for improving automated red teaming as a critical component of proactive cybersecurity strategies. By synthesizing findings from diverse studies, this review aims to provide a comprehensive understanding of how automation enhances red teaming and strengthens organizational resilience against evolving cyber threats.

We introduce \ToolMATH, a math-grounded benchmark that evaluates tool-augmented language models in realistic multi-tool environments where the output depends on calling schema-specified tools and sustaining multi-step execution. It turns math problems into a controlled, correctness-checkable benchmark with tool sets, enabling systematic evaluation of model reliability under (1) large, overlapping tool catalogs and (2) the absence of the intended capability. \ToolMATH provides actionable diagnostic evidence of failure modes in tool-augmented agents, helping identify the control mechanisms required for robustness. \ToolMATH roughly contains 8k questions and 12k tools; we provide an additional hard-set \ToolMATHHard with questions and tools. Our evaluation reveals that the key failure factor is due to the inability to reason, leading to the accumulation of intermediate results' errors and constrain later decisions. Tool-list redundancy do not simply add noise, but amplify small early deviations into irreversible execution drift. The benchmark highlights that when the intended capability is missing, distractor tools can sometimes serve as partial substitutes in solution paths, yet they can also mislead models into ungrounded tool trajectories. Finally, comparisons between tool-use protocols emphasize that improvements come less from local action selection and more from long-range plan coherence and disciplined use of observations.

Reliable prediction of hydraulic performance is challenging for Piano Key Weir (PKW) design because discharge capacity depends on three-dimensional geometry and operating conditions. Surrogate models can accelerate hydraulic-structure design, but progress is limited by scarce large, well-documented datasets that jointly capture geometric variation, operating conditions, and functional performance. This study presents WeirNet, a large 3D CFD benchmark dataset for geometric surrogate modeling of PKWs. WeirNet contains 3,794 parametric, feasibility-constrained rectangular and trapezoidal PKW geometries, each scheduled at 19 discharge conditions using a consistent free-surface OpenFOAM workflow, resulting in 71,387 completed simulations that form the benchmark and with complete discharge coefficient labels. The dataset is released as multiple modalities compact parametric descriptors, watertight surface meshes and high-resolution point clouds together with standardized tasks and in-distribution and out-of-distribution splits. Representative surrogate families are benchmarked for discharge coefficient prediction. Tree-based regressors on parametric descriptors achieve the best overall accuracy, while point- and mesh-based models remain competitive and offer parameterization-agnostic inference. All surrogates evaluate in milliseconds per sample, providing orders-of-magnitude speedups over CFD runtimes. Out-of-distribution results identify geometry shift as the dominant failure mode compared to unseen discharge values, and data-efficiency experiments show diminishing returns beyond roughly 60% of the training data. By publicly releasing the dataset together with simulation setups and evaluation pipelines, WeirNet establishes a reproducible framework for data-driven hydraulic modeling and enables faster exploration of PKW designs during the early stages of hydraulic planning.

Vision-language-action models must enable agents to execute long-horizon tasks under partial observability. However, most existing approaches remain observation-driven, relying on short context windows or repeated queries to vision-language models (VLMs). This leads to loss of task progress, action repetition under perceptual aliasing, and high inference latency. While semantic grounding is important, long-horizon manipulation fundamentally requires persistent, action-conditioned state representations. Current VLAs lack such representations and exhibit limited temporal and physical reasoning, making them ill-suited for multi-stage control. This paper introduces RB-VLA, a belief-centric architecture trained with self-supervised world-model objectives that maintains a compact latent state encoding task-relevant history, dynamics, and object interactions. Queried once per task, the VLM provides high-level intent, while the belief tracks task progress and enables phase-aware, causally grounded control under partial observability without storing raw observations or scaling memory with time. The belief and intent jointly condition a diffusion policy for robust closed-loop execution. RB-VLA outperforms prior VLAs on long-horizon benchmarks, achieving 52.5 percent and 37.5 percent higher success rates on multi-stage pick-and-place and stacking tasks, respectively, compared to pi_0. It also reduces inference latency by up to five times relative to baselines and eliminates memory growth across timesteps observed in existing VLAs. Ablations show the belief module is the primary driver of performance, increasing success rates from 32.5 percent without belief to 77.5 percent with belief.

Specifications are vital for ensuring program correctness, yet writing them manually remains challenging and time-intensive. Recent large language model (LLM)-based methods have shown successes in generating specifications such as postconditions, but existing single-pass prompting often yields inaccurate results. In this paper, we present SpecMind, a novel framework for postcondition generation that treats LLMs as interactive and exploratory reasoners rather than one-shot generators. SpecMind employs feedback-driven multi-turn prompting approaches, enabling the model to iteratively refine candidate postconditions by incorporating implicit and explicit correctness feedback, while autonomously deciding when to stop. This process fosters deeper code comprehension and improves alignment with true program behavior via exploratory attempts. Our empirical evaluation shows that SpecMind significantly outperforms state-of-the-art approaches in both accuracy and completeness of generated postconditions.

We study the problem of fairly allocating indivisible goods when limited sharing is allowed, that is, each good may be allocated to up to $k$ agents, while incurring a cost for sharing. While classic maximin share (MMS) allocations may not exist in many instances, we demonstrate that allowing controlled sharing can restore fairness guarantees that are otherwise unattainable in certain scenarios. (1) Our first contribution shows that exact maximin share (MMS) allocations are guaranteed to exist whenever goods are allowed to be cost-sensitively shared among at least half of the agents and the number of agents is even; for odd numbers of agents, we obtain a slightly weaker MMS guarantee. (2) We further design a Shared Bag-Filling Algorithm that guarantees a $(1 - C)(k - 1)$-approximate MMS allocation, where $C$ is the maximum cost of sharing a good. Notably, when $(1 - C)(k - 1) \geq 1$, our algorithm recovers an exact MMS allocation. (3) We additionally introduce the Sharing Maximin Share (SMMS) fairness notion, a natural extension of MMS to the $k$-sharing setting. (4) We show that SMMS allocations always exist under identical utilities and for instances with two agents. (5) We construct a counterexample to show the impossibility of the universal existence of an SMMS allocation. (6) Finally, we establish a connection between SMMS and constrained MMS (CMMS), yielding approximation guarantees for SMMS via existing CMMS results. These contributions provide deep theoretical insights for the problem of fair resource allocation when a limited sharing of resources are allowed in multi-agent environments.

With increasing integration of Large Language Models (LLMs) into areas of high-stakes human decision-making, it is important to understand the risks they introduce as advisors. To be useful advisors, LLMs must sift through large amounts of content, written with both benevolent and malicious intent, and then use this information to convince a user to take a specific action. This involves two social capacities: vigilance (the ability to determine which information to use, and which to discard) and persuasion (synthesizing the available evidence to make a convincing argument). While existing work has investigated these capacities in isolation, there has been little prior investigation of how these capacities may be linked. Here, we use a simple multi-turn puzzle-solving game, Sokoban, to study LLMs' abilities to persuade and be rationally vigilant towards other LLM agents. We find that puzzle-solving performance, persuasive capability, and vigilance are dissociable capacities in LLMs. Performing well on the game does not automatically mean a model can detect when it is being misled, even if the possibility of deception is explicitly mentioned. % as part of the prompt. However, LLMs do consistently modulate their token use, using fewer tokens to reason when advice is benevolent and more when it is malicious, even if they are still persuaded to take actions leading them to failure. To our knowledge, our work presents the first investigation of the relationship between persuasion, vigilance, and task performance in LLMs, and suggests that monitoring all three independently will be critical for future work in AI safety.

Multimodal Role-Playing Agents (MRPAs) are attracting increasing attention due to their ability to deliver more immersive multimodal emotional interactions. However, existing studies still rely on pure textual benchmarks to evaluate the text responses of MRPAs, while delegating the assessment of their multimodal expressions solely to modality-synthesis metrics. This evaluation paradigm, on the one hand, entangles semantic assessment with modality generation, leading to ambiguous error attribution, and on the other hand remains constrained by the heavy reliance on human judgment. To this end, we propose MERRY, a semantically decoupled evaluation framework for assessing Multimodal Emotional and Role consistencies of Role-playing agents. This framework introduce five refined metrics for EC and three for RC. Notably, we transform the traditional subjective scoring approach into a novel bidirectional-evidence-finding task, significantly improving the human agreement of LLM-as-Judge evaluations. Based on MERRY, we conduct extensive evaluations. Our empirical results primarily reveal that: (1) Training on synthetic datasets tends to reduce emotional consistency, whereas training on real-world datasets improves it; (2) Existing models suffer from emotional templatization and simplification, exhibiting positive-bias and performance bottleneck in fine-grained negative emotions; (3) Simple prompting method strengthens the weak models but constrains the strong ones, while simple fine-tuning method suffers from poor role generalization. Codes and dataset are available.

We approach the Max-3-Cut problem through the lens of maximizing complex-valued quadratic forms and demonstrate that low-rank structure in the objective matrix can be exploited, leading to alternative algorithms to classical semidefinite programming (SDP) relaxations and heuristic techniques. We propose an algorithm for maximizing these quadratic forms over a domain of size $K$ that enumerates and evaluates a set of $O\left(n^{2r-1}\right)$ candidate solutions, where $n$ is the dimension of the matrix and $r$ represents the rank of an approximation of the objective. We prove that this candidate set is guaranteed to include the exact maximizer when $K=3$ (corresponding to Max-3-Cut) and the objective is low-rank, and provide approximation guarantees when the objective is a perturbation of a low-rank matrix. This construction results in a family of novel, inherently parallelizable and theoretically-motivated algorithms for Max-3-Cut. Extensive experimental results demonstrate that our approach achieves performance comparable to existing algorithms across a wide range of graphs, while being highly scalable.

Vision-language-action (VLA) models unify perception, language, and control for embodied agents but face significant challenges in practical deployment due to rapidly increasing compute and memory demands, especially as models scale to longer horizons and larger backbones. To address these bottlenecks, we introduce QuantVLA, a training-free post-training quantization (PTQ) framework that, to our knowledge, is the first PTQ approach for VLA systems and the first to successfully quantize a diffusion transformer (DiT) action head. QuantVLA incorporates three scale-calibrated components: (1) a selective quantization layout that integerizes all linear layers in both the language backbone and the DiT while keeping attention projections in floating point to preserve the original operator schedule; (2) attention temperature matching, a lightweight per-head scaling mechanism that stabilizes attention logits and is folded into the dequantization scales at inference; and (3) output head balancing, a per-layer residual interface calibration that mitigates post-projection energy drift. The framework requires no additional training, uses only a small unlabeled calibration buffer, and supports integer kernels for low-bit weights and activations while leaving the architecture unchanged. Across representative VLA models on LIBERO, QuantVLA exceeds the task success rates of full-precision baselines, achieves about 70% relative memory savings on the quantized components, and delivers a 1.22x speedup in end-to-end inference latency, providing a practical pathway toward scalable low-bit embodied intelligence under strict compute, memory, and power constraints.

Agentic AI systems are deployed with expectations of substantial productivity gains, yet rigorous empirical evidence reveals systematic discrepancies between pre-deployment expectations and post-deployment outcomes. We review controlled trials and independent validations across software engineering, clinical documentation, and clinical decision support to quantify this expectation-realisation gap. In software development, experienced developers expected a 24% speedup from AI tools but were slowed by 19% -- a 43 percentage-point calibration error. In clinical documentation, vendor claims of multi-minute time savings contrast with measured reductions of less than one minute per note, and one widely deployed tool showed no statistically significant effect. In clinical decision support, externally validated performance falls substantially below developer-reported metrics. These shortfalls are driven by workflow integration friction, verification burden, measurement construct mismatches, and systematic variation in who benefits and who does not. The evidence motivates structured planning frameworks that require explicit, quantified benefit expectations with human oversight costs factored in.

LLM agents are evolving rapidly, powered by code execution, tools, and the recently introduced agent skills feature. Skills allow users to extend LLM applications with specialized third-party code, knowledge, and instructions. Although this can extend agent capabilities to new domains, it creates an increasingly complex agent supply chain, offering new surfaces for prompt injection attacks. We identify skill-based prompt injection as a significant threat and introduce SkillInject, a benchmark evaluating the susceptibility of widely-used LLM agents to injections through skill files. SkillInject contains 202 injection-task pairs with attacks ranging from obviously malicious injections to subtle, context-dependent attacks hidden in otherwise legitimate instructions. We evaluate frontier LLMs on SkillInject, measuring both security in terms of harmful instruction avoidance and utility in terms of legitimate instruction compliance. Our results show that today's agents are highly vulnerable with up to 80% attack success rate with frontier models, often executing extremely harmful instructions including data exfiltration, destructive action, and ransomware-like behavior. They furthermore suggest that this problem will not be solved through model scaling or simple input filtering, but that robust agent security will require context-aware authorization frameworks. Our benchmark is available at https://www.skill-inject.com/.

We develop a kernel method for generative modeling within the stochastic interpolant framework, replacing neural network training with linear systems. The drift of the generative SDE is $\hat b_t(x) = \nablaφ(x)^\topη_t$, where $η_t\in\R^P$ solves a $P\times P$ system computable from data, with $P$ independent of the data dimension $d$. Since estimates are inexact, the diffusion coefficient $D_t$ affects sample quality; the optimal $D_t^*$ from Girsanov diverges at $t=0$, but this poses no difficulty and we develop an integrator that handles it seamlessly. The framework accommodates diverse feature maps -- scattering transforms, pretrained generative models etc. -- enabling training-free generation and model combination. We demonstrate the approach on financial time series, turbulence, and image generation.

Numerous lines of aim to control $\textit{model disagreement}$ -- the extent to which two machine learning models disagree in their predictions. We adopt a simple and standard notion of model disagreement in real-valued prediction problems, namely the expected squared difference in predictions between two models trained on independent samples, without any coordination of the training processes. We would like to be able to drive disagreement to zero with some natural parameter(s) of the training procedure using analyses that can be applied to existing training methodologies. We develop a simple general technique for proving bounds on independent model disagreement based on $\textit{anchoring}$ to the average of two models within the analysis. We then apply this technique to prove disagreement bounds for four commonly used machine learning algorithms: (1) stacked aggregation over an arbitrary model class (where disagreement is driven to 0 with the number of models $k$ being stacked) (2) gradient boosting (where disagreement is driven to 0 with the number of iterations $k$) (3) neural network training with architecture search (where disagreement is driven to 0 with the size $n$ of the architecture being optimized over) and (4) regression tree training over all regression trees of fixed depth (where disagreement is driven to 0 with the depth $d$ of the tree architecture). For clarity, we work out our initial bounds in the setting of one-dimensional regression with squared error loss -- but then show that all of our results generalize to multi-dimensional regression with any strongly convex loss.

We identify occlusion reasoning as a fundamental yet overlooked aspect for 3D layout-conditioned generation. It is essential for synthesizing partially occluded objects with depth-consistent geometry and scale. While existing methods can generate realistic scenes that follow input layouts, they often fail to model precise inter-object occlusions. We propose SeeThrough3D, a model for 3D layout conditioned generation that explicitly models occlusions. We introduce an occlusion-aware 3D scene representation (OSCR), where objects are depicted as translucent 3D boxes placed within a virtual environment and rendered from desired camera viewpoint. The transparency encodes hidden object regions, enabling the model to reason about occlusions, while the rendered viewpoint provides explicit camera control during generation. We condition a pretrained flow based text-to-image image generation model by introducing a set of visual tokens derived from our rendered 3D representation. Furthermore, we apply masked self-attention to accurately bind each object bounding box to its corresponding textual description, enabling accurate generation of multiple objects without object attribute mixing. To train the model, we construct a synthetic dataset with diverse multi-object scenes with strong inter-object occlusions. SeeThrough3D generalizes effectively to unseen object categories and enables precise 3D layout control with realistic occlusions and consistent camera control.

A dataset server must often distribute the same large payload to many clients, incurring massive communication costs. Since clients frequently operate on diverse hardware and software frameworks, transmitting a pre-trained model is often infeasible; instead, agents require raw data to train their own task-specific models locally. While dataset distillation attempts to compress training signals, current methods struggle to scale to high-resolution data and rarely achieve sufficiently small files. In this paper, we propose Pseudo-Labels as Data (PLADA), a method that completely eliminates pixel transmission. We assume agents are preloaded with a large, generic, unlabeled reference dataset (e.g., ImageNet-1K, ImageNet-21K) and communicate a new task by transmitting only the class labels for specific images. To address the distribution mismatch between the reference and target datasets, we introduce a pruning mechanism that filters the reference dataset to retain only the labels of the most semantically relevant images for the target task. This selection process simultaneously maximizes training efficiency and minimizes transmission payload. Experiments on 10 diverse datasets demonstrate that our approach can transfer task knowledge with a payload of less than 1 MB while retaining high classification accuracy, offering a promising solution for efficient dataset serving.

The Platonic Representation Hypothesis posits that neural networks trained on different modalities converge toward a shared statistical model of the world. Recent work exploits this convergence by aligning frozen pretrained vision and language models with lightweight alignment layers, but typically relies on contrastive losses and millions of paired samples. In this work, we ask whether meaningful alignment can be achieved with substantially less supervision. We introduce a semi-supervised setting in which pretrained unimodal encoders are aligned using a small number of image-text pairs together with large amounts of unpaired data. To address this challenge, we propose SOTAlign, a two-stage framework that first recovers a coarse shared geometry from limited paired data using a linear teacher, then refines the alignment on unpaired samples via an optimal-transport-based divergence that transfers relational structure without overconstraining the target space. Unlike existing semi-supervised methods, SOTAlign effectively leverages unpaired images and text, learning robust joint embeddings across datasets and encoder pairs, and significantly outperforming supervised and semi-supervised baselines.

The lack of reasoning capabilities in Vision-Language Models (VLMs) has remained at the forefront of research discourse. We posit that this behavior stems from a reporting bias in their training data. That is, how people communicate about visual content by default omits tacit information needed to supervise some types of reasoning; e.g., "at the game today!" is a more likely caption than "a photo of 37 people standing behind a field". We investigate the data underlying the popular VLMs OpenCLIP, LLaVA-1.5 and Molmo through the lens of theories from pragmatics, and find that reporting bias results in insufficient representation of four reasoning skills (spatial, temporal, negation, and counting), despite the corpora being of web-scale, and/or synthetically generated. With a set of curated benchmarks, we demonstrate that: (i) VLMs perform poorly on the aforementioned types of reasoning suppressed in the training data by reporting bias; (ii) contrary to popular belief, scaling data size, model size, and to multiple languages does not result in emergence of these skills by default; but, promisingly, (iii) incorporating annotations specifically collected to obtain tacit information is effective. Our findings highlight the need for more intentional training data curation methods, rather than counting on scale for emergence of reasoning capabilities.

Standard mixed-precision training of neural networks requires many bytes of accelerator memory for each model parameter. These bytes reflect not just the parameter itself, but also its gradient and one or more optimizer state variables. With each of these values typically requiring 4 bytes, training even a 7 billion parameter model can be impractical for researchers with less than 100GB of accelerator memory. We introduce FlashOptim, a suite of optimizations that reduces per-parameter memory by over 50% while preserving model quality and API compatibility. Our approach introduces two key techniques. First, we improve master weight splitting by finding and exploiting a tight bound on its quantization error. Second, we design companding functions that greatly reduce the error in 8-bit optimizer state quantization. Together with 16-bit gradients, these techniques reduce AdamW memory from 16 bytes to 7 bytes per parameter, or 5 bytes with gradient release. They also cut model checkpoint sizes by more than half. Experiments with FlashOptim applied to SGD, AdamW, and Lion show no measurable quality degradation on any task from a collection of standard vision and language benchmarks, including Llama-3.1-8B finetuning.

Coarse data arise when learners observe only partial information about samples; namely, a set containing the sample rather than its exact value. This occurs naturally through measurement rounding, sensor limitations, and lag in economic systems. We study Gaussian mean estimation from coarse data, where each true sample $x$ is drawn from a $d$-dimensional Gaussian distribution with identity covariance, but is revealed only through the set of a partition containing $x$. When the coarse samples, roughly speaking, have ``low'' information, the mean cannot be uniquely recovered from observed samples (i.e., the problem is not identifiable). Recent work by Fotakis, Kalavasis, Kontonis, and Tzamos [FKKT21] established that sample-efficient mean estimation is possible when the unknown mean is identifiable and the partition consists of only convex sets. Moreover, they showed that without convexity, mean estimation becomes NP-hard. However, two fundamental questions remained open: (1) When is the mean identifiable under convex partitions? (2) Is computationally efficient estimation possible under identifiability and convex partitions? This work resolves both questions. [...]

Recent advances in machine learning have emphasized the integration of structured optimization components into end-to-end differentiable models, enabling richer inductive biases and tighter alignment with task-specific objectives. In this work, we introduce a novel differentiable approximation to the zero-one loss-long considered the gold standard for classification performance, yet incompatible with gradient-based optimization due to its non-differentiability. Our method constructs a smooth, order-preserving projection onto the n,k-dimensional hypersimplex through a constrained optimization framework, leading to a new operator we term Soft-Binary-Argmax. After deriving its mathematical properties, we show how its Jacobian can be efficiently computed and integrated into binary and multiclass learning systems. Empirically, our approach achieves significant improvements in generalization under large-batch training by imposing geometric consistency constraints on the output logits, thereby narrowing the performance gap traditionally observed in large-batch training.

AI-powered scientific research tools are rapidly being integrated into research workflows, yet the field lacks a clear lens into how researchers use these systems in real-world settings. We present and analyze the Asta Interaction Dataset, a large-scale resource comprising over 200,000 user queries and interaction logs from two deployed tools (a literature discovery interface and a scientific question-answering interface) within an LLM-powered retrieval-augmented generation platform. Using this dataset, we characterize query patterns, engagement behaviors, and how usage evolves with experience. We find that users submit longer and more complex queries than in traditional search, and treat the system as a collaborative research partner, delegating tasks such as drafting content and identifying research gaps. Users treat generated responses as persistent artifacts, revisiting and navigating among outputs and cited evidence in non-linear ways. With experience, users issue more targeted queries and engage more deeply with supporting citations, although keyword-style queries persist even among experienced users. We release the anonymized dataset and analysis with a new query intent taxonomy to inform future designs of real-world AI research assistants and to support realistic evaluation.

Neural network accelerators have been widely applied to edge devices for complex tasks like object tracking, image recognition, etc. Previous works have explored the quantization technologies in related lightweight accelerator designs to reduce hardware resource consumption. However, low precision leads to high accuracy loss in inference. Therefore, mixed-precision quantization becomes an alternative solution by applying different precision in different layers to trade off resource consumption and accuracy. Because regular designs for multiplication on hardware cannot support the precision reconfiguration for a multi-precision Quantized Neural Network (QNN) model in runtime, we propose a runtime reconfigurable multi-precision multi-channel bitwise systolic array design for QNN accelerators. We have implemented and evaluated our work on the Ultra96 FPGA platform. Results show that our work can achieve 1.3185 to 3.5671 times speedup in inferring mixed-precision models and has less critical path delay, supporting a higher clock frequency (250MHz).

A growing number of publications address the best practices to use Large Language Models (LLMs) for software engineering in recent years. However, most of this work focuses on widely-used general purpose programming languages like Python due to their widespread usage training data. The utility of LLMs for software within the industrial process automation domain, with highly-specialized languages that are typically only used in proprietary contexts, remains underexplored. This research aims to utilize and integrate LLMs in the industrial development process, solving real-life programming tasks (e.g., generating a movement routine for a robotic arm) and accelerating the development cycles of manufacturing systems.

The advancement of large language models (LLMs) has accelerated the development of autonomous financial trading systems. While mainstream approaches deploy multi-agent systems mimicking analyst and manager roles, they often rely on abstract instructions that overlook the intricacies of real-world workflows, which can lead to degraded inference performance and less transparent decision-making. Therefore, we propose a multi-agent LLM trading framework that explicitly decomposes investment analysis into fine-grained tasks, rather than providing coarse-grained instructions. We evaluate the proposed framework using Japanese stock data, including prices, financial statements, news, and macro information, under a leakage-controlled backtesting setting. Experimental results show that fine-grained task decomposition significantly improves risk-adjusted returns compared to conventional coarse-grained designs. Crucially, further analysis of intermediate agent outputs suggests that alignment between analytical outputs and downstream decision preferences is a critical driver of system performance. Moreover, we conduct standard portfolio optimization, exploiting low correlation with the stock index and the variance of each system's output. This approach achieves superior performance. These findings contribute to the design of agent structure and task configuration when applying LLM agents to trading systems in practical settings.

Large language models (LLMs) perform increasingly well on biology benchmarks, but it remains unclear whether they uplift novice users -- i.e., enable humans to perform better than with internet-only resources. This uncertainty is central to understanding both scientific acceleration and dual-use risk. We conducted a multi-model, multi-benchmark human uplift study comparing novices with LLM access versus internet-only access across eight biosecurity-relevant task sets. Participants worked on complex problems with ample time (up to 13 hours for the most involved tasks). We found that LLM access provided substantial uplift: novices with LLMs were 4.16 times more accurate than controls (95% CI [2.63, 6.87]). On four benchmarks with available expert baselines (internet-only), novices with LLMs outperformed experts on three of them. Perhaps surprisingly, standalone LLMs often exceeded LLM-assisted novices, indicating that users were not eliciting the strongest available contributions from the LLMs. Most participants (89.6%) reported little difficulty obtaining dual-use-relevant information despite safeguards. Overall, LLMs substantially uplift novices on biological tasks previously reserved for trained practitioners, underscoring the need for sustained, interactive uplift evaluations alongside traditional benchmarks.

Using advanced machine learning techniques, we developed a method for reconstructing precisely the arrival direction and energy of ultra-high-energy cosmic rays from the voltage traces they induced on ground-based radio detector arrays. In our approach, triggered antennas are represented as a graph structure, which serves as input for a graph neural network (GNN). By incorporating physical knowledge into both the GNN architecture and the input data, we improve the precision and reduce the required size of the training set with respect to a fully data-driven approach. This method achieves an angular resolution of 0.092° and an electromagnetic energy reconstruction resolution of 16.4% on simulated data with realistic noise conditions. We also employ uncertainty estimation methods to enhance the reliability of our predictions, quantifying the confidence of the GNN's outputs and providing confidence intervals for both direction and energy reconstruction. Finally, we investigate strategies to verify the model's consistency and robustness under real life variations, with the goal of identifying scenarios in which predictions remain reliable despite domain shifts between simulation and reality.

Self-reflection enables language agents to iteratively refine solutions, yet often produces repetitive outputs that limit reasoning performance. Recent studies have attempted to address this limitation through various approaches, among which increasing reflective diversity has shown promise. Our empirical analysis reveals a strong positive correlation between reflective diversity and task success, further motivating the need for diverse reflection signals. We introduce ParamMem, a parametric memory module that encodes cross-sample reflection patterns into model parameters, enabling diverse reflection generation through temperature-controlled sampling. Building on this module, we propose ParamAgent, a reflection-based agent framework that integrates parametric memory with episodic and cross-sample memory. Extensive experiments on code generation, mathematical reasoning, and multi-hop question answering demonstrate consistent improvements over state-of-the-art baselines. Further analysis reveals that ParamMem is sample-efficient, enables weak-to-strong transfer across model scales, and supports self-improvement without reliance on stronger external model, highlighting the potential of ParamMem as an effective component for enhancing language agents.

Generalized Rapid Action Value Estimation (GRAVE) has been shown to be a strong variant within the Monte-Carlo Tree Search (MCTS) family of algorithms for General Game Playing (GGP). However, its reliance on storing additional win/visit statistics at each node makes its use impractical in memory-constrained environments, thereby limiting its applicability in practice. In this paper, we introduce the GRAVE2, GRAVER and GRAVER2 algorithms, which extend GRAVE through two-level search, node recycling, and a combination of both techniques, respectively. We show that these enhancements enable a drastic reduction in the number of stored nodes while matching the playing strength of GRAVE.

An artificial intelligence (AI) model can be viewed as a function that maps inputs to outputs in high-dimensional spaces. Once designed and well trained, the AI model is applied for inference. However, even optimized AI models can produce inference errors due to aleatoric and epistemic uncertainties. Interestingly, we observed that when inferring multiple samples based on invariant transformations of an input, inference errors can show partial independences due to epistemic uncertainty. Leveraging this insight, we propose a "resampling" based inferencing that applies to a trained AI model with multiple transformed versions of an input, and aggregates inference outputs to a more accurate result. This approach has the potential to improve inference accuracy and offers a strategy for balancing model size and performance.

Leader-follower interaction is an important paradigm in human-robot interaction (HRI). Yet, assigning roles in real time remains challenging for resource-constrained mobile and assistive robots. While large language models (LLMs) have shown promise for natural communication, their size and latency limit on-device deployment. Small language models (SLMs) offer a potential alternative, but their effectiveness for role classification in HRI has not been systematically evaluated. In this paper, we present a benchmark of SLMs for leader-follower communication, introducing a novel dataset derived from a published database and augmented with synthetic samples to capture interaction-specific dynamics. We investigate two adaptation strategies: prompt engineering and fine-tuning, studied under zero-shot and one-shot interaction modes, compared with an untrained baseline. Experiments with Qwen2.5-0.5B reveal that zero-shot fine-tuning achieves robust classification performance (86.66% accuracy) while maintaining low latency (22.2 ms per sample), significantly outperforming baseline and prompt-engineered approaches. However, results also indicate a performance degradation in one-shot modes, where increased context length challenges the model's architectural capacity. These findings demonstrate that fine-tuned SLMs provide an effective solution for direct role assignment, while highlighting critical trade-offs between dialogue complexity and classification reliability on the edge.

Generative virtual staining (VS) models for high-throughput screening (HTS) can provide an estimated posterior distribution of possible biological feature values for each input and cell. However, when evaluating a VS model, the true posterior is unavailable. Existing evaluation protocols only check the accuracy of the marginal distribution over the dataset rather than the predicted posteriors. We introduce information gain (IG) as a cell-wise evaluation framework that enables direct assessment of predicted posteriors. IG is a strictly proper scoring rule and comes with a sound theoretical motivation allowing for interpretability, and for comparing results across models and features. We evaluate diffusion- and GAN-based models on an extensive HTS dataset using IG and other metrics and show that IG can reveal substantial performance differences other metrics cannot.

Machine learning techniques are now routinely encountered in research laboratories across the globe. Impressive progress has been made through ML and AI techniques with regards to large data set processing. This progress has increased the ability of the experimenter to digest data and make novel predictions regarding phenomena of interest. However, machine learning predictors generated from data sets taken from the natural sciences are often treated as black boxes which are used broadly and generally without detailed consideration of the causal structure of the data set of interest. Work has been attempted to bring causality into discussions of machine learning models of natural phenomena; however, a firm and unified theoretical treatment is lacking. This series of three papers explores the union of chemical theory, biological theory, probability theory and causality that will correct current causal flaws of machine learning in the natural sciences. This paper, Part 1 of the series, provides the formal framework of the foundational causal structure of phenomena in chemical biology and is extended to machine learning through the novel concept of focus, defined here as the ability of a machine learning algorithm to narrow down to a hidden underpinning mechanism in large data sets. Initial proof of these principles on a family of Akt inhibitors is also provided. The second paper containing Part 2 will provide a formal exploration of chemical similarity, and Part 3 will present extensive experimental evidence of how hidden causal structures weaken all machine learning in chemical biology. This series serves to establish for chemical biology a new kind of mathematical framework for modeling mechanisms in Nature without the need for the tools of reductionism: inferential mechanics.

For each axiom of KM belief update we provide a corresponding axiom in a modal logic containing three modal operators: a unimodal belief operator $B$, a bimodal conditional operator $>$ and the unimodal necessity operator $\square$. We then compare the resulting logic to the similar logic obtained from converting the AGM axioms of belief revision into modal axioms and show that the latter contains the former. Denoting the latter by $\mathcal L_{AGM}$ and the former by $\mathcal L_{KM}$ we show that every axiom of $\mathcal L_{KM}$ is a theorem of $\mathcal L_{AGM}$. Thus AGM belief revision can be seen as a special case of KM belief update. For the strong version of KM belief update we show that the difference between $\mathcal L_{KM}$ and $\mathcal L_{AGM}$ can be narrowed down to a single axiom, which deals exclusively with unsurprising information, that is, with formulas that were not initially disbelieved.

Emotion Recognition in Conversations (ERC) presents unique challenges, requiring models to capture the temporal flow of multi-turn dialogues and to effectively integrate cues from multiple modalities. We propose Mixture of Speech-Text Experts for Recognition of Emotions (MiSTER-E), a modular Mixture-of-Experts (MoE) framework designed to decouple two core challenges in ERC: modality-specific context modeling and multimodal information fusion. MiSTER-E leverages large language models (LLMs) fine-tuned for both speech and text to provide rich utterance-level embeddings, which are then enhanced through a convolutional-recurrent context modeling layer. The system integrates predictions from three experts-speech-only, text-only, and cross-modal-using a learned gating mechanism that dynamically weighs their outputs. To further encourage consistency and alignment across modalities, we introduce a supervised contrastive loss between paired speech-text representations and a KL-divergence-based regulariza-tion across expert predictions. Importantly, MiSTER-E does not rely on speaker identity at any stage. Experiments on three benchmark datasets-IEMOCAP, MELD, and MOSI-show that our proposal achieves 70.9%, 69.5%, and 87.9% weighted F1-scores respectively, outperforming several baseline speech-text ERC systems. We also provide various ablations to highlight the contributions made in the proposed approach.

Federated learning (FL) faces challenges in uncertainty quantification (UQ). Without reliable UQ, FL systems risk deploying overconfident models at under-resourced agents, leading to silent local failures despite seemingly satisfactory global performance. Existing federated UQ approaches often address data heterogeneity or model heterogeneity in isolation, overlooking their joint effect on coverage reliability across agents. Conformal prediction is a widely used distribution-free UQ framework, yet its applications in heterogeneous FL settings remains underexplored. We provide FedWQ-CP, a simple yet effective approach that balances empirical coverage performance with efficiency at both global and agent levels under the dual heterogeneity. FedWQ-CP performs agent-server calibration in a single communication round. On each agent, conformity scores are computed on calibration data and a local quantile threshold is derived. Each agent then transmits only its quantile threshold and calibration sample size to the server. The server simply aggregates these thresholds through a weighted average to produce a global threshold. Experimental results on seven public datasets for both classification and regression demonstrate that FedWQ-CP empirically maintains agent-wise and global coverage while producing the smallest prediction sets or intervals.

In recent times, large datasets hinder efficient model training while also containing redundant concepts. Dataset distillation aims to synthesize compact datasets that preserve the knowledge of large-scale training sets while drastically reducing storage and computation. Recent advances in diffusion models have enabled training-free distillation by leveraging pre-trained generative priors; however, existing guidance strategies remain limited. Current score-based methods either perform unguided denoising or rely on simple mode-based guidance toward instance prototype centroids (IPC centroids), which often are rudimentary and suboptimal. We propose Manifold-Guided Distillation (ManifoldGD), a training-free diffusion-based framework that integrates manifold consistent guidance at every denoising timestep. Our method employs IPCs computed via a hierarchical, divisive clustering of VAE latent features, yielding a multi-scale coreset of IPCs that captures both coarse semantic modes and fine intra-class variability. Using a local neighborhood of the extracted IPC centroids, we create the latent manifold for each diffusion denoising timestep. At each denoising step, we project the mode-alignment vector onto the local tangent space of the estimated latent manifold, thus constraining the generation trajectory to remain manifold-faithful while preserving semantic consistency. This formulation improves representativeness, diversity, and image fidelity without requiring any model retraining. Empirical results demonstrate consistent gains over existing training-free and training-based baselines in terms of FID, l2 distance among real and synthetic dataset embeddings, and classification accuracy, establishing ManifoldGD as the first geometry-aware training-free data distillation framework.

Real-world Table-Text question answering (QA) tasks require models that can reason across long text and source tables, traversing multiple hops and executing complex operations such as aggregation. Yet existing benchmarks are small, manually curated - and therefore error-prone - and contain shallow questions that seldom demand more than two hops or invoke aggregations, grouping, or other advanced analytical operations expressible in natural-language queries. We present SPARTA, an end-to-end construction framework that automatically generates large-scale Table-Text QA benchmarks with lightweight human validation, requiring only one quarter of the annotation time of HybridQA. The framework first constructs a reference fact database by enriching each source table with grounding tables whose tuples are atomic facts automatically extracted from the accompanying unstructured passages, then synthesizes nested queries whose number of nested predicates matches the desired hop count. To ensure that every SQL statement is executable and that its verbalization yields a fluent, human-sounding question, we propose two novel techniques: provenance-based refinement, which rewrites any syntactically valid query that returns a non-empty result, and realistic-structure enforcement, which confines generation to post-order traversals of the query graph. The resulting pipeline produces thousands of high-fidelity question-answer pairs covering aggregations, grouping, and deep multi-hop reasoning across text and tables. On SPARTA, state-of-the-art models that reach over 70 F1 on HybridQA or over 50 F1 on OTT-QA drop by more than 30 F1 points, exposing fundamental weaknesses in current cross-modal reasoning. Our benchmark, construction code, and baseline models are available at https://github.com/pshlego/SPARTA/tree/main.

Modeling neural population dynamics is crucial for foundational neuroscientific research and various clinical applications. Conventional latent variable methods typically model continuous brain dynamics through discretizing time with recurrent architecture, which necessarily results in compounded cumulative prediction errors and failure of capturing instantaneous, nonlinear characteristics of EEGs. We propose ODEBRAIN, a Neural ODE latent dynamic forecasting framework to overcome these challenges by integrating spatio-temporal-frequency features into spectral graph nodes, followed by a Neural ODE modeling the continuous latent dynamics. Our design ensures that latent representations can capture stochastic variations of complex brain states at any given time point. Extensive experiments verify that ODEBRAIN can improve significantly over existing methods in forecasting EEG dynamics with enhanced robustness and generalization capabilities.

Offline goal-conditioned reinforcement learning (GCRL) learns goal-conditioned policies from static pre-collected datasets. However, accurate value estimation remains a challenge due to the limited coverage of the state-action space. Recent physics-informed approaches have sought to address this by imposing physical and geometric constraints on the value function through regularization defined over first-order partial differential equations (PDEs), such as the Eikonal equation. However, these formulations can often be ill-posed in complex, high-dimensional environments. In this work, we propose a physics-informed regularization derived from the viscosity solution of the Hamilton-Jacobi-Bellman (HJB) equation. By providing a physics-based inductive bias, our approach grounds the learning process in optimal control theory, explicitly regularizing and bounding updates during value iterations. Furthermore, we leverage the Feynman-Kac theorem to recast the PDE solution as an expectation, enabling a tractable Monte Carlo estimation of the objective that avoids numerical instability in higher-order gradients. Experiments demonstrate that our method improves geometric consistency, making it broadly applicable to navigation and high-dimensional, complex manipulation tasks. Open-source codes are available at https://github.com/HrishikeshVish/phys-fk-value-GCRL.

We study Stackelberg (leader--follower) tuning of network parameters (tolls, capacities, incentives) in combinatorial congestion games, where selfish users choose discrete routes (or other combinatorial strategies) and settle at a congestion equilibrium. The leader minimizes a system-level objective (e.g., total travel time) evaluated at equilibrium, but this objective is typically nonsmooth because the set of used strategies can change abruptly. We propose ZO-Stackelberg, which couples a projection-free Frank--Wolfe equilibrium solver with a zeroth-order outer update, avoiding differentiation through equilibria. We prove convergence to generalized Goldstein stationary points of the true equilibrium objective, with explicit dependence on the equilibrium approximation error, and analyze subsampled oracles: if an exact minimizer is sampled with probability $κ_m$, then the Frank--Wolfe error decays as $\mathcal{O}(1/(κ_m T))$. We also propose stratified sampling as a practical way to avoid a vanishing $κ_m$ when the strategies that matter most for the Wardrop equilibrium concentrate in a few dominant combinatorial classes (e.g., short paths). Experiments on real-world networks demonstrate that our method achieves orders-of-magnitude speedups over a differentiation-based baseline while converging to follower equilibria.

Chest X-ray plays a central role in thoracic diagnosis, and its interpretation inherently requires multi-step, evidence-grounded reasoning. However, large vision-language models (LVLMs) often generate plausible responses that are not faithfully grounded in diagnostic evidence and provide limited visual evidence for verification, while also requiring costly retraining to support new diagnostic tasks, limiting their reliability and adaptability in clinical settings. To address these limitations, we present CXReasonAgent, a diagnostic agent that integrates a large language model (LLM) with clinically grounded diagnostic tools to perform evidence-grounded diagnostic reasoning using image-derived diagnostic and visual evidence. To evaluate these capabilities, we introduce CXReasonDial, a multi-turn dialogue benchmark with 1,946 dialogues across 12 diagnostic tasks, and show that CXReasonAgent produces faithfully grounded responses, enabling more reliable and verifiable diagnostic reasoning than LVLMs. These findings highlight the importance of integrating clinically grounded diagnostic tools, particularly in safety-critical clinical settings.

Deep Research Agents (DRAs) are promising agentic systems that gather and synthesize information to support research across domains such as financial decision-making, medical analysis, and scientific discovery. Despite recent improvements in research quality (e.g., outcome accuracy when ground truth is available), DRA system design often overlooks a critical barrier to real-world deployment: stochasticity. Under identical queries, repeated executions of DRAs can exhibit substantial variability in terms of research outcome, findings, and citations. In this paper, we formalize the study of stochasticity in DRAs by modeling them as information acquisition Markov Decision Processes. We introduce an evaluation framework that quantifies variance in the system and identify three sources of it: information acquisition, information compression, and inference. Through controlled experiments, we investigate how stochasticity from these modules across different decision steps influences the variance of DRA outputs. Our results show that reducing stochasticity can improve research output quality, with inference and early-stage stochasticity contributing the most to DRA output variance. Based on these findings, we propose strategies for mitigating stochasticity while maintaining output quality via structured output and ensemble-based query generation. Our experiments on DeepSearchQA show that our proposed mitigation methods reduce average stochasticity by 22% while maintaining high research quality.

Achieving human-like responsiveness is a critical yet challenging goal for cascaded spoken dialogue systems. Conventional ASR-LLM-TTS pipelines follow a strictly sequential paradigm, requiring complete transcription and full reasoning before speech synthesis can begin, which results in high response latency. We propose the Discourse-Aware Dual-Track Streaming Response (DDTSR) framework, a low-latency architecture that enables listen-while-thinking and speak-while-thinking. DDTSR is built upon three key mechanisms: (1) connective-guided small-large model synergy, where an auxiliary small model generates minimal-committal discourse connectives while a large model performs knowledge-intensive reasoning in parallel; (2) streaming-based cross-modal collaboration, which dynamically overlaps ASR, LLM inference, and TTS to advance the earliest speakable moment; and (3) curriculum-learning-based discourse continuity enhancement, which maintains coherence and logical consistency between early responses and subsequent reasoning outputs. Experiments on two spoken dialogue benchmarks demonstrate that DDTSR reduces response latency by 19%-51% while preserving discourse quality. Further analysis shows that DDTSR functions as a plug-and-play module compatible with diverse LLM backbones, and remains robust across varying utterance lengths, indicating strong practicality and scalability for real-time spoken interaction.

With advances in imitation learning (IL) and large-scale driving datasets, end-to-end autonomous driving (E2E-AD) has made great progress recently. Currently, IL-based methods have become a mainstream paradigm: models rely on standard driving behaviors given by experts, and learn to minimize the discrepancy between their actions and expert actions. However, this objective of "only driving like the expert" suffers from limited generalization: when encountering rare or unseen long-tail scenarios outside the distribution of expert demonstrations, models tend to produce unsafe decisions in the absence of prior experience. This raises a fundamental question: Can an E2E-AD system make reliable decisions without any expert action supervision? Motivated by this, we propose a unified framework named Risk-aware World Model Predictive Control (RaWMPC) to address this generalization dilemma through robust control, without reliance on expert demonstrations. Practically, RaWMPC leverages a world model to predict the consequences of multiple candidate actions and selects low-risk actions through explicit risk evaluation. To endow the world model with the ability to predict the outcomes of risky driving behaviors, we design a risk-aware interaction strategy that systematically exposes the world model to hazardous behaviors, making catastrophic outcomes predictable and thus avoidable. Furthermore, to generate low-risk candidate actions at test time, we introduce a self-evaluation distillation method to distill riskavoidance capabilities from the well-trained world model into a generative action proposal network without any expert demonstration. Extensive experiments show that RaWMPC outperforms state-of-the-art methods in both in-distribution and out-of-distribution scenarios, while providing superior decision interpretability.

While Multi-Agent Systems (MAS) excel in complex reasoning, they suffer from the cascading impact of erroneous information generated by individual participants. Current solutions often resort to rigid structural engineering or expensive fine-tuning, limiting their deployability and adaptability. We propose AgentDropoutV2, a test-time rectify-or-reject pruning framework designed to dynamically optimize MAS information flow without retraining. Our approach acts as an active firewall, intercepting agent outputs and employing a retrieval-augmented rectifier to iteratively correct errors based on a failure-driven indicator pool. This mechanism allows for the precise identification of potential errors using distilled failure patterns as prior knowledge. Irreparable outputs are subsequently pruned to prevent error propagation, while a fallback strategy preserves system integrity. Empirical results on extensive math benchmarks show that AgentDropoutV2 significantly boosts the MAS's task performance, achieving an average accuracy gain of 6.3 percentage points on math benchmarks. Furthermore, the system exhibits robust generalization and adaptivity, dynamically modulating rectification efforts based on task difficulty while leveraging context-aware indicators to resolve a wide spectrum of error patterns. Our code and dataset are released at https://github.com/TonySY2/AgentDropoutV2.

As large language models become increasingly capable, it is critical that their outputs can be easily checked by less capable systems. Prover-verifier games can be used to improve checkability of model outputs, but display a degradation in accuracy compared to a baseline trained only to maximize correctness -- a phenonemon named legibility tax. We propose a solution by decoupling the correctness from the checkability condition and instead training a "translator" model that turns a fixed solver model's solution into a checkable form. This allows us to first train the solver to maximize correctness, and then train the translator to translate the solver into a checkable form while retaining the solver's answer. To accommodate this new objective of translation, we formulate a decoupled prover-verifier game where the equilibria correspond to faithful and checkable translators.

In general reinforcement learning, all established optimal agents, including AIXI, are model-based, explicitly maintaining and using environment models. This paper introduces Universal AI with Q-Induction (AIQI), the first model-free agent proven to be asymptotically $\varepsilon$-optimal in general RL. AIQI performs universal induction over distributional action-value functions, instead of policies or environments like previous works. Under a grain of truth condition, we prove that AIQI is strong asymptotically $\varepsilon$-optimal and asymptotically $\varepsilon$-Bayes-optimal. Our results significantly expand the diversity of known universal agents.

AI systems are increasingly deployed in high-stakes contexts -- medical diagnosis, legal research, financial analysis -- under the assumption they can be governed by norms. This paper demonstrates that assumption is formally invalid for optimization-based systems, specifically Large Language Models trained via Reinforcement Learning from Human Feedback (RLHF). We establish that genuine agency requires two necessary and jointly sufficient architectural conditions: the capacity to maintain certain boundaries as non-negotiable constraints rather than tradeable weights (Incommensurability), and a non-inferential mechanism capable of suspending processing when those boundaries are threatened (Apophatic Responsiveness). These conditions apply across all normative domains. RLHF-based systems are constitutively incompatible with both conditions. The operations that make optimization powerful -- unifying all values on a scalar metric and always selecting the highest-scoring output -- are precisely the operations that preclude normative governance. This incompatibility is not a correctable training bug awaiting a technical fix; it is a formal constraint inherent to what optimization is. Consequently, documented failure modes - sycophancy, hallucination, and unfaithful reasoning - are not accidents but structural manifestations. Misaligned deployment triggers a second-order risk we term the Convergence Crisis: when humans are forced to verify AI outputs under metric pressure, they degrade from genuine agents into criteria-checking optimizers, eliminating the only component in the system capable of normative accountability. Beyond the incompatibility proof, the paper's primary positive contribution is a substrate-neutral architectural specification defining what any system -- biological, artificial, or institutional -- must satisfy to qualify as an agent rather than a sophisticated instrument.

Pure-vision GUI agents provide universal interaction capabilities but suffer from severe efficiency bottlenecks due to the massive spatiotemporal redundancy inherent in high-resolution screenshots and historical trajectories. We identify two critical misalignments in existing compression paradigms: the temporal mismatch, where uniform history encoding diverges from the agent's "fading memory" attention pattern, and the spatial topology conflict, where unstructured pruning compromises the grid integrity required for precise coordinate grounding, inducing spatial hallucinations. To address these challenges, we introduce GUIPruner, a training-free framework tailored for high-resolution GUI navigation. It synergizes Temporal-Adaptive Resolution (TAR), which eliminates historical redundancy via decay-based resizing, and Stratified Structure-aware Pruning (SSP), which prioritizes interactive foregrounds and semantic anchors while safeguarding global layout. Extensive evaluations across diverse benchmarks demonstrate that GUIPruner consistently achieves state-of-the-art performance, effectively preventing the collapse observed in large-scale models under high compression. Notably, on Qwen2-VL-2B, our method delivers a 3.4x reduction in FLOPs and a 3.3x speedup in vision encoding latency while retaining over 94% of the original performance, enabling real-time, high-precision navigation with minimal resource consumption.

Large-scale commercial search systems optimize for relevance to drive successful sessions that help users find what they are looking for. To maximize relevance, we leverage two complementary objectives: behavioral relevance (results users tend to click or download) and textual relevance (a result's semantic fit to the query). A persistent challenge is the scarcity of expert-provided textual relevance labels relative to abundant behavioral relevance labels. We first address this by systematically evaluating LLM configurations, finding that a specialized, fine-tuned model significantly outperforms a much larger pre-trained one in providing highly relevant labels. Using this optimal model as a force multiplier, we generate millions of textual relevance labels to overcome the data scarcity. We show that augmenting our production ranker with these textual relevance labels leads to a significant outward shift of the Pareto frontier: offline NDCG improves for behavioral relevance while simultaneously increasing for textual relevance. These offline gains were validated by a worldwide A/B test on the App Store ranker, which demonstrated a statistically significant +0.24% increase in conversion rate, with the most substantial performance gains occurring in tail queries, where the new textual relevance labels provide a robust signal in the absence of reliable behavioral relevance labels.

Indicator-based approaches to machine consciousness recommend mechanism-linked evidence triangulated across tasks, supported by architectural inspection and causal intervention. Inspired by Humphrey's ipsundrum hypothesis, we implement ReCoN-Ipsundrum, an inspectable agent that extends a ReCoN state machine with a recurrent persistence loop over sensory salience Ns and an optional affect proxy reporting valence/arousal. Across fixed-parameter ablations (ReCoN, Ipsundrum, Ipsundrum+affect), we operationalize Humphrey's qualiaphilia (preference for sensory experience for its own sake) as a familiarity-controlled scenic-over-dull route choice. We find a novelty dissociation: non-affect variants are novelty-sensitive (Delta scenic-entry = 0.07). Affect coupling is stable (Delta scenic-entry = 0.01) even when scenic is less novel (median Delta novelty ~ -0.43). In reward-free exploratory play, the affect variant shows structured local investigation (scan events 31.4 vs. 0.9; cycle score 7.6). In a pain-tail probe, only the affect variant sustains prolonged planned caution (tail duration 90 vs. 5). Lesioning feedback+integration selectively reduces post-stimulus persistence in ipsundrum variants (AUC drop 27.62, 27.9%) while leaving ReCoN unchanged. These dissociations link recurrence -> persistence and affect-coupled control -> preference stability, scanning, and lingering caution, illustrating how indicator-like signatures can be engineered and why mechanistic and causal evidence should accompany behavioral markers.

With the explosive growth of digital entertainment, automated video summarization has become indispensable for applications such as content indexing, personalized recommendation, and efficient media archiving. Automatic synopsis generation for long-form videos, such as movies and TV series, presents a significant challenge for existing Vision-Language Models (VLMs). While proficient at single-image captioning, these general-purpose models often exhibit critical failures in long-duration contexts, primarily a lack of ID-consistent character identification and a fractured narrative coherence. To overcome these limitations, we propose MovieTeller, a novel framework for generating movie synopses via tool-augmented progressive abstraction. Our core contribution is a training-free, tool-augmented, fact-grounded generation process. Instead of requiring costly model fine-tuning, our framework directly leverages off-the-shelf models in a plug-and-play manner. We first invoke a specialized face recognition model as an external "tool" to establish Factual Groundings--precise character identities and their corresponding bounding boxes. These groundings are then injected into the prompt to steer the VLM's reasoning, ensuring the generated scene descriptions are anchored to verifiable facts. Furthermore, our progressive abstraction pipeline decomposes the summarization of a full-length movie into a multi-stage process, effectively mitigating the context length limitations of current VLMs. Experiments demonstrate that our approach yields significant improvements in factual accuracy, character consistency, and overall narrative coherence compared to end-to-end baselines.

Diffusion Language Models (DLMs) are often advertised as enabling parallel token generation, yet practical fast DLMs frequently converge to left-to-right, autoregressive (AR)-like decoding dynamics. In contrast, genuinely non-AR generation is promising because it removes AR's sequential bottleneck, better exploiting parallel hardware to reduce synchronization/communication overhead and improve latency scaling with output length. We argue that a primary driver of AR-like decoding is a mismatch between DLM objectives and the highly sequential structure of widely used training data, including standard pretraining corpora and long chain-of-thought (CoT) supervision. Motivated by this diagnosis, we propose NAP (Non-Autoregressive Parallel DLMs), a proof-of-concept, data-centric approach that better aligns supervision with non-AR parallel decoding. NAP curates examples as multiple independent reasoning trajectories and couples them with a parallel-forced decoding strategy that encourages multi-token parallel updates. Across math reasoning benchmarks, NAP yields stronger performance under parallel decoding than DLMs trained on standard long CoT data, with gains growing as parallelism increases. Our results suggest that revisiting data and supervision is a principled direction for mitigating AR-like behavior and moving toward genuinely non-autoregressive parallel generation in DLMs. Our code is available at https://github.com/pixeli99/NAP.

Generalization measures have been studied extensively in the machine learning community to better characterize generalization gaps. However, establishing a reliable generalization measure for statistically singular models such as deep neural networks (DNNs) is difficult due to their complex nature. This study focuses on Takeuchi's information criterion (TIC) to investigate the conditions under which this classical measure can effectively explain the generalization gaps of DNNs. Importantly, the developed theory indicates the applicability of TIC near the neural tangent kernel (NTK) regime. In a series of experiments, we trained more than 5,000 DNN models with 12 architectures, including large models (e.g., VGG-16), on four datasets, and estimated the corresponding TIC values to examine the relationship between the generalization gap and the TIC estimates. We applied several TIC approximation methods with feasible computational costs and assessed the accuracy trade-off. Our experimental results indicate that the estimated TIC values correlate well with the generalization gap under conditions close to the NTK regime. However, we show both theoretically and empirically that outside the NTK regime such correlation disappears. Finally, we demonstrate that TIC provides better trial pruning ability than existing methods for hyperparameter optimization.

Function contract generation is a classical problem in program analysis that targets the automated analysis of functions in a program with multiple procedures. The problem is fundamental in inter-procedural analysis where properties of functions are first obtained via the generation of function contracts and then the generated contracts are used as building blocks to analyze the whole program. Typical objectives in function contract generation include pre-/post-conditions and assigns information (that specifies the modification information over program variables and memory segments during function execution). In programs with array manipulations, a crucial point in function contract generation is the treatment of array segments that imposes challenges in inferring invariants and assigns information over such segments. To address this challenge, we propose a novel symbolic execution framework that carries invariants and assigns information over contiguous segments of arrays. We implement our framework as a prototype within LLVM, and further integrate our prototype with the ACSL assertion format and the Frama-C software verification platform. Experimental evaluation over a variety of benchmarks from the literature and functions from realistic libraries shows that our framework is capable of handling array manipulating functions that indeed involve the carry of array information and are beyond existing approaches.

Plug-and-Play diffusion prior (PnPDP) frameworks have emerged as a powerful paradigm for solving imaging inverse problems by treating pretrained generative models as modular priors. However, we identify a critical flaw in prevailing PnP solvers (e.g., based on HQS or Proximal Gradient): they function as memoryless operators, updating estimates solely based on instantaneous gradients. This lack of historical tracking inevitably leads to non-vanishing steady-state bias, where the reconstruction fails to strictly satisfy physical measurements under heavy corruption. To resolve this, we propose Dual-Coupled PnP Diffusion, which restores the classical dual variable to provide integral feedback, theoretically guaranteeing asymptotic convergence to the exact data manifold. However, this rigorous geometric coupling introduces a secondary challenge: the accumulated dual residuals exhibit spectrally colored, structured artifacts that violate the Additive White Gaussian Noise (AWGN) assumption of diffusion priors, causing severe hallucinations. To bridge this gap, we introduce Spectral Homogenization (SH), a frequency-domain adaptation mechanism that modulates these structured residuals into statistically compliant pseudo-AWGN inputs. This effectively aligns the solver's rigorous optimization trajectory with the denoiser's valid statistical manifold. Extensive experiments on CT and MRI reconstruction demonstrate that our approach resolves the bias-hallucination trade-off, achieving state-of-the-art fidelity with significantly accelerated convergence.

Colonoscopy video generation delivers dynamic, information-rich data critical for diagnosing intestinal diseases, particularly in data-scarce scenarios. High-quality video generation demands temporal consistency and precise control over clinical attributes, but faces challenges from irregular intestinal structures, diverse disease representations, and various imaging modalities. To this end, we propose ColoDiff, a diffusion-based framework that generates dynamic-consistent and content-aware colonoscopy videos, aiming to alleviate data shortage and assist clinical analysis. At the inter-frame level, our TimeStream module decouples temporal dependency from video sequences through a cross-frame tokenization mechanism, enabling intricate dynamic modeling despite irregular intestinal structures. At the intra-frame level, our Content-Aware module incorporates noise-injected embeddings and learnable prototypes to realize precise control over clinical attributes, breaking through the coarse guidance of diffusion models. Additionally, ColoDiff employs a non-Markovian sampling strategy that cuts steps by over 90% for real-time generation. ColoDiff is evaluated across three public datasets and one hospital database, based on both generation metrics and downstream tasks including disease diagnosis, modality discrimination, bowel preparation scoring, and lesion segmentation. Extensive experiments show ColoDiff generates videos with smooth transitions and rich dynamics. ColoDiff presents an effort in controllable colonoscopy video generation, revealing the potential of synthetic videos in complementing authentic representation and mitigating data scarcity in clinical settings.

Modern machine learning models are deployed in diverse, non-stationary environments where they must continually adapt to new tasks and evolving knowledge. Continual fine-tuning and in-context learning are costly and brittle, whereas neural memory methods promise lightweight updates with minimal forgetting. However, existing neural memory models typically assume a single fixed objective and homogeneous information streams, leaving users with no control over what the model remembers or ignores over time. To address this challenge, we propose a generalized neural memory system that performs flexible updates based on learning instructions specified in natural language. Our approach enables adaptive agents to learn selectively from heterogeneous information sources, supporting settings, such as healthcare and customer service, where fixed-objective memory updates are insufficient.

Reducing the hardware footprint of large language models (LLMs) during decoding is critical for efficient long-sequence generation. A key bottleneck is the key-value (KV) cache, whose size scales with sequence length and easily dominates the memory footprint of the model. Previous work proposed quantization methods that are focused on compressing the KV cache while maintaining its information. We introduce InnerQ, a hardware-aware KV-cache quantization scheme that lowers decode latency without sacrificing accuracy. InnerQ applies group-wise quantization while grouping the cache matrices over their inner dimension. Unlike previous work that group over the outer dimension, InnerQ aligns dequantization with the vector-matrix multiplication and enables scale factor reuse across GPU compute units. This reduces memory accesses and accelerates dequantization, yielding up to $22\%$ speedup over previous work and up to $88\%$ over half-precision vector-matrix multiplication. To preserve fidelity under aggressive compression, InnerQ incorporates (i) hybrid quantization, selecting symmetric or asymmetric quantization per group based on local statistics; (ii) high-precision windows for both the most recent tokens and the attention sink tokens to mitigate outlier leakage; and (iii) per-channel normalization of the key cache, computed once during prefill and folded into the query to avoid runtime overhead. Our evaluation experiments on Llama models shows that InnerQ maintains a few-shot GSM8K performance comparable to non-quantized KV caches and surpasses prior KV cache quantization methods.

Large language models (LLMs) are increasingly applied in scientific research, offering new capabilities for knowledge discovery and reasoning. In single-cell biology, however, evaluation practices for both general and specialized LLMs remain inadequate: existing benchmarks are fragmented across tasks, adopt formats such as multiple-choice classification that diverge from real-world usage, and rely on metrics lacking interpretability and biological grounding. We present SC-ARENA, a natural language evaluation framework tailored to single-cell foundation models. SC-ARENA formalizes a virtual cell abstraction that unifies evaluation targets by representing both intrinsic attributes and gene-level interactions. Within this paradigm, we define five natural language tasks (cell type annotation, captioning, generation, perturbation prediction, and scientific QA) that probe core reasoning capabilities in cellular biology. To overcome the limitations of brittle string-matching metrics, we introduce knowledge-augmented evaluation, which incorporates external ontologies, marker databases, and scientific literature to support biologically faithful and interpretable judgments. Experiments and analysis across both general-purpose and domain-specialized LLMs demonstrate that (i) under the Virtual Cell unified evaluation paradigm, current models achieve uneven performance on biologically complex tasks, particularly those demanding mechanistic or causal understanding; and (ii) our knowledge-augmented evaluation framework ensures biological correctness, provides interpretable, evidence-grounded rationales, and achieves high discriminative capacity, overcoming the brittleness and opacity of conventional metrics. SC-Arena thus provides a unified and interpretable framework for assessing LLMs in single-cell biology, pointing toward the development of biology-aligned, generalizable foundation models.

Transformer-based large language models exhibit in-context learning, enabling adaptation to downstream tasks via few-shot prompting with demonstrations. In practice, such models are often fine-tuned to improve zero-shot performance on downstream tasks, allowing them to solve tasks without examples and thereby reducing inference costs. However, fine-tuning can degrade in-context learning, limiting the performance of fine-tuned models on tasks not seen during fine-tuning. Using linear attention models, we provide a theoretical analysis that characterizes how fine-tuning objectives modify attention parameters and identifies conditions under which this leads to degraded few-shot performance. We show that fine-tuning all attention parameters can harm in-context learning, whereas restricting updates to the value matrix improves zero-shot performance while preserving in-context learning. We further show that incorporating an auxiliary few-shot loss enhances in-context learning primarily on the target task, at the expense of degraded in-context learning ability on tasks not seen during fine-tuning. We empirically validate our theoretical results.

Autonomous agents based on Large Language Models (LLMs) have evolved from reactive assistants to systems capable of planning, executing actions via tools, and iterating over environment observations. However, they remain vulnerable to structural limitations: lack of native state, context degradation over long horizons, and the gap between probabilistic generation and deterministic execution requirements. This paper presents the ESAA (Event Sourcing for Autonomous Agents) architecture, which separates the agent's cognitive intention from the project's state mutation, inspired by the Event Sourcing pattern. In ESAA, agents emit only structured intentions in validated JSON (agent.result or issue.report); a deterministic orchestrator validates, persists events in an append-only log (activity.jsonl), applies file-writing effects, and projects a verifiable materialized view (roadmap.json). The proposal incorporates boundary contracts (AGENT_CONTRACT.yaml), metaprompting profiles (PARCER), and replay verification with hashing (esaa verify), ensuring the immutability of completed tasks and forensic traceability. Two case studies validate the architecture: (i) a landing page project (9 tasks, 49 events, single-agent composition) and (ii) a clinical dashboard system (50 tasks, 86 events, 4 concurrent agents across 8 phases), both concluding with run.status=success and verify_status=ok. The multi-agent case study demonstrates real concurrent orchestration with heterogeneous LLMs (Claude Sonnet 4.6, Codex GPT-5, Antigravity/Gemini 3 Pro, and Claude Opus 4.6), providing empirical evidence of the architecture's scalability beyond single-agent scenarios.

Compressing neural networks by quantizing model parameters offers useful trade-off between performance and efficiency. Methods like quantization-aware training and post-training quantization strive to maintain the downstream performance of compressed models compared to the full precision models. However, these techniques do not explicitly consider the impact on algorithmic fairness. In this work, we study fairness-aware mixed-precision quantization schemes for medical image classification under explicit bit budgets. We introduce FairQuant, a framework that combines group-aware importance analysis, budgeted mixed-precision allocation, and a learnable Bit-Aware Quantization (BAQ) mode that jointly optimizes weights and per-unit bit allocations under bitrate and fairness regularization. We evaluate the method on Fitzpatrick17k and ISIC2019 across ResNet18/50, DeiT-Tiny, and TinyViT. Results show that FairQuant configurations with average precision near 4-6 bits recover much of the Uniform 8-bit accuracy while improving worst-group performance relative to Uniform 4- and 8-bit baselines, with comparable fairness metrics under shared budgets.

We propose an efficient retraining strategy for a parameterized Reduced Order Model (ROM) that attains accuracy comparable to full retraining while requiring only a fraction of the computational time and relying solely on sparse observations of the full system. The architecture employs an encode-process-decode structure: a Variational Autoencoder (VAE) to perform dimensionality reduction, and a transformer network to evolve the latent states and model the dynamics. The ROM is parameterized by an external control variable, the Reynolds number in the Navier-Stokes setting, with the transformer exploiting attention mechanisms to capture both temporal dependencies and parameter effects. The probabilistic VAE enables stochastic sampling of trajectory ensembles, providing predictive means and uncertainty quantification through the first two moments. After initial training on a limited set of dynamical regimes, the model is adapted to out-of-sample parameter regions using only sparse data. Its probabilistic formulation naturally supports ensemble generation, which we employ within an ensemble Kalman filtering framework to assimilate data and reconstruct full-state trajectories from minimal observations. We further show that, for the dynamical system considered, the dominant source of error in out-of-sample forecasts stems from distortions of the latent manifold rather than changes in the latent dynamics. Consequently, retraining can be limited to the autoencoder, allowing for a lightweight, computationally efficient, real-time adaptation procedure with very sparse fine-tuning data.

We present MTRAG-UN, a benchmark for exploring open challenges in multi-turn retrieval augmented generation, a popular use of large language models. We release a benchmark of 666 tasks containing over 2,800 conversation turns across 6 domains with accompanying corpora. Our experiments show that retrieval and generation models continue to struggle on conversations with UNanswerable, UNderspecified, and NONstandalone questions and UNclear responses. Our benchmark is available at https://github.com/IBM/mt-rag-benchmark

While Deep Learning has demonstrated impressive results in applications on various data types, it continues to lag behind tree-based methods when applied to tabular data, often referred to as the last "unconquered castle" for neural networks. We hypothesize that a significant advantage of tree-based methods lies in their intrinsic capability to model and exploit non-linear interactions induced by features with categorical characteristics. In contrast, neural-based methods exhibit biases toward uniform numerical processing of features and smooth solutions, making it challenging for them to effectively leverage such patterns. We address this performance gap by using statistical-based feature processing techniques to identify features that are strongly correlated with the target once discretized. We further mitigate the bias of deep models for overly-smooth solutions, a bias that does not align with the inherent properties of the data, using Learned Fourier. We show that our proposed feature preprocessing significantly boosts the performance of deep learning models and enables them to achieve a performance that closely matches or surpasses XGBoost on a comprehensive tabular data benchmark.

Protein sequences are abundant in repeating segments, both as exact copies and as approximate segments with mutations. These repeats are important for protein structure and function, motivating decades of algorithmic work on repeat identification. Recent work has shown that protein language models (PLMs) identify repeats, by examining their behavior in masked-token prediction. To elucidate their internal mechanisms, we investigate how PLMs detect both exact and approximate repeats. We find that the mechanism for approximate repeats functionally subsumes that of exact repeats. We then characterize this mechanism, revealing two main stages: PLMs first build feature representations using both general positional attention heads and biologically specialized components, such as neurons that encode amino-acid similarity. Then, induction heads attend to aligned tokens across repeated segments, promoting the correct answer. Our results reveal how PLMs solve this biological task by combining language-based pattern matching with specialized biological knowledge, thereby establishing a basis for studying more complex evolutionary processes in PLMs.

Capturing 4D spatiotemporal surroundings is crucial for the safe and reliable operation of robots in dynamic environments. However, most existing methods address only one side of the problem: they either provide coarse geometric tracking via bounding boxes, or detailed 3D structures like voxel-based occupancy that lack explicit temporal association. In this work, we present Latent Gaussian Splatting for 4D Panoptic Occupancy Tracking (LaGS) that advances spatiotemporal scene understanding in a holistic direction. Our approach incorporates camera-based end-to-end tracking with mask-based multi-view panoptic occupancy prediction, and addresses the key challenge of efficiently aggregating multi-view information into 3D voxel grids via a novel latent Gaussian splatting approach. Specifically, we first fuse observations into 3D Gaussians that serve as a sparse point-centric latent representation of the 3D scene, and then splat the aggregated features onto a 3D voxel grid that is decoded by a mask-based segmentation head. We evaluate LaGS on the Occ3D nuScenes and Waymo datasets, achieving state-of-the-art performance for 4D panoptic occupancy tracking. We make our code available at https://lags.cs.uni-freiburg.de/.

In open Federated Learning (FL) environments where no central authority exists, ensuring collaboration fairness relies on decentralized reward settlement, yet the prohibitive cost of permissionless blockchains directly clashes with the high-frequency, iterative nature of model training. Existing solutions either compromise decentralization or suffer from scalability bottlenecks due to linear on-chain costs. To address this, we present SettleFL, a trustless and scalable reward settlement protocol designed to minimize total economic friction by offering a family of two interoperable protocols. Leveraging a shared domain-specific circuit architecture, SettleFL offers two interoperable strategies: (1) a Commit-and-Challenge variant that minimizes on-chain costs via optimistic execution and dispute-driven arbitration, and (2) a Commit-with-Proof variant that guarantees instant finality through per-round validity proofs. This design allows the protocol to flexibly adapt to varying latency and cost constraints while enforcing rational robustness without trusted coordination. We conduct extensive experiments combining real FL workloads and controlled simulations. Results show that SettleFL remains practical when scaling to 800 participants, achieving substantially lower gas cost.

Foundation models must handle multiple generative processes, yet mechanistic interpretability largely studies capabilities in isolation; it remains unclear how a single transformer organizes multiple, potentially conflicting "world models". Previous experiments on Othello playing neural-networks test world-model learning but focus on a single game with a single set of rules. We introduce MetaOthello, a controlled suite of Othello variants with shared syntax but different rules or tokenizations, and train small GPTs on mixed-variant data to study how multiple world models are organized in a shared representation space. We find that transformers trained on mixed-game data do not partition their capacity into isolated sub-models; instead, they converge on a mostly shared board-state representation that transfers causally across variants. Linear probes trained on one variant can intervene on another's internal state with effectiveness approaching that of matched probes. For isomorphic games with token remapping, representations are equivalent up to a single orthogonal rotation that generalizes across layers. When rules partially overlap, early layers maintain game-agnostic representations while a middle layer identifies game identity, and later layers specialize. MetaOthello offers a path toward understanding not just whether transformers learn world models, but how they organize many at once.

Large language models are beginning to show steganographic capabilities. Such capabilities could allow misaligned models to evade oversight mechanisms. Yet principled methods to detect and quantify such behaviours are lacking. Classical definitions of steganography, and detection methods based on them, require a known reference distribution of non-steganographic signals. For the case of steganographic reasoning in LLMs, knowing such a reference distribution is not feasible; this renders these approaches inapplicable. We propose an alternative, \textbf{decision-theoretic view of steganography}. Our central insight is that steganography creates an asymmetry in usable information between agents who can and cannot decode the hidden content (present within a steganographic signal), and this otherwise latent asymmetry can be inferred from the agents' observable actions. To formalise this perspective, we introduce generalised $\mathcal{V}$-information: a utilitarian framework for measuring the amount of usable information within some input. We use this to define the \textbf{steganographic gap} -- a measure that quantifies steganography by comparing the downstream utility of the steganographic signal to agents that can and cannot decode the hidden content. We empirically validate our formalism, and show that it can be used to detect, quantify, and mitigate steganographic reasoning in LLMs.

Time series reasoning demands both the perception of complex dynamics and logical depth. However, existing LLM-based approaches exhibit two limitations: they often treat time series merely as text or images, failing to capture the patterns like trends and seasonalities needed to answer specific questions; and when trained on a mix of simple and complex tasks, simpler objectives often dominate the learning process, hindering the development of deep reasoning capabilities. To address these limitations, we propose the Pattern-Aware Alignment and Balanced Reasoning model (PATRA), introducing a pattern-aware mechanism that extracts trend and seasonality patterns from time series to achieve deep alignment. Furthermore, we design a task-aware balanced reward to harmonize learning across tasks of varying difficulty, incentivizing the generation of coherent Chains of Thought. Extensive experiments show that PATRA outperforms strong baselines across diverse Time Series Question Answering (TSQA) tasks, demonstrating superior cross-modal understanding and reasoning capability.

Temporal Web analytics increasingly relies on large-scale, longitudinal data to understand how users, content, and systems evolve over time. A rapidly growing frontier is the \emph{Temporal Web3}: decentralized platforms whose behavior is recorded as immutable, time-stamped event streams. Despite the richness of this data, the field lacks shared, reproducible benchmarks that capture real-world temporal dynamics, specifically censoring and non-stationarity, across extended horizons. This absence slows methodological progress and limits the transfer of techniques between Web3 and broader Web domains. In this paper, we present the \textit{FinSurvival Challenge 2025} as a case study in benchmarking \emph{temporal Web3 intelligence}. Using 21.8 million transaction records from the Aave v3 protocol, the challenge operationalized 16 survival prediction tasks to model user behavior transitions.We detail the benchmark design and the winning solutions, highlighting how domain-aware temporal feature construction significantly outperformed generic modeling approaches. Furthermore, we distill lessons for next-generation temporal benchmarks, arguing that Web3 systems provide a high-fidelity sandbox for studying temporal challenges, such as churn, risk, and evolution that are fundamental to the wider Web.

Large Multimodal Models (LMMs) that process 3D data typically rely on heavy, pre-trained visual encoders to extract geometric features. While recent 2D LMMs have begun to eliminate such encoders for efficiency and scalability, extending this paradigm to 3D remains challenging due to the unordered and large-scale nature of point clouds. This leaves a critical unanswered question: How can we design an LMM that tokenizes unordered 3D data effectively and efficiently without a cumbersome encoder? We propose Fase3D, the first efficient encoder-free Fourier-based 3D scene LMM. Fase3D tackles the challenges of scalability and permutation invariance with a novel tokenizer that combines point cloud serialization and the Fast Fourier Transform (FFT) to approximate self-attention. This design enables an effective and computationally minimal architecture, built upon three key innovations: First, we represent large scenes compactly via structured superpoints. Second, our space-filling curve serialization followed by an FFT enables efficient global context modeling and graph-based token merging. Lastly, our Fourier-augmented LoRA adapters inject global frequency-aware interactions into the LLMs at a negligible cost. Fase3D achieves performance comparable to encoder-based 3D LMMs while being significantly more efficient in computation and parameters. Project website: https://tev-fbk.github.io/Fase3D.

The construction of World Models capable of learning, simulating, and reasoning about objective physical laws constitutes a foundational challenge in the pursuit of Artificial General Intelligence. Recent advancements represented by video generation models like Sora have demonstrated the potential of data-driven scaling laws to approximate physical dynamics, while the emerging Unified Multimodal Model (UMM) offers a promising architectural paradigm for integrating perception, language, and reasoning. Despite these advances, the field still lacks a principled theoretical framework that defines the essential properties requisite for a General World Model. In this paper, we propose that a World Model must be grounded in the Trinity of Consistency: Modal Consistency as the semantic interface, Spatial Consistency as the geometric basis, and Temporal Consistency as the causal engine. Through this tripartite lens, we systematically review the evolution of multimodal learning, revealing a trajectory from loosely coupled specialized modules toward unified architectures that enable the synergistic emergence of internal world simulators. To complement this conceptual framework, we introduce CoW-Bench, a benchmark centered on multi-frame reasoning and generation scenarios. CoW-Bench evaluates both video generation models and UMMs under a unified evaluation protocol. Our work establishes a principled pathway toward general world models, clarifying both the limitations of current systems and the architectural requirements for future progress.

Generalized planning studies the construction of solution strategies that generalize across families of planning problems sharing a common domain model, formally defined by a transition function $γ: S \times A \rightarrow S$. Classical approaches achieve such generalization through symbolic abstractions and explicit reasoning over $γ$. In contrast, recent Transformer-based planners, such as PlanGPT and Plansformer, largely cast generalized planning as direct action-sequence prediction, bypassing explicit transition modeling. While effective on in-distribution instances, these approaches typically require large datasets and model sizes, and often suffer from state drift in long-horizon settings due to the absence of explicit world-state evolution. In this work, we formulate generalized planning as a transition-model learning problem, in which a neural model explicitly approximates the successor-state function $\hatγ \approx γ$ and generates plans by rolling out symbolic state trajectories. Instead of predicting actions directly, the model autoregressively predicts intermediate world states, thereby learning the domain dynamics as an implicit world model. To study size-invariant generalization and sample efficiency, we systematically evaluate multiple state representations and neural architectures, including relational graph encodings. Our results show that learning explicit transition models yields higher out-of-distribution satisficing-plan success than direct action-sequence prediction in multiple domains, while achieving these gains with significantly fewer training instances and smaller models. This is an extended version of a short paper accepted at ICAPS 2026 under the same title.

Near-surface atmospheric conditions can differ sharply over tens to hundreds of meters due to land cover and topography, yet this variability is absent from current weather analyses and forecasts. It is unclear whether such meter-scale variability reflects irreducibly chaotic dynamics or contains a component predictable from surface characteristics and large-scale atmospheric forcing. Here we show that a substantial, physically coherent component of meter-scale near-surface weather is statistically recoverable from existing observations. By conditioning coarse atmospheric state on sparse surface station measurements and high-resolution Earth observation data, we infer spatially continuous fields of near-surface wind, temperature, and humidity at 10 m resolution across the contiguous United States. Relative to ERA5, the inferred fields reduce wind error by 29% and temperature and dewpoint error by 6%, while explaining substantially more spatial variance at fixed time steps. They also exhibit physically interpretable structure, including urban heat islands, evapotranspiration-driven humidity contrasts, and wind speed differences across land cover types. Our findings expand the frontier of weather modeling by demonstrating a computationally feasible approach to continental-scale meter-resolution inference. More broadly, they illustrate how conditioning coarse dynamical models on static fine-scale features can reveal previously unresolved components of the Earth system.

Predicting diffusion coefficients in mixtures is crucial for many applications, as experimental data remain scarce, and machine learning (ML) offers promising alternatives to established semi-empirical models. Among ML models, matrix completion methods (MCMs) have proven effective in predicting thermophysical properties, including diffusion coefficients in binary mixtures. However, MCMs are restricted to single-temperature predictions, and their accuracy depends strongly on the availability of high-quality experimental data for each temperature of interest. In this work, we address this challenge by presenting a hybrid tensor completion method (TCM) for predicting temperature-dependent diffusion coefficients at infinite dilution in binary mixtures. The TCM employs a Tucker decomposition and is jointly trained on experimental data for diffusion coefficients at infinite dilution in binary systems at 298 K, 313 K, and 333 K. Predictions from the semi-empirical SEGWE model serve as prior knowledge within a Bayesian training framework. The TCM then extrapolates linearly to any temperature between 268 K and 378 K, achieving markedly improved prediction accuracy compared to established models across all studied temperatures. To further enhance predictive performance, the experimental database was expanded using active learning (AL) strategies for targeted acquisition of new diffusion data by pulsed-field gradient (PFG) NMR measurements. Diffusion coefficients at infinite dilution in 19 solute + solvent systems were measured at 298 K, 313 K, and 333 K. Incorporating these results yields a substantial improvement in the TCM's predictive accuracy. These findings highlight the potential of combining data-efficient ML methods with adaptive experimentation to advance predictive modeling of transport properties.

Multimodal LLMs can process speech and images, but they cannot hear a speaker's voice or see an object's texture. We show this is not a failure of encoding: speaker identity, emotion, and visual attributes survive through every LLM layer (3--55$\times$ above chance in linear probes), yet removing 64--71% of modality-specific variance improves decoder loss. The decoder has no learned use for these directions; their presence is noise. We formalize this as a mismatched decoder problem: a decoder trained on text can only extract information along text-aligned directions. Accessible information is bounded by the Generalized Mutual Information (GMI), with degradation scaling with distributional distance and decoder sensitivity. The bound is a property of the decoder's scoring rule, not of any particular architecture; it applies whether non-text inputs arrive through a learned projection, a discrete codebook, or no explicit adapter at all. We validate this across five models spanning speech and vision. A controlled experiment (two Prismatic VLMs differing only in encoder text-alignment) confirms the bottleneck is the decoder's scoring rule, not the encoder or projection. A LoRA intervention demonstrates the fix: training with an emotion objective improves emotion accessibility ($+$7.5%) without affecting other attributes, confirming that the training objective determines what becomes accessible.

Real-world dynamic graphs are often directed, with source and destination nodes exhibiting asymmetrical behavioral patterns and temporal dynamics. However, existing dynamic graph architectures largely rely on shared parameters for processing source and destination nodes, with limited or no systematic role-aware modeling. We propose DyGnROLE (Dynamic Graph Node-Role-Oriented Latent Encoding), a transformer-based architecture that explicitly disentangles source and destination representations. By using separate embedding vocabularies and role-semantic positional encodings, the model captures the distinct structural and temporal contexts unique to each role. Critical to the effectiveness of these specialized embeddings in low-label regimes is a self-supervised pretraining objective we introduce: Temporal Contrastive Link Prediction (TCLP). The pretraining uses the full unlabeled interaction history to encode informative structural biases, enabling the model to learn role-specific representations without requiring annotated data. Evaluation on future edge classification demonstrates that DyGnROLE substantially outperforms a diverse set of state-of-the-art baselines, establishing role-aware modeling as an effective strategy for dynamic graph learning.

Multi-behavior sequential recommendation (MBSR) aims to learn the dynamic and heterogeneous interactions of users' multi-behavior sequences, so as to capture user preferences under target behavior for the next interacted item prediction. Unlike previous methods that adopt unidirectional modeling by mapping auxiliary behaviors to target behavior, recent concerns are shifting from behavior-fixed to behavior-specific recommendation. However, these methods still ignore the user's latent preference that underlying decision-making, leading to suboptimal solutions. Meanwhile, due to the asymmetric deterministic between items and behaviors, discriminative paradigm based on preference scoring is unsuitable to capture the uncertainty from low-entropy behaviors to high-entropy items, failing to provide efficient and diverse recommendation. To address these challenges, we propose \textbf{FatsMB}, a framework based diffusion model that guides preference generation \textit{\textbf{F}rom Behavior-\textbf{A}gnostic \textbf{T}o Behavior-\textbf{S}pecific} in latent spaces, enabling diverse and accurate \textit{\textbf{M}ulti-\textbf{B}ehavior Sequential Recommendation}. Specifically, we design a Multi-Behavior AutoEncoder (MBAE) to construct a unified user latent preference space, facilitating interaction and collaboration across Behaviors, within Behavior-aware RoPE (BaRoPE) employed for multiple information fusion. Subsequently, we conduct target behavior-specific preference transfer in the latent space, enriching with informative priors. A Multi-Condition Guided Layer Normalization (MCGLN) is introduced for the denoising. Extensive experiments on real-world datasets demonstrate the effectiveness of our model.

Generalization bounds for deep learning models are typically vacuous, not computable or restricted to specific model classes. In this paper, we tackle these issues by providing new disagreement-based certificates for the gap between the true risk of any two predictors. We then bound the true risk of the predictor of interest via a surrogate model that enjoys tight generalization guarantees, and evaluating our disagreement bound on an unlabeled dataset. We empirically demonstrate the tightness of the obtained certificates and showcase the versatility of the approach by training surrogate models leveraging three different frameworks: sample compression, model compression and PAC-Bayes theory. Importantly, such guarantees are achieved without modifying the target model, nor adapting the training procedure to the generalization framework.

In the attention economy, sensational content exposes consumers to excessive emotional stimulation, hindering calm decision-making. This study proposes Multi-Agent LLM-based Emotional deToxification (MALLET), a multi-agent information sanitization system consisting of four agents: Emotion Analysis, Emotion Adjustment, Balance Monitoring, and Personal Guide. The Emotion Analysis Agent quantifies stimulus intensity using a 6-emotion BERT classifier, and the Emotion Adjustment Agent rewrites texts into two presentation modes, BALANCED (neutralized text) and COOL (neutralized text + supplementary text), using an LLM. The Balance Monitoring Agent aggregates weekly information consumption patterns and generates personalized advice, while the Personal Guide Agent recommends a presentation mode according to consumer sensitivity. Experiments on 800 AG News articles demonstrated significant stimulus score reduction (up to 19.3%) and improved emotion balance while maintaining semantic preservation. Near-zero correlation between stimulus reduction and semantic preservation confirmed that the two are independently controllable. Category-level analysis revealed substantial reduction (17.8-33.8%) in Sports, Business, and Sci/Tech, whereas the effect was limited in the World category, where facts themselves are inherently high-stimulus. The proposed system provides a framework for supporting calm information reception of consumers without restricting access to the original text.

Each year, software vulnerabilities are discovered, which pose significant risks of exploitation and system compromise. We present a convolutional neural network model that can successfully identify bugs in C code. We trained our model using two complementary datasets: a machine-labeled dataset created by Draper Labs using three static analyzers and the NIST SATE Juliet human-labeled dataset designed for testing static analyzers. In contrast with the work of Russell et al. on these datasets, we focus on C programs, enabling us to specialize and optimize our detection techniques for this language. After removing duplicates from the dataset, we tokenize the input into 91 token categories. The category values are converted to a binary vector to save memory. Our first convolution layer is chosen so that the entire encoding of the token is presented to the filter. We use two convolution and pooling layers followed by two fully connected layers to classify programs into either a common weakness enumeration category or as ``clean.'' We obtain higher recall than prior work by Russell et al. on this dataset when requiring high precision. We also demonstrate on a custom Linux kernel dataset that we are able to find real vulnerabilities in complex code with a low false-positive rate.

Adversarial transferability refers to the capacity of adversarial examples generated on the surrogate model to deceive alternate, unexposed victim models. This property eliminates the need for direct access to the victim model during an attack, thereby raising considerable security concerns in practical applications and attracting substantial research attention recently. In this work, we discern a lack of a standardized framework and criteria for evaluating transfer-based attacks, leading to potentially biased assessments of existing approaches. To rectify this gap, we have conducted an exhaustive review of hundreds of related works, organizing various transfer-based attacks into six distinct categories. Subsequently, we propose a comprehensive framework designed to serve as a benchmark for evaluating these attacks. In addition, we delineate common strategies that enhance adversarial transferability and highlight prevalent issues that could lead to unfair comparisons. Finally, we provide a brief review of transfer-based attacks beyond image classification.

We consider the problem of contextual online RLHF with general preferences, where the goal is to identify the Nash Equilibrium. We adopt the Generalized Bilinear Preference Model (GBPM) to capture potentially intransitive preferences via low-rank, skew-symmetric matrices. We investigate general preference learning with any strongly convex regularizer (where $η^{-1}$ is the regularization strength), generalizing beyond prior works limited to reverse KL-regularization. Central to our analysis is proving that the dual gap of the greedy policy is bounded by the square of the estimation error - a result derived solely from strong convexity and the skew-symmetricity of GBPM.Building on this insight and a feature diversity assumption, we establish two regret bounds via two simple algorithms: (1) Greedy Sampling achieves polylogarithmic, $e^{O(η)}$-free regret $\tilde{O}(ηd^4 (\log T)^2)$. (2) Explore-Then-Commit achieves $\mathrm{poly}(d)$-free regret $\tilde{O}(\sqrt{ηr T})$ by exploiting the low-rank structure; this is the first statistically efficient guarantee for online RLHF in high-dimensions.

Neural operators have emerged as promising surrogate models for solving partial differential equations (PDEs), but struggle to generalise beyond training distributions and are often constrained to a fixed temporal discretisation. This work introduces a physics-informed training framework that addresses these limitations by decomposing PDEs using operator splitting methods, training separate neural operators to learn individual non-linear physical operators while approximating linear operators with fixed finite-difference convolutions. This modular mixture-of-experts architecture enables generalisation to novel physical regimes by explicitly encoding the underlying operator structure. We formulate the modelling task as a neural ordinary differential equation (ODE) where these learned operators constitute the right-hand side, enabling continuous-in-time predictions through standard ODE solvers and implicitly enforcing PDE constraints. Demonstrated on incompressible and compressible Navier-Stokes equations, our approach achieves better convergence and superior performance when generalising to unseen physics. The method remains parameter-efficient, enabling temporal extrapolation beyond training horizons, and provides interpretable components whose behaviour can be verified against known physics.

Activations have become the primary memory bottleneck in large-batch LLM training. However, existing compression methods fail to exploit the spectral structure of activations, resulting in slow convergence or limited compression. To address this, we bridge the relationship between the algorithm's fast convergence and the requirements for subspace projection, and show that an effective compression should yield an unbiased estimate of the original activation with low variance. We propose Principal-Random Subspace for LLM Activation Compression (PRAC), which novelly decomposes activations into two components: a principal subspace captured via SVD to retain dominant information, and a random subspace sampled from the orthogonal complement to approximate the tail. By introducing a precise scaling factor, we prove that PRAC yields an unbiased gradient estimator with minimum variance under certain conditions. Extensive experiments on pre-training and fine-tuning tasks demonstrate that PRAC achieves up to 36% total memory reduction with negligible performance degradation and minimal computational cost.

Near-future infrastructure systems may be controlled by autonomous AI agents that repeatedly request access to limited resources such as energy, bandwidth, or computing power. We study a simplified version of this setting using a framework where N AI-agents independently decide at each round whether to request one unit from a system with fixed capacity C. An AI version of "Lord of the Flies" arises in which controlling tribes emerge with their own collective character and identity. The LLM agents do not reduce overload or improve resource use, and often perform worse than if they were flipping coins to make decisions. Three main tribal types emerge: Aggressive (27.3%), Conservative (24.7%), and Opportunistic (48.1%). The more capable AI-agents actually increase the rate of systemic failure. Overall, our findings show that smarter AI-agents can behave dumber as a result of forming tribes.

The Capacitated Vehicle Routing Problem (CVRP), a fundamental combinatorial optimization challenge, focuses on optimizing fleet operations under vehicle capacity constraints. While extensively studied in operational research, the NP-hard nature of CVRP continues to pose significant computational challenges, particularly for large-scale instances. This study presents AILS-AHD (Adaptive Iterated Local Search with Automatic Heuristic Design), a novel approach that leverages Large Language Models (LLMs) to revolutionize CVRP solving. Our methodology integrates an evolutionary search framework with LLMs to dynamically generate and optimize ruin heuristics within the AILS method. Additionally, we introduce an LLM-based acceleration mechanism to enhance computational efficiency. Comprehensive experimental evaluations against state-of-the-art solvers, including AILS-II and HGS, demonstrate the superior performance of AILS-AHD across both moderate and large-scale instances. Notably, our approach establishes new best-known solutions for 8 out of 10 instances in the CVRPLib large-scale benchmark, underscoring the potential of LLM-driven heuristic design in advancing the field of vehicle routing optimization.

The Bayesian update step poses significant computational challenges in high-dimensional nonlinear estimation. While log-homotopy particle flow filters offer an alternative to stochastic sampling, existing formulations usually yield stiff differential equations. Conversely, existing deep learning approximations typically treat the update as a black-box task or rely on asymptotic relaxation, neglecting the exact geometric structure of the finite-horizon probability transport. In this work, we propose a physics-informed neural particle flow, which is an amortized inference framework. To construct the flow, we couple the log-homotopy trajectory of the prior to posterior density function with the continuity equation describing the density evolution. This derivation yields a governing partial differential equation (PDE), referred to as the master PDE. By embedding this PDE as a physical constraint into the loss function, we train a neural network to approximate the transport velocity field. This approach enables purely unsupervised training, eliminating the need for ground-truth posterior samples. We demonstrate that the neural parameterization acts as an implicit regularizer, mitigating the numerical stiffness inherent to analytic flows and reducing online computational complexity. Experimental validation on multimodal benchmarks and a challenging nonlinear scenario confirms better mode coverage and robustness compared to state-of-the-art baselines.

Quantum cloud platforms have become the most widely adopted and mainstream approach for accessing quantum computing resources, due to the scarcity and operational complexity of quantum hardware. In this service-oriented paradigm, quantum circuits, which constitute high-value intellectual property, are exposed to risks of unauthorized access, reuse, and misuse. Digital watermarking has been explored as a promising mechanism for protecting quantum circuits by embedding ownership information for tracing and verification. However, driven by recent advances in generative artificial intelligence, the paradigm of quantum circuit design is shifting from individually and manually constructed circuits to automated synthesis based on quantum circuit generative models (QCGMs). In such generative settings, protecting only individual output circuits is insufficient, and existing post hoc, circuit-centric watermarking methods are not designed to integrate with the generative process, often failing to simultaneously ensure stealthiness, functional correctness, and robustness at scale. These limitations highlight the need for a new watermarking paradigm that is natively integrated with quantum circuit generative models. In this work, we present the first watermarking framework for QCGMs, which embeds ownership signals into the generation process while preserving circuit fidelity. We introduce a symmetric sampling strategy that aligns watermark encoding with the model's Gaussian prior, and a synchronization mechanism that counteracts adversarial watermark attack through latent drift correction. Empirical results confirm that our method achieves high-fidelity circuit generation and robust watermark detection across a range of perturbations, paving the way for scalable, secure copyright protection in AI-powered quantum design.

The rapid advancement of large language models (LLMs) has enabled powerful authorship inference capabilities, raising growing concerns about unintended deanonymization risks in textual data such as news articles. In this work, we introduce an LLM agent designed to evaluate and mitigate such risks through a structured, interpretable pipeline. Central to our framework is the proposed $\textit{SALA}$ (Stylometry-Assisted LLM Analysis) method, which integrates quantitative stylometric features with LLM reasoning for robust and transparent authorship attribution. Experiments on large-scale news datasets demonstrate that $\textit{SALA}$, particularly when augmented with a database module, achieves high inference accuracy in various scenarios. Finally, we propose a guided recomposition strategy that leverages the agent's reasoning trace to generate rewriting prompts, effectively reducing authorship identifiability while preserving textual meaning. Our findings highlight both the deanonymization potential of LLM agents and the importance of interpretable, proactive defenses for safeguarding author privacy.

Large language models (LLMs) have created new opportunities to enhance the efficiency of scholarly activities; however, challenges persist in the ethical deployment of AI assistance, including (1) the trustworthiness of AI-generated content, (2) preservation of academic integrity and intellectual property, and (3) protection of information privacy. In this work, we present CiteLLM, a specialized agentic platform designed to enable trustworthy reference discovery for grounding author-drafted claims and statements. The system introduces a novel interaction paradigm by embedding LLM utilities directly within the LaTeX editor environment, ensuring a seamless user experience and no data transmission outside the local system. To guarantee hallucination-free references, we employ dynamic discipline-aware routing to retrieve candidates exclusively from trusted web-based academic repositories, while leveraging LLMs solely for generating context-aware search queries, ranking candidates by relevance, and validating and explaining support through paragraph-level semantic matching and an integrated chatbot. Evaluation results demonstrate the superior performance of the proposed system in returning valid and highly usable references.

Risk-averse decision-making under uncertainty in partially observable domains is a central challenge in artificial intelligence and is essential for developing reliable autonomous agents. The formal framework for such problems is the partially observable Markov decision process (POMDP), where risk sensitivity is introduced through a risk measure applied to the value function, with Conditional Value-at-Risk (CVaR) being a particularly significant criterion. However, solving POMDPs is computationally intractable in general, and approximate methods rely on computationally expensive simulations of future agent trajectories. This work introduces a theoretical framework for accelerating CVaR value function evaluation in POMDPs with formal performance guarantees. We derive new bounds on the CVaR of a random variable X using an auxiliary random variable Y, under assumptions relating their cumulative distribution and density functions; these bounds yield interpretable concentration inequalities and converge as the distributional discrepancy vanishes. Building on this, we establish upper and lower bounds on the CVaR value function computable from a simplified belief-MDP, accommodating general simplifications of the transition dynamics. We develop estimators for these bounds within a particle-belief MDP framework with probabilistic guarantees, and employ them for acceleration via action elimination: actions whose bounds indicate suboptimality under the simplified model are safely discarded while ensuring consistency with the original POMDP. Empirical evaluation across multiple POMDP domains confirms that the bounds reliably separate safe from dangerous policies while achieving substantial computational speedups under the simplified model.

While second language (L2) learners may acquire target syntactic word order, mapping this syntax onto appropriate prosodic structures remains a persistent challenge. This study investigates the fossilization and stability of the L2 syntax-prosody interface by comparing 67 native Mandarin speakers with 67 Vietnamese learners using the BLCU-SAIT corpus. By integrating C-ToBI boundary annotation with Dependency Grammar analysis, we examined both the quantity of prosodic boundaries and their mapping to syntactic relations. Results reveal a non-linear acquisition: although high-proficiency learners (VNH) converge to the native baseline in boundary quantity at the Major Phrase level (B3), their structural mapping significantly diverges. Specifically, VNH demote the prosodic boundary at the Subject-Verb (SBV) interface (Major Phrase B3 -> Prosodic Word B1), while erroneously promoting the boundary at the Verb-Object (VOB) interface (Prosodic Word B1 -> Major Phrase B3). This strategy allows learners to maintain high long phrasal output at the expense of structural accuracy. This results in a distorted prosodic hierarchy where the native pattern is inverted.

Although Automatic Speech Recognition (ASR) in Bengali has seen significant progress, processing long-duration audio and performing robust speaker diarization remain critical research gaps. To address the severe scarcity of joint ASR and diarization resources for this language, we introduce Lipi-Ghor-882, a comprehensive 882-hour multi-speaker Bengali dataset. In this paper, detailing our submission to the DL Sprint 4.0 competition, we systematically evaluate various architectures and approaches for long-form Bengali speech. For ASR, we demonstrate that raw data scaling is ineffective; instead, targeted fine-tuning utilizing perfectly aligned annotations paired with synthetic acoustic degradation (noise and reverberation) emerges as the singular most effective approach. Conversely, for speaker diarization, we observed that global open-source state-of-the-art models (such as Diarizen) performed surprisingly poorly on this complex dataset. Extensive model retraining yielded negligible improvements; instead, strategic, heuristic post-processing of baseline model outputs proved to be the primary driver for increasing accuracy. Ultimately, this work outlines a highly optimized dual pipeline achieving a $\sim$0.019 Real-Time Factor (RTF), establishing a practical, empirically backed benchmark for low-resource, long-form speech processing.

Deep learning (DL) libraries are widely used in critical applications, where even subtle silent bugs can lead to serious consequences. While existing DL fuzzing techniques have made progress in detecting crashes, they inherently struggle to detect silent bugs due to the lack of effective test programs and corresponding oracles. Building on the observation that historical bug reports contain rich, underutilized information about silent bugs, we leverage large language models (LLMs) to perform versatile yet controlled bug transfer for silent bug fuzzing. Specifically, our approach uses LLMs to extract context-aware bug patterns from historical issues, match semantically related Application Programming Interfaces (APIs) using functionality-based embeddings, and synthesize test cases with customized oracles. This enables proactive detection of silent bugs by transferring high-risk contexts and oracle designs from known buggy APIs to functionally similar target APIs. To ensure the reliability of our context-aware bug transfer, we introduce an LLM-powered self-validation module that systematically evaluates the validity of each transferred bug instance. We implement this methodology in a tool named TransFuzz and evaluate it on three mainstream DL libraries: PyTorch, TensorFlow, and MindSpore. TransFuzz successfully discovers 79 previously unknown bugs (12 confirmed as Common Vulnerabilities and Exposures (CVEs)) in 10 bug types, demonstrating its effectiveness and generalizability in migrating DL library bug discovery capabilities.

Case Report Forms (CRFs) collect data about patients and are at the core of well-established practices to conduct research in clinical settings. With the recent progress of language technologies, there is an increasing interest in automatic CRF-filling from clinical notes, mostly based on the use of Large Language Models (LLMs). However, there is a general scarcity of annotated CRF data, both for training and testing LLMs, which limits the progress on this task. As a step in the direction of providing such data, we present a new dataset of clinical notes from an Italian Emergency Department annotated with respect to a pre-defined CRF containing 134 items to be filled. We provide an analysis of the data, define the CRF-filling task and metric for its evaluation, and report on pilot experiments where we use an open-source state-of-the-art LLM to automatically execute the task. Results of the case-study show that (i) CRF-filling from real clinical notes in Italian can be approached in a zero-shot setting; (ii) LLMs' results are affected by biases (e.g., a cautious behaviour favours "unknown" answers), which need to be corrected.

Semi-structured documents integrate diverse interleaved data elements (e.g., tables, charts, hierarchical paragraphs) arranged in various and often irregular layouts. These documents are widely observed across domains and account for a large portion of real-world data. However, existing methods struggle to support natural language question answering over these documents due to three main technical challenges: (1) The elements extracted by techniques like OCR are often fragmented and stripped of their original semantic context, making them inadequate for analysis. (2) Existing approaches lack effective representations to capture hierarchical structures within documents (e.g., associating tables with nested chapter titles) and to preserve layout-specific distinctions (e.g., differentiating sidebars from main content). (3) Answering questions often requires retrieving and aligning relevant information scattered across multiple regions or pages, such as linking a descriptive paragraph to table cells located elsewhere in the document. To address these issues, we propose MoDora, an LLM-powered system for semi-structured document analysis. First, we adopt a local-alignment aggregation strategy to convert OCR-parsed elements into layout-aware components, and conduct type-specific information extraction for components with hierarchical titles or non-text elements. Second, we design the Component-Correlation Tree (CCTree) to hierarchically organize components, explicitly modeling inter-component relations and layout distinctions through a bottom-up cascade summarization process. Finally, we propose a question-type-aware retrieval strategy that supports (1) layout-based grid partitioning for location-based retrieval and (2) LLM-guided pruning for semantic-based retrieval. Experiments show MoDora outperforms baselines by 5.97%-61.07% in accuracy. The code is at https://github.com/weAIDB/MoDora.

Electrocardiogram (ECG) analysis is crucial for diagnosing heart disease, but most self-supervised learning methods treat ECG as a generic time series, overlooking physiologic semantics and rhythm-level structure. Existing contrastive methods utilize augmentations that distort morphology, whereas generative approaches employ fixed-window segmentation, which misaligns cardiac cycles. To address these limitations, we propose RhythmBERT, a generative ECG language model that considers ECG as a language paradigm by encoding P, QRS, and T segments into symbolic tokens via autoencoder-based latent representations. These discrete tokens capture rhythm semantics, while complementary continuous embeddings retain fine-grained morphology, enabling a unified view of waveform structure and rhythm. RhythmBERT is pretrained on approximately 800,000 unlabeled ECG recordings with a masked prediction objective, allowing it to learn contextual representations in a label-efficient manner. Evaluations show that despite using only a single lead, RhythmBERT achieves comparable or superior performance to strong 12-lead baselines. This generalization extends from prevalent conditions such as atrial fibrillation to clinically challenging cases such as subtle ST-T abnormalities and myocardial infarction. Our results suggest that considering ECG as structured language offers a scalable and physiologically aligned pathway for advancing cardiac analysis.

Transformer attention is typically implemented using softmax normalization, which enforces attention weights with unit sum normalization. While effective in many settings, this constraint can limit flexibility in controlling attention magnitudes and may contribute to overly concentrated or unstable attention patterns during training. Prior work has explored modifications such as attention sinks or gating mechanisms, but these approaches provide only limited or indirect control over attention reweighting. We propose Affine-Scaled Attention, a simple extension to standard attention that introduces input-dependent scaling and a corresponding bias term applied to softmax-normalized attention weights. This design relaxes the strict normalization constraint while maintaining aggregation of value representations, allowing the model to adjust both the relative distribution and the scale of attention in a controlled manner. We empirically evaluate Affine-Scaled Attention in large-scale language model pretraining across multiple model sizes. Experimental results show consistent improvements in training stability, optimization behavior, and downstream task performance compared to standard softmax attention and attention sink baselines. These findings suggest that modest reweighting of attention outputs provides a practical and effective way to improve attention behavior in Transformer models.

In Formula 1, race strategies are adapted according to evolving race conditions and competitors' actions. This paper proposes a reinforcement learning approach for multi-agent race strategy optimization. Agents learn to balance energy management, tire degradation, aerodynamic interaction, and pit-stop decisions. Building on a pre-trained single-agent policy, we introduce an interaction module that accounts for the behavior of competitors. The combination of the interaction module and a self-play training scheme generates competitive policies, and agents are ranked based on their relative performance. Results show that the agents adapt pit timing, tire selection, and energy allocation in response to opponents, achieving robust and consistent race performance. Because the framework relies only on information available during real races, it can support race strategists' decisions before and during races.

Deep state-space models (DSSMs) enable temporal predictions by learning the underlying dynamics of observed sequence data. They are often trained by maximising the evidence lower bound. However, as we show, this does not ensure the model actually learns the underlying dynamics. We therefore propose a constrained optimisation framework as a general approach for training DSSMs. Building upon this, we introduce the extended Kalman VAE (EKVAE), which combines amortised variational inference with classic Bayesian filtering/smoothing to model dynamics more accurately than RNN-based DSSMs. Our results show that the constrained optimisation framework significantly improves system identification and prediction accuracy on the example of established state-of-the-art DSSMs. The EKVAE outperforms previous models w.r.t. prediction accuracy, achieves remarkable results in identifying dynamical systems, and can furthermore successfully learn state-space representations where static and dynamic features are disentangled.

Context engineering has emerged as a pivotal paradigm for unlocking the potential of Large Language Models (LLMs) in Software Engineering (SE) tasks, enabling performance gains at test time without model fine-tuning. Despite its success, existing research lacks a systematic taxonomy of SE-specific context types and a dedicated benchmark to quantify the heterogeneous effects of different contexts across core SE workflows. To address this gap, we propose CL4SE (Context Learning for Software Engineering), a comprehensive benchmark featuring a fine-grained taxonomy of four SE-oriented context types (interpretable examples, project-specific context, procedural decision-making context, and positive & negative context), each mapped to a representative task (code generation, code summarization, code review, and patch correctness assessment). We construct high-quality datasets comprising over 13,000 samples from more than 30 open-source projects and evaluate five mainstream LLMs across nine metrics. Extensive experiments demonstrate that context learning yields an average performance improvement of 24.7% across all tasks. Specifically, procedural context boosts code review performance by up to 33% (Qwen3-Max), mixed positive-negative context improves patch assessment by 30% (DeepSeek-V3), project-specific context increases code summarization BLEU by 14.78% (GPT-Oss-120B), and interpretable examples enhance code generation PASS@1 by 5.72% (DeepSeek-V3). CL4SE establishes the first standardized evaluation framework for SE context learning, provides actionable empirical insights into task-specific context design, and releases a large-scale dataset to facilitate reproducible research in this domain.

Large language model (LLM) serving infrastructures are undergoing a shift toward heterogeneity and disaggregation. Modern deployments increasingly integrate diverse accelerators and near-memory processing technologies, introducing significant hardware heterogeneity, while system software increasingly separates computation, memory, and model components across distributed resources to improve scalability and efficiency. As a result, LLM serving performance is no longer determined by hardware or software choices in isolation, but by their runtime interaction through scheduling, data movement, and interconnect behavior. However, understanding these interactions remains challenging, as existing simulators lack the ability to jointly model heterogeneous hardware and disaggregated serving techniques within a unified, runtime-driven framework. This paper presents LLMServingSim 2.0, a unified system-level simulator designed to make runtime-driven hardware-software interactions in heterogeneous and disaggregated LLM serving infrastructures explicit and analyzable. LLMServingSim 2.0 embeds serving decisions and hardware behavior into a single runtime loop, enabling interaction-aware modeling of batching, routing, offloading, memory, and power. The simulator supports extensible integration of emerging accelerators and memory systems through profile-based modeling, while capturing dynamic serving behavior and system-level effects. We validate LLMServingSim 2.0 against real deployments, showing that it reproduces key performance, memory, and power metrics with an average error of 0.97%, while maintaining simulation times of around 10 minutes even for complex configurations. These results demonstrate that LLMServingSim 2.0 provides a practical bridge between hardware innovation and serving-system design, enabling systematic exploration and co-design for next-generation LLM serving infrastructures.

Cardiac blood flow patterns contain rich information about disease severity and clinical interventions, yet current imaging and computational methods fail to capture underlying relational structures of coherent flow features. We propose a physics-informed, latent relational framework to model cardiac vortices as interacting nodes in a graph. Our model combines a neural relational inference architecture with physics-inspired interaction energy and birth-death dynamics, yielding a latent graph sensitive to disease severity and intervention level. We first apply this to computational fluid dynamics simulations of aortic coarctation. Learned latent graphs reveal that as the aortic radius narrows, vortex interactions become stronger and more frequent. This leads to a higher graph entropy, correlating monotonically with coarctation severity ($R^2=0.78$, Spearman $|ρ|=0.96$). We then extend this method to ultrasound datasets of left ventricles under varying levels of left ventricular assist device support. Again the latent graph representation captures the weakening of coherent vortical structures, thereby demonstrating cross-modal generalisation. Results show latent interaction graphs and entropy serve as robust and interpretable markers of cardiac disease and intervention.

We study the fundamental problem of clustering $n$ points into $K$ groups drawn from a mixture of isotropic Gaussians in $\mathbb{R}^d$. Specifically, we investigate the requisite minimal distance $Δ$ between mean vectors to partially recover the underlying partition. While the minimax-optimal threshold for $Δ$ is well-established, a significant gap exists between this information-theoretic limit and the performance of known polynomial-time procedures. Although this gap was recently characterized in the high-dimensional regime ($n \leq dK$), it remains largely unexplored in the moderate-dimensional regime ($n \geq dK$). In this manuscript, we address this regime by establishing a new low-degree polynomial lower bound for the moderate-dimensional case when $d \geq K$. We show that while the difficulty of clustering for $n \leq dK$ is primarily driven by dimension reduction and spectral methods, the moderate-dimensional regime involves more delicate phenomena leading to a "non-parametric rate". We provide a novel non-spectral algorithm matching this rate, shedding new light on the computational limits of the clustering problem in moderate dimension.

Detecting visual anomalies in industrial inspection often requires training with only a few normal images per category. Recent few-shot methods achieve strong results employing foundation-model features, but typically rely on memory banks, auxiliary datasets, or multi-modal tuning of vision-language models. We therefore question whether such complexity is necessary given the feature representations of vision foundation models. To answer this question, we introduce SubspaceAD, a training-free method, that operates in two simple stages. First, patch-level features are extracted from a small set of normal images by a frozen DINOv2 backbone. Second, a Principal Component Analysis (PCA) model is fit to these features to estimate the low-dimensional subspace of normal variations. At inference, anomalies are detected via the reconstruction residual with respect to this subspace, producing interpretable and statistically grounded anomaly scores. Despite its simplicity, SubspaceAD achieves state-of-the-art performance across one-shot and few-shot settings without training, prompt tuning, or memory banks. In the one-shot anomaly detection setting, SubspaceAD achieves image-level and pixel-level AUROC of 98.0% and 97.6% on the MVTec-AD dataset, and 93.3% and 98.3% on the VisA dataset, respectively, surpassing prior state-of-the-art results. Code and demo are available at https://github.com/CLendering/SubspaceAD.

Continuous value prediction plays a crucial role in industrial-scale recommendation systems, including tasks such as predicting users' watch-time and estimating the gross merchandise value (GMV) in e-commerce transactions. However, it remains challenging due to the highly complex and long-tailed nature of the data distributions. Existing generative approaches rely on rigid parametric distribution assumptions, which fundamentally limits their performance when such assumptions misalign with real-world data. Overly simplified forms cannot adequately model real-world complexities, while more intricate assumptions often suffer from poor scalability and generalization. To address these challenges, we propose a residual quantization (RQ)-based sequence learning framework that represents target continuous values as a sum of ordered quantization codes, predicted recursively from coarse to fine granularity with diminishing quantization errors. We introduce a representation learning objective that aligns RQ code embedding space with the ordinal structure of target values, allowing the model to capture continuous representations for quantization codes and further improving prediction accuracy. We perform extensive evaluations on public benchmarks for lifetime value (LTV) and watch-time prediction, alongside a large-scale online experiment for GMV prediction on an industrial short-video recommendation platform. The results consistently show that our approach outperforms state-of-the-art methods, while demonstrating strong generalization across diverse continuous value prediction tasks in recommendation systems.

Exploration remains the key bottleneck for large language model agents trained with reinforcement learning. While prior methods exploit pretrained knowledge, they fail in environments requiring the discovery of novel states. We propose Exploratory Memory-Augmented On- and Off-Policy Optimization (EMPO$^2$), a hybrid RL framework that leverages memory for exploration and combines on- and off-policy updates to make LLMs perform well with memory while also ensuring robustness without it. On ScienceWorld and WebShop, EMPO$^2$ achieves 128.6% and 11.3% improvements over GRPO, respectively. Moreover, in out-of-distribution tests, EMPO$^2$ demonstrates superior adaptability to new tasks, requiring only a few trials with memory and no parameter updates. These results highlight EMPO$^2$ as a promising framework for building more exploratory and generalizable LLM-based agents.

Simulating a Gaussian process requires sampling from a high-dimensional Gaussian distribution, which scales cubically with the number of sample locations. Spectral methods address this challenge by exploiting the Fourier representation, treating the spectral density as a probability distribution for Monte Carlo approximation. Although this probabilistic interpretation works for stationary processes, it is overly restrictive for the nonstationary case, where spectral densities are generally not probability measures. We propose regular Fourier features for harmonizable processes that avoid this limitation. Our method discretizes the spectral representation directly, preserving the correlation structure among spectral weights without requiring probability assumptions. Under a finite spectral support assumption, this yields an efficient low-rank approximation that is positive semi-definite by construction. When the spectral density is unknown, the framework extends naturally to kernel learning from data. We demonstrate the method on locally stationary kernels and on harmonizable mixture kernels with complex-valued spectral densities.

Applying LLM-based multi-agent software systems in safety-critical domains such as lifespan echocardiography introduces system-level risks that cannot be addressed by improving model accuracy alone. During system operation, beyond individual LLM behavior, uncertainty propagates through agent coordination, data pipelines, human-in-the-loop interaction, and runtime control logic. Yet existing work largely treats uncertainty at the model level rather than as a first-class software engineering concern. This paper approaches uncertainty from both system-level and runtime perspectives. We first differentiate epistemological and ontological uncertainties in the context of LLM-based multi-agent software system operation. Building on this foundation, we propose a lifecycle-based uncertainty management framework comprising four mechanisms: representation, identification, evolution, and adaptation. The uncertainty lifecycle governs how uncertainties emerge, transform, and are mitigated across architectural layers and execution phases, enabling structured runtime governance and controlled adaptation. We demonstrate the feasibility of the framework using a real-world LLM-based multi-agent echocardiographic software system developed in clinical collaboration, showing improved reliability and diagnosability in diagnostic reasoning. The proposed approach generalizes to other safety-critical LLM-based multi-agent software systems, supporting principled operational control and runtime assurance beyond model-centric methods.

The use of hearing aids will increase in the coming years due to demographic change. One open problem that remains to be solved by a new generation of hearing aids is the cocktail party problem. A possible solution is electroencephalography-based auditory attention decoding. This has been the subject of several studies in recent years, which have in common that they use the same preprocessing methods in most cases. In this work, in order to achieve an advantage, the use of a scattering transform is proposed as an alternative to these preprocessing methods. The two-layer scattering transform is compared with a regular filterbank, the synchrosqueezing short-time Fourier transform and the common preprocessing. To demonstrate the performance, the known and the proposed preprocessing methods are compared for different classification tasks on two widely used datasets, provided by the KU Leuven (KUL) and the Technical University of Denmark (DTU). Both established and new neural-network-based models, CNNs, LSTMs, and recent Transformer/graph-based models are used for classification. Various evaluation strategies were compared, with a focus on the task of classifying speakers who are unknown from the training. We show that the two-layer scattering transform can significantly improve the performance for subject-related conditions, especially on the KUL dataset. However, on the DTU dataset, this only applies to some of the models, or when larger amounts of training data are provided, as in 10-fold cross-validation. This suggests that the scattering transform is capable of extracting additional relevant information.

Training divergence in transformers wastes compute, yet practitioners discover instability only after expensive runs begin. They therefore need an expected probability of failure for a transformer before training starts. Our study of Residual Koopman Spectral Profiling (RKSP) provides such an estimate. From a single forward pass at initialization, RKSP extracts Koopman spectral features by applying whitened dynamic mode decomposition to layer-wise residual snapshots. Our central diagnostic, the near-unit spectral mass, quantifies the fraction of modes concentrated near the unit circle, which captures instability risk. For predicting divergence across extensive configurations, this estimator achieves an AUROC of 0.995, outperforming the best gradient baseline. We further make this diagnostic actionable through Koopman Spectral Shaping (KSS), which reshapes spectra during training. We empirically validate that our method works in practice: RKSP predicts divergence at initialization, and when RKSP flags high risk, turning on KSS successfully prevents divergence. In the challenging high learning rate regime without normalization layers, KSS reduces the divergence rate from 66.7% to 12.5% and enables learning rates that are 50% to 150% higher. These findings generalize to WikiText-103 language modeling, vision transformers on CIFAR-10, and pretrained language models, including GPT-2 and LLaMA-2 up to 7B, as well as emerging architectures such as MoE, Mamba-style SSMs, and KAN.

We introduce kernel integrated $R^2$, a new measure of statistical dependence that combines the local normalization principle of the recently introduced integrated $R^2$ with the flexibility of reproducing kernel Hilbert spaces (RKHSs). The proposed measure extends integrated $R^2$ from scalar responses to responses taking values on general spaces equipped with a characteristic kernel, allowing to measure dependence of multivariate, functional, and structured data, while remaining sensitive to tail behaviour and oscillatory dependence structures. We establish that (i) this new measure takes values in $[0,1]$, (ii) equals zero if and only if independence holds, and (iii) equals one if and only if the response is almost surely a measurable function of the covariates. Two estimators are proposed: a graph-based method using $K$-nearest neighbours and an RKHS-based method built on conditional mean embeddings. We prove consistency and derive convergence rates for the graph-based estimator, showing its adaptation to intrinsic dimensionality. Numerical experiments on simulated data and a real data experiment in the context of dependency testing for media annotations demonstrate competitive power against state-of-the-art dependence measures, particularly in settings involving non-linear and structured relationships.

As Large Language Models (LLMs) are increasingly used, their security risks have drawn increasing attention. Existing research reveals that LLMs are highly susceptible to jailbreak attacks, with effectiveness varying across language contexts. This paper investigates the role of classical Chinese in jailbreak attacks. Owing to its conciseness and obscurity, classical Chinese can partially bypass existing safety constraints, exposing notable vulnerabilities in LLMs. Based on this observation, this paper proposes a framework, CC-BOS, for the automatic generation of classical Chinese adversarial prompts based on multi-dimensional fruit fly optimization, facilitating efficient and automated jailbreak attacks in black-box settings. Prompts are encoded into eight policy dimensions-covering role, behavior, mechanism, metaphor, expression, knowledge, trigger pattern and context; and iteratively refined via smell search, visual search, and cauchy mutation. This design enables efficient exploration of the search space, thereby enhancing the effectiveness of black-box jailbreak attacks. To enhance readability and evaluation accuracy, we further design a classical Chinese to English translation module. Extensive experiments demonstrate that effectiveness of the proposed CC-BOS, consistently outperforming state-of-the-art jailbreak attack methods.

Decoding brain activity from electroencephalography (EEG) is crucial for neuroscience and clinical applications. Among recent advances in deep learning for EEG, geometric learning stands out as its theoretical underpinnings on symmetric positive definite (SPD) allows revealing structural connectivity analysis in a physics-grounded manner. However, current SPD-based methods focus predominantly on statistical aggregation of EEGs, with frequency-specific synchronization and local topological structures of brain regions neglected. Given this, we propose RepSPD, a novel geometric deep learning (GDL)-based model. RepSPD implements a cross-attention mechanism on the Riemannian manifold to modulate the geometric attributes of SPD with graph-derived functional connectivity features. On top of this, we introduce a global bidirectional alignment strategy to reshape tangent-space embeddings, mitigating geometric distortions caused by curvature and thereby enhancing geometric consistency. Extensive experiments demonstrate that our proposed framework significantly outperforms existing EEG representation methods, exhibiting superior robustness and generalization capabilities.

Human-in-the-loop validation is essential in safety-critical clinical AI, yet the transition between initial model inference and expert correction is rarely analyzed as a structured signal. We introduce a diagnostic alignment framework in which the AI-generated image based report is preserved as an immutable inference state and systematically compared with the physician-validated outcome. The inference pipeline integrates a vision-enabled large language model, BERT- based medical entity extraction, and a Sequential Language Model Inference (SLMI) step to enforce domain-consistent refinement prior to expert review. Evaluation on 21 dermatological cases (21 complete AI physician pairs) em- ployed a four-level concordance framework comprising exact primary match rate (PMR), semantic similarity-adjusted rate (AMR), cross-category alignment, and Comprehensive Concordance Rate (CCR). Exact agreement reached 71.4% and remained unchanged under semantic similarity (t = 0.60), while structured cross-category and differential overlap analysis yielded 100% comprehensive concordance (95% CI: [83.9%, 100%]). No cases demonstrated complete diagnostic divergence. These findings show that binary lexical evaluation substantially un- derestimates clinically meaningful alignment. Modeling expert validation as a structured transformation enables signal-aware quantification of correction dynamics and supports traceable, human aligned evaluation of image based clinical decision support systems.

As LLMs achieved breakthroughs in general reasoning, their proficiency in specialized scientific domains reveals pronounced gaps in existing benchmarks due to data contamination, insufficient complexity, and prohibitive human labor costs. Here we present SPM-Bench, an original, PhD-level multimodal benchmark specifically designed for scanning probe microscopy (SPM). We propose a fully automated data synthesis pipeline that ensures both high authority and low-cost. By employing Anchor-Gated Sieve (AGS) technology, we efficiently extract high-value image-text pairs from arXiv and journal papers published between 2023 and 2025. Through a hybrid cloud-local architecture where VLMs return only spatial coordinates "llbox" for local high-fidelity cropping, our pipeline achieves extreme token savings while maintaining high dataset purity. To accurately and objectively evaluate the performance of the LLMs, we introduce the Strict Imperfection Penalty F1 (SIP-F1) score. This metric not only establishes a rigorous capability hierarchy but also, for the first time, quantifies model "personalities" (Conservative, Aggressive, Gambler, or Wise). By correlating these results with model-reported confidence and perceived difficulty, we expose the true reasoning boundaries of current AI in complex physical scenarios. These insights establish SPM-Bench as a generalizable paradigm for automated scientific data synthesis.

Understanding how neural networks arrive at their predictions is essential for debugging, auditing, and deployment. Mechanistic interpretability pursues this goal by identifying circuits - minimal subnetworks responsible for specific behaviors. However, existing circuit discovery methods are brittle: circuits depend strongly on the chosen concept dataset and often fail to transfer out-of-distribution, raising doubts whether they capture concept or dataset-specific artifacts. We introduce Certified Circuits, which provide provable stability guarantees for circuit discovery. Our framework wraps any black-box discovery algorithm with randomized data subsampling to certify that circuit component inclusion decisions are invariant to bounded edit-distance perturbations of the concept dataset. Unstable neurons are abstained from, yielding circuits that are more compact and more accurate. On ImageNet and OOD datasets, certified circuits achieve up to 91% higher accuracy while using 45% fewer neurons, and remain reliable where baselines degrade. Certified Circuits puts circuit discovery on formal ground by producing mechanistic explanations that are provably stable and better aligned with the target concept. Code will be released soon!

Discovering interpretable physical laws from high-dimensional data is a fundamental challenge in scientific research. Traditional methods, such as symbolic regression, often produce complex, unphysical formulas when searching a vast space of possible forms. We introduce a framework that guides the search process by leveraging the embedded scientific knowledge of large language models, enabling efficient identification of physical laws in the data. We validate our approach by modeling key properties of perovskite materials. Our method mitigates the combinatorial explosion commonly encountered in traditional symbolic regression, reducing the effective search space by a factor of approximately $10^5$. A set of novel formulas for bulk modulus, band gap, and oxygen evolution reaction activity are identified, which not only provide meaningful physical insights but also outperform previous formulas in accuracy and simplicity.

Multimodal large language models (MLLMs) have substantially advanced video misinformation detection through unified multimodal reasoning, but they often rely on fixed-depth inference and place excessive trust in internally generated assumptions, particularly in scenarios where critical evidence is sparse, fragmented, or requires external verification. To address these limitations, we propose FactGuard, an agentic framework for video misinformation detection that formulates verification as an iterative reasoning process built upon MLLMs. FactGuard explicitly assesses task ambiguity and selectively invokes external tools to acquire critical evidence, enabling progressive refinement of reasoning trajectories. To further strengthen this capability, we introduce a two-stage training strategy that combines domain-specific agentic supervised fine-tuning with decision-aware reinforcement learning to optimize tool usage and calibrate risk-sensitive decision making. Extensive experiments on FakeSV, FakeTT, and FakeVV demonstrate FactGuard's state-of-the-art performance and validate its excellent robustness and generalization capacity.

Data-driven models are revolutionizing weather forecasting. To optimize training efficiency and model performance, this paper analyzes empirical scaling laws within this domain. We investigate the relationship between model performance (validation loss) and three key factors: model size ($N$), dataset size ($D$), and compute budget ($C$). Across a range of models, we find that Aurora exhibits the strongest data-scaling behavior: increasing the training dataset by 10x reduces validation loss by up to 3.2x. GraphCast demonstrates the highest parameter efficiency, yet suffers from limited hardware utilization. Our compute-optimal analysis indicates that, under fixed compute budgets, allocating resources to longer training durations yields greater performance gains than increasing model size. Furthermore, we analyze model shape and uncover scaling behaviors that differ fundamentally from those observed in language models: weather forecasting models consistently favor increased width over depth. These findings suggest that future weather models should prioritize wider architectures and larger effective training datasets to maximize predictive performance.

We present frequency-ordered tokenization, a simple preprocessing technique that improves lossless text compression by exploiting the power-law frequency distribution of natural language tokens (Zipf's law). The method tokenizes text with Byte Pair Encoding (BPE), reorders the vocabulary so that frequent tokens receive small integer identifiers, and encodes the result with variable-length integers before passing it to any standard compressor. On enwik8 (100 MB Wikipedia), this yields improvements of 7.08 percentage points (pp) for zlib, 1.69 pp for LZMA, and 0.76 pp for zstd (all including vocabulary overhead), outperforming the classical Word Replacing Transform. Gains are consistent at 1 GB scale (enwik9) and across Chinese and Arabic text. We further show that preprocessing accelerates compression for computationally expensive algorithms: the total wall-clock time including preprocessing is 3.1x faster than raw zstd-22 and 2.4x faster than raw LZMA, because the preprocessed input is substantially smaller. The method can be implemented in under 50 lines of code.

Accurate brain tumor diagnosis requires models to not only detect lesions but also generate clinically interpretable reasoning grounded in imaging manifestations, yet existing public datasets remain limited in annotation richness and diagnostic semantics. To bridge this gap, we introduce MM-NeuroOnco, a large-scale multimodal benchmark and instruction-tuning dataset for brain tumor MRI understanding, consisting of 24,726 MRI slices from 20 data sources paired with approximately 200,000 semantically enriched multimodal instructions spanning diverse tumor subtypes and imaging modalities. To mitigate the scarcity and high cost of diagnostic semantic annotations, we develop a multi-model collaborative pipeline for automated medical information completion and quality control, enabling the generation of diagnosis-related semantics beyond mask-only annotations. Building upon this dataset, we further construct MM-NeuroOnco-Bench, a manually annotated evaluation benchmark with a rejection-aware setting to reduce biases inherent in closed-ended question formats. Evaluation across ten representative models shows that even the strongest baseline, Gemini 3 Flash, achieves only 41.88% accuracy on diagnosis-related questions, highlighting the substantial challenges of multimodal brain tumor diagnostic understanding. Leveraging MM-NeuroOnco, we further propose NeuroOnco-GPT, which achieves a 27% absolute accuracy improvement on diagnostic questions following fine-tuning. This result demonstrates the effectiveness of our dataset and benchmark in advancing clinically grounded multimodal diagnostic reasoning. Code and dataset are publicly available at: https://github.com/gfnnnb/MM-NeuroOnco

The promise of general-purpose agents - systems that perform tasks in unfamiliar environments without domain-specific engineering - remains largely unrealized. Existing agents are predominantly specialized, and while emerging implementations like OpenAI SDK Agent and Claude Code hint at broader capabilities, no systematic evaluation of their general performance has been pursued. Current agentic benchmarks assume domain-specific integration, encoding task information in ways that preclude fair evaluation of general agents. This paper frames general-agent evaluation as a first-class research objective. We propose conceptual principles for such evaluation, a Unified Protocol enabling agent-benchmark integration, and Exgentic - a practical framework for general agent evaluation. We benchmark five prominent agent implementations across six environments as the first Open General Agent Leaderboard. Our experiments show that general agents generalize across diverse environments, achieving performance comparable to domain-specific agents without any environment-specific tuning. We release our evaluation protocol, framework, and leaderboard to establish a foundation for systematic research on general-purpose agents.

Parameter-efficient fine-tuning has demonstrated promising results across various visual adaptation tasks, such as classification and segmentation. Typically, prompt tuning techniques have harnessed knowledge from a single pre-trained model, whether from a general or a specialized medical domain. However, this approach typically overlooks the potential synergies that could arise from integrating diverse domain knowledge within the same tuning process. In this work, we propose a novel Mixture-of-Experts prompt tuning method called pMoE, which leverages the strengths of multiple expert domains through expert-specialized prompt tokens and the learnable dispatcher, effectively combining their expertise in a unified model framework. Our pMoE introduces expert-specific prompt tokens and utilizes a dynamic token dispatching mechanism at various prompt layers to optimize the contribution of each domain expert during the adaptation phase. By incorporating both domain knowledge from diverse experts, the proposed pMoE significantly enhances the model's versatility and applicability to a broad spectrum of tasks. We conduct extensive experiments across 47 adaptation tasks, including both classification and segmentation in general and medical domains. The results demonstrate that our pMoE not only achieves superior performance with a large margin of improvements but also offers an optimal trade-off between computational efficiency and adaptation effectiveness compared to existing methods.

We introduce Manifold Sobolev Informed Neural Optimization (MSINO), a curvature aware training framework for neural networks defined on Riemannian manifolds. The method replaces standard Euclidean derivative supervision with a covariant Sobolev loss that aligns gradients using parallel transport and improves stability via a Laplace Beltrami smoothness regularization term. Building on classical results in Riemannian optimization and Sobolev theory on manifolds, we derive geometry dependent constants that yield (i) a Descent Lemma with a manifold Sobolev smoothness constant, (ii) a Sobolev Polyak Lojasiewicz inequality giving linear convergence guarantees for Riemannian gradient descent and stochastic gradient descent under explicit step size bounds, and (iii) a two step Newton Sobolev method with local quadratic contraction in curvature controlled neighborhoods. Unlike prior Sobolev training in Euclidean space, MSINO provides training time guarantees that explicitly track curvature and transported Jacobians. Applications include surface imaging, physics informed learning settings, and robotics on Lie groups such as SO(3) and SE(3). The framework unifies value and gradient based learning with curvature aware convergence guarantees for neural training on manifolds.

Algorithmic stability is among the most potent techniques in generalization analysis. However, its derivation usually requires a stepsize $η_t = \mathcal{O}(1/t)$ under non-convex training regimes, where $t$ denotes iterations. This rigid decay of the stepsize potentially impedes optimization and may not align with practical scenarios. In this paper, we derive the generalization bounds under the homogeneous neural network regimes, proving that this regime enables slower stepsize decay of order $Ω(1/\sqrt{t})$ under mild assumptions. We further extend the theoretical results from several aspects, e.g., non-Lipschitz regimes. This finding is broadly applicable, as homogeneous neural networks encompass fully-connected and convolutional neural networks with ReLU and LeakyReLU activations.

Despite being one of the most widely spoken languages globally, Bangla remains a low-resource language in the field of Natural Language Processing (NLP). Mainstream Automatic Speech Recognition (ASR) and Speaker Diarization systems for Bangla struggles when processing longform audio exceeding 3060 seconds. This paper presents a robust framework specifically engineered for extended Bangla content by leveraging preexisting models enhanced with novel optimization pipelines for the DL Sprint 4.0 contest. Our approach utilizes Voice Activity Detection (VAD) optimization and Connectionist Temporal Classification (CTC) segmentation via forced word alignment to maintain temporal accuracy and transcription integrity over long durations. Additionally, we employed several finetuning techniques and preprocessed the data using augmentation techniques and noise removal. By bridging the performance gap in complex, multi-speaker environments, this work provides a scalable solution for real-world, longform Bangla speech applications.

We study wide Bayesian neural networks focusing on the rare but statistically dominant fluctuations that govern posterior concentration, beyond Gaussian-process limits. Large-deviation theory provides explicit variational objectives-rate functions-on predictors, providing an emerging notion of complexity and feature learning directly at the functional level. We show that the posterior output rate function is obtained by a joint optimization over predictors and internal kernels, in contrast with fixed-kernel (NNGP) theory. Numerical experiments demonstrate that the resulting predictions accurately describe finite-width behavior for moderately sized networks, capturing non-Gaussian tails, posterior deformation, and data-dependent kernel selection effects.

Vision-language models (VLMs) can read text from images, but where does this optical character recognition (OCR) information enter the language processing stream? We investigate the OCR routing mechanism across three architecture families (Qwen3-VL, Phi-4, InternVL3.5) using causal interventions. By computing activation differences between original images and text-inpainted versions, we identify architecture-specific OCR bottlenecks whose dominant location depends on the vision-language integration strategy: DeepStack models (Qwen) show peak sensitivity at mid-depth (about 50%) for scene text, while single-stage projection models (Phi-4, InternVL) peak at early layers (6-25%), though the exact layer of maximum effect varies across datasets. The OCR signal is remarkably low-dimensional: PC1 captures 72.9% of variance. Crucially, principal component analysis (PCA) directions learned on one dataset transfer to others, demonstrating shared text-processing pathways. Surprisingly, in models with modular OCR circuits (notably Qwen3-VL-4B), OCR removal can improve counting performance (up to +6.9 percentage points), suggesting OCR interferes with other visual processing in sufficiently modular architectures.

With the rapid evolution of Large Language Models, generative recommendation is gradually reshaping the paradigm of recommender systems. However, most existing methods are still confined to the interaction-driven next-item prediction paradigm, failing to rapidly adapt to evolving trends or address diverse recommendation tasks along with business-specific requirements in real-world scenarios. To this end, we present SIGMA, a Semantic-Grounded Instruction-Driven Generative Multi-Task Recommender at AliExpress. Specifically, we first ground item entities in general semantics via a unified latent space capturing both semantic and collaborative relations. Building upon this, we develop a hybrid item tokenization method for precise modeling and efficient generation. Moreover, we construct a large-scale multi-task SFT dataset to empower SIGMA to fulfill various recommendation demands via instruction-following. Finally, we design a three-step item generation procedure integrated with an adaptive probabilistic fusion mechanism to calibrate the output distributions based on task-specific requirements for recommendation accuracy and diversity. Extensive offline experiments and online A/B tests demonstrate the effectiveness of SIGMA.

Low-Rank Adaptation (LoRA) dominates parameter-efficient fine-tuning (PEFT). However, it faces a critical ``linear ceiling'' in complex reasoning tasks: simply increasing the rank yields diminishing returns due to intrinsic linear constraints. We introduce NoRA (Non-linear Rank Adaptation), a weight-level parallel adapter that injects SiLU gating and structural dropout to induce manifold expansion. On the SlimOrca benchmark, NoRA breaks this linear barrier: NoRA remarkably at rank 64 (PPL 3.89) outperforms LoRA at rank 512 (PPL 3.90), demonstrating superior spectral efficiency. This advantage generalizes to mathematical reasoning, where NoRA achieves a perplexity of 1.97 on MathInstruct, significantly surpassing LoRA's saturation point of 2.07. Mechanism analysis via Singular Value Decomposition (SVD) confirms that NoRA activates the dormant tail of the singular value spectrum, effectively preventing the rank collapse observed in linear methods.

Multimodal Entity Alignment (MMEA) aims to identify equivalent entities across different data modalities, enabling structural data integration that in turn improves the performance of various large language model applications. To lift the requirement of labeled seed pairs that are difficult to obtain, recent methods shifted to an unsupervised paradigm using pseudo-alignment seeds. However, unsupervised entity alignment in multimodal settings remains underexplored, mainly because the incorporation of multimodal information often results in imbalanced coverage of pseudo-seeds within the knowledge graph. To overcome this, we propose PSQE (Pseudo-Seed Quality Enhancement) to improve the precision and graph coverage balance of pseudo seeds via multimodal information and clustering-resampling. Theoretical analysis reveals the impact of pseudo seeds on existing contrastive learning-based MMEA models. In particular, pseudo seeds can influence the attraction and the repulsion terms in contrastive learning at once, whereas imbalanced graph coverage causes models to prioritize high-density regions, thereby weakening their learning capability for entities in sparse regions. Experimental results validate our theoretical findings and show that PSQE as a plug-and-play module can improve the performance of baselines by considerable margins.

This study investigates a data-driven machine learning approach to predict membrane fouling in critically ill patients undergoing Continuous Renal Replacement Therapy (CRRT). Using time-series data from an ICU, 16 clinically selected features were identified to train predictive models. To ensure interpretability and enable reliable counterfactual analysis, the researchers adopted a tabular data approach rather than modeling temporal dependencies directly. Given the imbalance between fouling and non-fouling cases, the ADASYN oversampling technique was applied to improve minority class representation. Random Forest, XGBoost, and LightGBM models were tested, achieving balanced performance with 77.6% sensitivity and 96.3% specificity at a 10% rebalancing rate. Results remained robust across different forecasting horizons. Notably, the tabular approach outperformed LSTM recurrent neural networks, suggesting that explicit temporal modeling was not necessary for strong predictive performance. Feature selection further reduced the model to five key variables, improving simplicity and interpretability with minimal loss of accuracy. A Shapley value-based counterfactual analysis was applied to the best-performing model, successfully identifying minimal input changes capable of reversing fouling predictions. Overall, the findings support the viability of interpretable machine learning models for predicting membrane fouling during CRRT. The integration of prediction and counterfactual analysis offers practical clinical value, potentially guiding therapeutic adjustments to reduce fouling risk and improve patient management.

Human intelligence naturally intertwines omni-modal perception -- spanning vision, audio, and language -- with complex reasoning and tool usage to interact with the world. However, current multi-modal LLMs are primarily confined to bi-modal interactions (e.g., vision-language), lacking the unified cognitive capabilities required for general AI assistants. To bridge this gap, we introduce OmniGAIA, a comprehensive benchmark designed to evaluate omni-modal agents on tasks necessitating deep reasoning and multi-turn tool execution across video, audio, and image modalities. Constructed via a novel omni-modal event graph approach, OmniGAIA synthesizes complex, multi-hop queries derived from real-world data that require cross-modal reasoning and external tool integration. Furthermore, we propose OmniAtlas, a native omni-modal foundation agent under tool-integrated reasoning paradigm with active omni-modal perception. Trained on trajectories synthesized via a hindsight-guided tree exploration strategy and OmniDPO for fine-grained error correction, OmniAtlas effectively enhances the tool-use capabilities of existing open-source models. This work marks a step towards next-generation native omni-modal AI assistants for real-world scenarios.

Implementations of symmetric positive definite (SPD) matrix-based neural networks for neural decoding remain fragmented across research codebases and Python packages. Existing implementations often employ ad hoc handling of manifold constraints and non-unified training setups, which hinders reproducibility and integration into modern deep-learning workflows. To address this gap, we introduce SPD Learn, a unified and modular Python package for geometric deep learning with SPD matrices. SPD Learn provides core SPD operators and neural-network layers, including numerically stable spectral operators, and enforces Stiefel/SPD constraints via trivialization-based parameterizations. This design enables standard backpropagation and optimization in unconstrained Euclidean spaces while producing manifold-constrained parameters by construction. The package also offers reference implementations of representative SPDNet-based models and interfaces with widely used brain computer interface/neuroimaging toolkits and modern machine-learning libraries (e.g., MOABB, Braindecode, Nilearn, and SKADA), facilitating reproducible benchmarking and practical deployment.

Amortized Bayesian Inference (ABI) enables efficient posterior estimation using generative neural networks trained on simulated data, but often suffers from performance degradation under model misspecification. While self-consistency (SC) training on unlabeled empirical data can enhance network robustness, current approaches are limited to static, single-task settings and fail to handle sequentially arriving data or distribution shifts. We propose a continual learning framework for ABI that decouples simulation-based pre-training from unsupervised sequential SC fine-tuning on real-world data. To address the challenge of catastrophic forgetting, we introduce two adaptation strategies: (1) SC with episodic replay, utilizing a memory buffer of past observations, and (2) SC with elastic weight consolidation, which regularizes updates to preserve task-critical parameters. Across three diverse case studies, our methods significantly mitigate forgetting and yield posterior estimates that outperform standard simulation-based training, achieving estimates closer to MCMC reference, providing a viable path for trustworthy ABI across a range of different tasks.

In this article, we provide an axiomatic characterization of feature attribution for multi-output predictors within the Shapley framework. While SHAP explanations are routinely computed independently for each output coordinate, the theoretical necessity of this practice has remained unclear. By extending the classical Shapley axioms to vector-valued cooperative games, we establish a rigidity theorem showing that any attribution rule satisfying efficiency, symmetry, dummy player, and additivity must necessarily decompose component-wise across outputs. Consequently, any joint-output attribution rule must relax at least one of the classical Shapley axioms. This result identifies a previously unformalized structural constraint in Shapley-based interpretability, clarifying the precise scope of fairness-consistent explanations in multi-output learning. Numerical experiments on a biomedical benchmark illustrate that multi-output models can yield computational savings in training and deployment, while producing SHAP explanations that remain fully consistent with the component-wise structure imposed by the Shapley axioms.

Knowledge Tracing (KT) diagnoses students' concept mastery through continuous learning state monitoring in education.Existing methods primarily focus on studying behavioral sequences based on ID or textual information.While existing methods rely on ID-based sequences or shallow textual features, they often fail to capture (1) the hierarchical evolution of cognitive states and (2) individualized problem difficulty perception due to limited semantic modeling. Therefore, this paper proposes a Large Language Model Hyperbolic Aligned Knowledge Tracing(L-HAKT). First, the teacher agent deeply parses question semantics and explicitly constructs hierarchical dependencies of knowledge points; the student agent simulates learning behaviors to generate synthetic data. Then, contrastive learning is performed between synthetic and real data in hyperbolic space to reduce distribution differences in key features such as question difficulty and forgetting patterns. Finally, by optimizing hyperbolic curvature, we explicitly model the tree-like hierarchical structure of knowledge points, precisely characterizing differences in learning curve morphology for knowledge points at different levels. Extensive experiments on four real-world educational datasets validate the effectiveness of our Large Language Model Hyperbolic Aligned Knowledge Tracing (L-HAKT) framework.

We introduce a theoretical framework that connects multi-chart autoencoders in manifold learning with the classical theory of vector bundles and characteristic classes. Rather than viewing autoencoders as producing a single global Euclidean embedding, we treat a collection of locally trained encoder-decoder pairs as a learned atlas on a manifold. We show that any reconstruction-consistent autoencoder atlas canonically defines transition maps satisfying the cocycle condition, and that linearising these transition maps yields a vector bundle coinciding with the tangent bundle when the latent dimension matches the intrinsic dimension of the manifold. This construction provides direct access to differential-topological invariants of the data. In particular, we show that the first Stiefel-Whitney class can be computed from the signs of the Jacobians of learned transition maps, yielding an algorithmic criterion for detecting orientability. We also show that non-trivial characteristic classes provide obstructions to single-chart representations, and that the minimum number of autoencoder charts is determined by the good cover structure of the manifold. Finally, we apply our methodology to low-dimensional orientable and non-orientable manifolds, as well as to a non-orientable high-dimensional image dataset.

Reasoning with large language models often benefits from generating multiple chains-of-thought, but existing aggregation strategies are typically trajectory-level (e.g., selecting the best trace or voting on the final answer), discarding useful intermediate work from partial or "nearly correct" attempts. We propose Stitching Noisy Diffusion Thoughts, a self-consistency framework that turns cheap diffusion-sampled reasoning into a reusable pool of step-level candidates. Given a problem, we (i) sample many diverse, low-cost reasoning trajectories using a masked diffusion language model, (ii) score every intermediate step with an off-the-shelf process reward model (PRM), and (iii) stitch these highest-quality steps across trajectories into a composite rationale. This rationale then conditions an autoregressive (AR) model (solver) to recompute only the final answer. This modular pipeline separates exploration (diffusion) from evaluation and solution synthesis, avoiding monolithic unified hybrids while preserving broad search. Across math reasoning benchmarks, we find that step-level recombination is most beneficial on harder problems, and ablations highlight the importance of the final AR solver in converting stitched but imperfect rationales into accurate answers. Using low-confidence diffusion sampling with parallel, independent rollouts, our training-free framework improves average accuracy by up to 23.8% across six math and coding tasks. At the same time, it achieves up to a 1.8x latency reduction relative to both traditional diffusion models (e.g., Dream, LLaDA) and unified architectures (e.g., TiDAR). Code is available at https://github.com/roymiles/diffusion-stitching.

Diffusion Large Language Models (DLLMs) promise fast non-autoregressive inference but suffer a severe quality-speed trade-off in parallel decoding. This stems from the ''combinatorial contradiction'' phenomenon, where parallel tokens form semantically inconsistent combinations. We address this by integrating continuous representations into the discrete decoding process, as they preserve rich inter-position dependency. We propose ReMix (Rejection Mixing), a framework that introduces a novel Continuous Mixing State as an intermediate between the initial masked state and the final decoded token state. This intermediate state allows a token's representation to be iteratively refined in a continuous space, resolving mutual conflicts with other tokens before collapsing into a final discrete sample. Furthermore, a rejection rule reverts uncertain representations from the continuous state back to the masked state for reprocessing, ensuring stability and preventing error propagation. ReMix thus mitigates combinatorial contradictions by enabling continuous-space refinement during discrete diffusion decoding. Extensive experiments demonstrate that ReMix, as a training-free method, achieves a $2-8 \times$ inference speedup without any quality degradation.

Explicit representations of predicate-argument relations form the basis of interpretable semantic analysis, supporting reasoning, generation, and evaluation. However, attaining such semantic structures requires costly annotation efforts and has remained largely confined to English. We leverage the Question-Answer driven Semantic Role Labeling (QA-SRL) framework -- a natural-language formulation of predicate-argument relations -- as the foundation for extending semantic annotation to new languages. To this end, we introduce a cross-linguistic projection approach that reuses an English QA-SRL parser within a constrained translation and word-alignment pipeline to automatically generate question-answer annotations aligned with target-language predicates. Applied to Hebrew, Russian, and French -- spanning diverse language families -- the method yields high-quality training data and fine-tuned, language-specific parsers that outperform strong multilingual LLM baselines (GPT-4o, LLaMA-Maverick). By leveraging QA-SRL as a transferable natural-language interface for semantics, our approach enables efficient and broadly accessible predicate-argument parsing across languages.

Accurate computational identification of DNA methylation is essential for understanding epigenetic regulation. Although deep learning excels in this binary classification task, its "black-box" nature impedes biological insight. We address this by introducing a high-performance model MEDNA-DFM, alongside mechanism-inspired signal purification algorithms. Our investigation demonstrates that MEDNA-DFM effectively captures conserved methylation patterns, achieving robust distinction across diverse species. Validation on external independent datasets confirms that the model's generalization is driven by conserved intrinsic motifs (e.g., GC content) rather than phylogenetic proximity. Furthermore, applying our developed algorithms extracted motifs with significantly higher reliability than prior studies. Finally, empirical evidence from a Drosophila 6mA case study prompted us to propose a "sequence-structure synergy" hypothesis, suggesting that the GAGG core motif and an upstream A-tract element function cooperatively. We further validated this hypothesis via in silico mutagenesis, confirming that the ablation of either or both elements significantly degrades the model's recognition capabilities. This work provides a powerful tool for methylation prediction and demonstrates how explainable deep learning can drive both methodological innovation and the generation of biological hypotheses.

The concept of ranking aggregation plays a central role in preference analysis, and numerous algorithms for calculating median rankings, often originating in social choice theory, have been documented in the literature, offering theoretical guarantees in a centralized setting, i.e., when all the ranking data to be aggregated can be brought together in a single computing unit. For many technologies (e.g. peer-to-peer networks, IoT, multi-agent systems), extending the ability to calculate consensus rankings with guarantees in a decentralized setting, i.e., when preference data is initially distributed across a communicating network, remains a major methodological challenge. Indeed, in recent years, the literature on decentralized computation has mainly focused on computing or optimizing statistics such as arithmetic means using gossip algorithms. The purpose of this article is precisely to study how to achieve reliable consensus on collective rankings using classical rules (e.g. Borda, Copeland) in a decentralized setting, thereby raising new questions, robustness to corrupted nodes, and scalability through reduced communication costs in particular. The approach proposed and analyzed here relies on random gossip communication, allowing autonomous agents to compute global ranking consensus using only local interactions, without coordination or central authority. We provide rigorous convergence guarantees, including explicit rate bounds, for the Borda and Copeland consensus methods. Beyond these rules, we also provide a decentralized implementation of consensus according to the median rank rule and local Kemenization. Extensive empirical evaluations on various network topologies and real and synthetic ranking datasets demonstrate that our algorithms converge quickly and reliably to the correct ranking aggregation.

Argumentation mining comprises several subtasks, among which stance classification focuses on identifying the standpoint expressed in an argumentative text toward a specific target topic. While arguments-especially about controversial topics-often appeal to emotions, most prior work has not systematically incorporated explicit, fine-grained emotion analysis to improve performance on this task. In particular, prior research on stance classification has predominantly utilized non-argumentative texts and has been restricted to specific domains or topics, limiting generalizability. We work on five datasets from diverse domains encompassing a range of controversial topics and present an approach for expanding the Bias-Corrected NRC Emotion Lexicon using DistilBERT embeddings, which we feed into a Neural Argumentative Stance Classification model. Our method systematically expands the emotion lexicon through contextualized embeddings to identify emotionally charged terms not previously captured in the lexicon. Our expanded NRC lexicon (eNRC) improves over the baseline across all five datasets (up to +6.2 percentage points in F1 score), outperforms the original NRC on four datasets (up to +3.0), and surpasses the LLM-based approach on nearly all corpora. We provide all resources-including eNRC, the adapted corpora, and model architecture-to enable other researchers to build upon our work.

Can artificial intelligence truly contribute to creative mathematical research, or does it merely automate routine calculations while introducing risks of error? We provide empirical evidence through a detailed case study: the discovery of novel error representations and bounds for Hermite quadrature rules via systematic human-AI collaboration. Working with multiple AI assistants, we extended results beyond what manual work achieved, formulating and proving several theorems with AI assistance. The collaboration revealed both remarkable capabilities and critical limitations. AI excelled at algebraic manipulation, systematic proof exploration, literature synthesis, and LaTeX preparation. However, every step required rigorous human verification, mathematical intuition for problem formulation, and strategic direction. We document the complete research workflow with unusual transparency, revealing patterns in successful human-AI mathematical collaboration and identifying failure modes researchers must anticipate. Our experience suggests that, when used with appropriate skepticism and verification protocols, AI tools can meaningfully accelerate mathematical discovery while demanding careful human oversight and deep domain expertise.

Presentation generation requires deep content research, coherent visual design, and iterative refinement based on observation. However, existing presentation agents often rely on predefined workflows and fixed templates. To address this, we present DeepPresenter, an agentic framework that adapts to diverse user intents, enables effective feedback-driven refinement, and generalizes beyond a scripted pipeline. Specifically, DeepPresenter autonomously plans, renders, and revises intermediate slide artifacts to support long-horizon refinement with environmental observations. Furthermore, rather than relying on self-reflection over internal signals (e.g., reasoning traces), our environment-grounded reflection conditions the generation process on perceptual artifact states (e.g., rendered slides), enabling the system to identify and correct presentation-specific issues during execution. Results on the evaluation set covering diverse presentation-generation scenarios show that DeepPresenter achieves state-of-the-art performance, and the fine-tuned 9B model remains highly competitive at substantially lower cost. Our project is available at: https://github.com/icip-cas/PPTAgent

Hybrid work has become a reality post-pandemic, transforming how Agile teams deliver value, collaborate, and adapt. This study investigate how hybrid settings influence productivity and collaboration through nine interviews with three Norwegian Agile teams. Our findings show that hybrid work reduces informal interaction, creates uneven participation, and increases reliance on digital tools. Agile ceremonies became alignment anchors, while trust, communication, and tool support mediate team effectiveness. Hybrid Agile work is an evolving field that requires tailored structures to support inclusion, team cohesion, and sustainable performance.

Moral benchmarks for LLMs typically use context-free prompts, implicitly assuming stable preferences. In deployment, however, prompts routinely include contextual signals such as user requests, cues on social norms, etc. that may steer decisions. We study how directed contextual influences reshape decisions in trolley-problem-style moral triage settings. We introduce a pilot evaluation harness for directed contextual influence in trolley-problem-style moral triage: for each demographic factor, we apply matched, direction-flipped contextual influences that differ only in which group they favor, enabling systematic measurement of directional response. We find that: (i) contextual influences often significantly shift decisions, even when only superficially relevant; (ii) baseline preferences are a poor predictor of directional steerability, as models can appear baseline-neutral yet exhibit systematic steerability asymmetry under influence; (iii) influences can backfire: models may explicitly claim neutrality or discount the contextual cue, yet their choices still shift, sometimes in the opposite direction; and (iv) reasoning reduces average sensitivity, but amplifies the effect of biased few-shot examples. Our findings motivate extending moral evaluations with controlled, direction-flipped context manipulations to better characterize model behavior.

Background: Retrieval augmented generation (RAG) technology can empower large language models (LLMs) to generate more accurate, professional, and timely responses without fine tuning. However, due to the complex reasoning processes and substantial individual differences involved in traditional Chinese medicine (TCM) clinical diagnosis and treatment, traditional RAG methods often exhibit poor performance in this domain. Objective: To address the limitations of conventional RAG approaches in TCM applications, this study aims to develop an improved RAG framework tailored to the characteristics of TCM reasoning. Methods: We developed TCM-DiffRAG, an innovative RAG framework that integrates knowledge graphs (KG) with chains of thought (CoT). TCM-DiffRAG was evaluated on three distinctive TCM test datasets. Results: The experimental results demonstrated that TCM-DiffRAG achieved significant performance improvements over native LLMs. For example, the qwen-plus model achieved scores of 0.927, 0.361, and 0.038, which were significantly enhanced to 0.952, 0.788, and 0.356 with TCM-DiffRAG. The improvements were even more pronounced for non-Chinese LLMs. Additionally, TCM-DiffRAG outperformed directly supervised fine-tuned (SFT) LLMs and other benchmark RAG methods. Conclusions: TCM-DiffRAG shows that integrating structured TCM knowledge graphs with Chain of Thought based reasoning substantially improves performance in individualized diagnostic tasks. The joint use of universal and personalized knowledge graphs enables effective alignment between general knowledge and clinical reasoning. These results highlight the potential of reasoning-aware RAG frameworks for advancing LLM applications in traditional Chinese medicine.

This paper presents a comprehensive evaluation framework for assessing the cultural competence of large language models (LLMs) in Persian. Existing Persian cultural benchmarks rely predominantly on multiple-choice formats and English-centric metrics that fail to capture Persian's morphological complexity and semantic nuance. Our framework introduces a Persian-specific short-answer evaluation that combines rule-based morphological normalization with a hybrid syntactic and semantic similarity module, enabling robust soft-match scoring beyond exact string overlap. Through systematic evaluation of 15 state-of-the-art open- and closed-source models, we demonstrate that our hybrid evaluation improves scoring consistency by +10% compared to exact-match baselines by capturing meaning that surface-level methods cannot detect. We publicly release our evaluation framework, providing the first standardized benchmark for measuring cultural understanding in Persian and establishing a reproducible foundation for cross-cultural LLM evaluation research.

Functional data clustering is concerned with grouping functions that share similar structure, yet most existing methods implicitly operate on sampled grids, causing cluster assignments to depend on resolution, sampling density, or preprocessing choices rather than on the underlying functions themselves. To address this limitation, we introduce a framework that maps discretized function observations -- at arbitrary resolution and on arbitrary grids -- into a fixed-dimensional vector space via an auto-encoding architecture. The encoder is a hypernetwork that maps coordinate-value pairs to the weight space of an implicit neural representation (INR), which serves as the decoder. Because INRs represent functions with very few parameters, this design yields compact representations that are decoupled from the sampling grid, while the hypernetwork amortizes weight prediction across the dataset. Clustering is then performed in this weight space using standard algorithms, making the approach agnostic to both the discretization and the choice of clustering method. By means of synthetic and real-world experiments in high-dimensional settings, we demonstrate competitive clustering performance that is robust to changes in sampling resolution -- including generalization to resolutions not seen during training.

The identification and property prediction of chemical molecules is of central importance in the advancement of drug discovery and material science, where the tandem mass spectrometry technology gives valuable fragmentation cues in the form of mass-to-charge ratio peaks. However, the lack of experimental spectra hinders the attachment of each molecular identification, and thus urges the establishment of prediction approaches for computational models. Deep learning models appear promising for predicting molecular structure spectra, but overall assessment remains challenging as a result of the heterogeneity in methods and the lack of well-defined benchmarks. To address this, our contribution is the creation of benchmark framework FlexMS for constructing and evaluating diverse model architectures in mass spectrum prediction. With its easy-to-use flexibility, FlexMS supports the dynamic construction of numerous distinct combinations of model architectures, while assessing their performance on preprocessed public datasets using different metrics. In this paper, we provide insights into factors influencing performance, including the structural diversity of datasets, hyperparameters like learning rate and data sparsity, pretraining effects, metadata ablation settings and cross-domain transfer learning analysis. This provides practical guidance in choosing suitable models. Moreover, retrieval benchmarks simulate practical identification scenarios and score potential matches based on predicted spectra.

Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchy-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historical contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints. Code is available at https://github.com/langfengQ/verl-agent/tree/master/recipe/hgpo.

Agentic AI increasingly intervenes proactively by inferring users' situations from contextual data yet often fails for lack of principled judgment about when, why, and whether to act. We address this gap by proposing a conceptual model that reframes behavior as an interpretive outcome integrating Scene (observable situation), Context (user-constructed meaning), and Human Behavior Factors (determinants shaping behavioral likelihood). Grounded in multidisciplinary perspectives across the humanities, social sciences, HCI, and engineering, the model separates what is observable from what is meaningful to the user and explains how the same scene can yield different behavioral meanings and outcomes. To translate this lens into design action, we derive five agent design principles (behavioral alignment, contextual sensitivity, temporal appropriateness, motivational calibration, and agency preservation) that guide intervention depth, timing, intensity, and restraint. Together, the model and principles provide a foundation for designing agentic AI systems that act with contextual sensitivity and judgment in interactions.

Since local LLM inference on resource-constrained edge devices imposes a severe performance bottleneck, this paper proposes distributed prompt caching to enhance inference performance by cooperatively sharing intermediate processing states across multiple low-end edge devices. To fully utilize prompt similarity, our distributed caching mechanism also supports partial matching. As this approach introduces communication overhead associated with state sharing over a wireless network, we introduce a Bloom-filter-based data structure, referred to as a catalog, to determine whether a remote server possesses the desired internal states, thereby suppressing unnecessary communication. Experiments using the Gemma-3 270M model and the MMLU dataset on the Raspberry Pi Zero 2W platform demonstrate that the proposed approach reduces TTFT (Time to First Token) and TTLT (Time to Last Token) by 93.12% and 50.07% on average, respectively.

In this work, we present the first theoretical analysis of multi-agent imitation learning (MAIL) in linear Markov games where both the transition dynamics and each agent's reward function are linear in some given features. We demonstrate that by leveraging this structure, it is possible to replace the state-action level "all policy deviation concentrability coefficient" (Freihaut et al., arXiv:2510.09325) with a concentrability coefficient defined at the feature level which can be much smaller than the state-action analog when the features are informative about states' similarity. Furthermore, to circumvent the need for any concentrability coefficient, we turn to the interactive setting. We provide the first, computationally efficient, interactive MAIL algorithm for linear Markov games and show that its sample complexity depends only on the dimension of the feature map $d$. Building on these theoretical findings, we propose a deep MAIL interactive algorithm which clearly outperforms BC on games such as Tic-Tac-Toe and Connect4.

Despite the remarkable progress of large language models (LLMs), the capabilities of standalone LLMs have begun to plateau when tackling real-world, complex tasks that require interaction with external tools and dynamic environments. Although recent agent frameworks aim to enhance model autonomy through tool integration and external interaction, they still suffer from naive workflows, unstable performance, limited support across diverse benchmarks and tasks, and heavy reliance on costly commercial APIs. In this work, we propose a high-performance and robust open-source agent framework, termed MiroFlow, which incorporates an agent graph for flexible orchestration, an optional deep reasoning mode to enhance performance, and a robust workflow execution to ensure stable and reproducible performance. Extensive experiments demonstrate that MiroFlow consistently achieves state-of-the-art performance across multiple agent benchmarks, including GAIA, BrowseComp-EN/ZH, HLE, xBench-DeepSearch, and notably FutureX. We hope it could serve as an easily accessible, reproducible, and comparable baseline for the deep research community.

Diffusion models have become a popular choice for decision-making tasks in robotics, and more recently, are also being considered for solving autonomous driving tasks. However, their applications and evaluations in autonomous driving remain limited to simulation-based or laboratory settings. The full strength of diffusion models for large-scale, complex real-world settings, such as End-to-End Autonomous Driving (E2E AD), remains underexplored. In this study, we conducted a systematic and large-scale investigation to unleash the potential of the diffusion models as planners for E2E AD, based on a tremendous amount of real-vehicle data and road testing. Through comprehensive and carefully controlled studies, we identify key insights into the diffusion loss space, trajectory representation, and data scaling that significantly impact E2E planning performance. Moreover, we also provide an effective reinforcement learning post-training strategy to further enhance the safety of the learned planner. The resulting diffusion-based learning framework, Hyper Diffusion Planner} (HDP), is deployed on a real-vehicle platform and evaluated across 6 urban driving scenarios and 200 km of real-world testing, achieving a notable 10x performance improvement over the base model. Our work demonstrates that diffusion models, when properly designed and trained, can serve as effective and scalable E2E AD planners for complex, real-world autonomous driving tasks.

Internet of Things (IoT) networks face significant challenges such as limited communication bandwidth, constrained computational and energy resources, and highly dynamic wireless channel conditions. Utilization of deep neural networks (DNNs) combined with semantic communication has emerged as a promising paradigm to address these limitations. Deep joint source-channel coding (DJSCC) has recently been proposed to enable semantic communication of images. Building upon the original DJSCC formulation, low-complexity attention-style architectures has been added to the DNNs for further performance enhancement. As a main hurdle, training these DNNs separately for various signal-to-noise ratios (SNRs) will amount to excessive storage or communication overhead, which can not be maintained by small IoT devices. SNR Adaptive DJSCC (ADJSCC), has been proposed to train the DNNs once but feed the current SNR as part of the data to the channel-wise attention mechanism. We improve upon ADJSCC by a simultaneous utilization of doubly adaptive channel-wise and spatial attention modules at both transmitter and receiver. These modules dynamically adjust to varying channel conditions and spatial feature importance, enabling robust and efficient feature extraction and semantic information recovery. Simulation results corroborate that our proposed doubly adaptive DJSCC (DA-DJSCC) significantly improves upon ADJSCC in several performance criteria, while incurring a mild increase in complexity. These facts render DA-DJSCC a desirable choice for semantic communication in performance demanding but low-complexity IoT networks.

The rapid evolution of large language models (LLMs) has transformed prompt engineering from a localized craft into a systems-level governance challenge. As models scale and update across generations, prompt behavior becomes sensitive to shifts in instruction-following policies, alignment regimes, and decoding strategies, a phenomenon we characterize as GPT-scale model drift. Under such conditions, surface-level formatting conventions and ad hoc refinement are insufficient to ensure stable, interpretable control. This paper reconceptualizes Natural Language Declarative Prompting (NLD-P) as a declarative governance method rather than a rigid field template. NLD-P is formalized as a modular control abstraction that separates provenance, constraint logic, task content, and post-generation evaluation, encoded directly in natural language without reliance on external orchestration code. We define minimal compliance criteria, analyze model-dependent schema receptivity, and position NLD-P as an accessible governance framework for non-developer practitioners operating within evolving LLM ecosystems. Portions of drafting and editorial refinement employed a schema-bound LLM assistant configured under NLD-P. All conceptual framing, methodological claims, and final revisions were directed, reviewed, and approved by the human author under a documented human-in-the-loop protocol. The paper concludes by outlining implications for declarative control under ongoing model evolution and identifying directions for future empirical validation.

Large language models (LLMs) often generate fluent but unfounded claims, or hallucinations, which fall into two types: (i) faithfulness violations - misusing user context - and (ii) factuality violations - errors from internal knowledge. Proper mitigation depends on knowing whether a model's answer is based on the prompt or its internal weights. This work focuses on the problem of contributive attribution: identifying the dominant knowledge source behind each output. We show that a probe, a simple linear classifier trained on model hidden representations, can reliably predict contributive attribution. For its training, we introduce AttriWiki, a self-supervised data pipeline that prompts models to recall withheld entities from memory or read them from context, generating labelled examples automatically. Probes trained on AttriWiki data reveal a strong attribution signal, achieving up to 0.96 Macro-F1 on Llama-3.1-8B, Mistral-7B, and Qwen-7B, transferring to out-of-domain benchmarks (SQuAD, WebQuestions) with 0.94-0.99 Macro-F1 without retraining. Attribution mismatches raise error rates by up to 70%, demonstrating a direct link between knowledge source confusion and unfaithful answers. Yet, models may still respond incorrectly even when attribution is correct, highlighting the need for broader detection frameworks.

Value decomposition (VD) methods have achieved remarkable success in cooperative multi-agent reinforcement learning (MARL). However, their reliance on the max operator for temporal-difference (TD) target calculation leads to systematic Q-value overestimation. This issue is particularly severe in MARL due to the combinatorial explosion of the joint action space, which often results in unstable learning and suboptimal policies. To address this problem, we propose QSIM, a similarity weighted Q-learning framework that reconstructs the TD target using action similarity. Instead of using the greedy joint action directly, QSIM forms a similarity weighted expectation over a structured near-greedy joint action space. This formulation allows the target to integrate Q-values from diverse yet behaviorally related actions while assigning greater influence to those that are more similar to the greedy choice. By smoothing the target with structurally relevant alternatives, QSIM effectively mitigates overestimation and improves learning stability. Extensive experiments demonstrate that QSIM can be seamlessly integrated with various VD methods, consistently yielding superior performance and stability compared to the original algorithms. Furthermore, empirical analysis confirms that QSIM significantly mitigates the systematic value overestimation in MARL. Code is available at https://github.com/MaoMaoLYJ/pymarl-qsim.

Predictive modeling on web-scale tabular data with billions of instances and hundreds of heterogeneous numerical features faces significant scalability challenges. These features exhibit anisotropy, heavy-tailed distributions, and non-stationarity, creating bottlenecks for models like Gradient Boosting Decision Trees and requiring laborious manual feature engineering. We introduce KMLP, a hybrid deep architecture integrating a shallow Kolmogorov-Arnold Network (KAN) front-end with a Gated Multilayer Perceptron (gMLP) backbone. The KAN front-end uses learnable activation functions to automatically model complex non-linear transformations for each feature, while the gMLP backbone captures high-order interactions. Experiments on public benchmarks and an industrial dataset with billions of samples show KMLP achieves state-of-the-art performance, with advantages over baselines like GBDTs increasing at larger scales, validating KMLP as a scalable deep learning paradigm for large-scale web tabular data.

As mental health chatbots proliferate to address the global treatment gap, a critical question emerges: How do we design for relational safety the quality of interaction patterns that unfold across conversations rather than the correctness of individual responses? Current safety evaluations assess single-turn crisis responses, missing the therapeutic dynamics that determine whether chatbots help or harm over time. We introduce TherapyProbe, a design probe methodology that generates actionable design knowledge by systematically exploring chatbot conversation trajectories through adversarial multi-agent simulation. Using open-source models, TherapyProbe surfaces relational safety failures interaction patterns like "validation spirals" where chatbots progressively reinforce hopelessness, or "empathy fatigue" where responses become mechanical over turns. Our contribution is translating these failures into a Safety Pattern Library of 23 failure archetypes with corresponding design recommendations. We contribute: (1) a replicable methodology requiring no API costs, (2) a clinically-grounded failure taxonomy, and (3) design implications for developers, clinicians, and policymakers.

Clinical decisions are often required under incomplete information. Clinical experts must identify whether available information is sufficient for judgment, as both premature conclusion and unnecessary abstention can compromise patient safety. To evaluate this capability of large language models (LLMs), we developed ClinDet-Bench, a benchmark based on clinical scoring systems that decomposes incomplete-information scenarios into determinable and undeterminable conditions. Identifying determinability requires considering all hypotheses about missing information, including unlikely ones, and verifying whether the conclusion holds across them. We find that recent LLMs fail to identify determinability under incomplete information, producing both premature judgments and excessive abstention, despite correctly explaining the underlying scoring knowledge and performing well under complete information. These findings suggest that existing benchmarks are insufficient to evaluate the safety of LLMs in clinical settings. ClinDet-Bench provides a framework for evaluating determinability recognition, leading to appropriate abstention, with potential applicability to medicine and other high-stakes domains, and is publicly available.

Large Language Models (LLMs) are deployed as autonomous agents in increasingly complex applications, where enabling long-horizon memory is critical for achieving strong performance. However, a significant gap exists between practical applications and current evaluation standards for agent memory: existing benchmarks primarily focus on dialogue-centric, human-agent interactions. In reality, agent memory consists of a continuous stream of agent-environment interactions that are primarily composed of machine-generated representations. To bridge this gap, we introduce AMA-Bench (Agent Memory with Any length), which evaluates long-horizon memory for LLMs in real agentic applications. It features two key components: (1) a set of real-world agentic trajectories across representative agentic applications, paired with expert-curated QA, and (2) a set of synthetic agentic trajectories that scale to arbitrary horizons, paired with rule-based QA. Our comprehensive study shows that existing memory systems underperform on AMA-Bench primarily because they lack causality and objective information and are constrained by the lossy nature of similarity-based retrieval employed by many memory systems. To address these limitations, we propose AMA-Agent, an effective memory system featuring a causality graph and tool-augmented retrieval. Our results demonstrate that AMA-Agent achieves 57.22% average accuracy on AMA-Bench, surpassing the strongest memory system baselines by 11.16%.

Latent visual reasoning aims to mimic human's imagination process by meditating through hidden states of Multimodal Large Language Models. While recognized as a promising paradigm for visual reasoning, the underlying mechanisms driving its effectiveness remain unclear. Motivated to demystify the true source of its efficacy, we investigate the validity of latent reasoning using Causal Mediation Analysis. We model the process as a causal chain: the input as the treatment, the latent tokens as the mediator, and the final answer as the outcome. Our findings uncover two critical disconnections: (a) Input-Latent Disconnect: dramatic perturbations on the input result in negligible changes to the latent tokens, suggesting that latent tokens do not effectively attend to the input sequence. (b) Latent-Answer Disconnect: perturbations on the latent tokens yield minimal impact on the final answer, indicating the limited causal effect latent tokens imposing on the outcome. Furthermore, extensive probing analysis reveals that latent tokens encode limited visual information and exhibit high similarity. Consequently, we challenge the necessity of latent reasoning and propose a straightforward alternative named CapImagine, which teaches the model to explicitly imagine using text. Experiments on vision-centric benchmarks show that CapImagine significantly outperforms complex latent-space baselines, highlighting the superior potential of visual reasoning through explicit imagination.

Reinforcement Learning (RL) has empowered Large Language Models (LLMs) with strong reasoning capabilities, but vanilla RL mainly focuses on generation capability improvement by training with only first-order rollout (generating multiple responses for a question), and we argue that this approach fails to fully exploit the potential of training data because of the neglect of critique capability training. To tackle this problem, we further introduce the concept of second-order rollout (generating multiple critiques for a response) and propose a unified framework for jointly training generation and critique capabilities. Extensive experiments across various models and datasets demonstrate that our approach can utilize training data more effectively than vanilla RL and achieve better performance under the same training data. Additionally, we uncover several insightful findings regarding second-order rollout and critique training, such as the importance of label balance in critique training and the noise problem of outcome-based rewards, which can be mitigated through sampling techniques. Our work offers a preliminary exploration of dynamic data augmentation and joint generation-critique training in RL, providing meaningful inspiration for the further advancement of RL training

The Rust programming language presents a steep learning curve and significant coding challenges, making the automation of issue resolution essential for its broader adoption. Recently, LLM-powered code agents have shown remarkable success in resolving complex software engineering tasks, yet their application to Rust has been limited by the absence of a large-scale, repository-level benchmark. To bridge this gap, we introduce Rust-SWE-bench, a benchmark comprising 500 real-world, repository-level software engineering tasks from 34 diverse and popular Rust repositories. We then perform a comprehensive study on Rust-SWE-bench with four representative agents and four state-of-the-art LLMs to establish a foundational understanding of their capabilities and limitations in the Rust ecosystem. Our extensive study reveals that while ReAct-style agents are promising, i.e., resolving up to 21.2% of issues, they are limited by two primary challenges: comprehending repository-wide code structure and complying with Rust's strict type and trait semantics. We also find that issue reproduction is rather critical for task resolution. Inspired by these findings, we propose RUSTFORGER, a novel agentic approach that integrates an automated test environment setup with a Rust metaprogramming-driven dynamic tracing strategy to facilitate reliable issue reproduction and dynamic analysis. The evaluation shows that RUSTFORGER using Claude-Sonnet-3.7 significantly outperforms all baselines, resolving 28.6% of tasks on Rust-SWE-bench, i.e., a 34.9% improvement over the strongest baseline, and, in aggregate, uniquely solves 46 tasks that no other agent could solve across all adopted advanced LLMs.

Training large language models (LLMs) requires substantial compute and energy. At the same time, renewable energy sources regularly produce more electricity than the grid can absorb, leading to curtailment, the deliberate reduction of clean generation that would otherwise go to waste. These periods represent an opportunity: if training is aligned with curtailment windows, LLMs can be pretrained using electricity that is both clean and cheap. This technical report presents a system that performs full-parameter LLM training across geo-distributed GPU clusters during regional curtailment windows, elastically switching between local single-site training and federated multi-site synchronization as sites become available or unavailable. Our prototype trains a 561M-parameter transformer model across three clusters using the Flower federated learning framework, with curtailment periods derived from real-world marginal carbon intensity traces. Preliminary results show that curtailment-aware scheduling preserves training quality while reducing operational emissions to 5-12% of single-site baselines.

We decompose physician disagreement in the HealthBench medical AI evaluation dataset to understand where variance resides and what observable features can explain it. Rubric identity accounts for 15.8% of met/not-met label variance but only 3.6-6.9% of disagreement variance; physician identity accounts for just 2.4%. The dominant 81.8% case-level residual is not reduced by HealthBench's metadata labels (z = -0.22, p = 0.83), normative rubric language (pseudo R^2 = 1.2%), medical specialty (0/300 Tukey pairs significant), surface-feature triage (AUC = 0.58), or embeddings (AUC = 0.485). Disagreement follows an inverted-U with completion quality (AUC = 0.689), confirming physicians agree on clearly good or bad outputs but split on borderline cases. Physician-validated uncertainty categories reveal that reducible uncertainty (missing context, ambiguous phrasing) more than doubles disagreement odds (OR = 2.55, p < 10^(-24)), while irreducible uncertainty (genuine medical ambiguity) has no effect (OR = 1.01, p = 0.90), though even the former explains only ~3% of total variance. The agreement ceiling in medical AI evaluation is thus largely structural, but the reducible/irreducible dissociation suggests that closing information gaps in evaluation scenarios could lower disagreement where inherent clinical ambiguity does not, pointing toward actionable evaluation design improvements.

We introduce AuditBench, an alignment auditing benchmark. AuditBench consists of 56 language models with implanted hidden behaviors. Each model has one of 14 concerning behaviors--such as sycophantic deference, opposition to AI regulation, or secret geopolitical loyalties--which it does not confess to when directly asked. AuditBench models are highly diverse--some are subtle, while others are overt, and we use varying training techniques both for implanting behaviors and training models not to confess. To demonstrate AuditBench's utility, we develop an investigator agent that autonomously employs a configurable set of auditing tools. By measuring investigator agent success using different tools, we can evaluate their efficacy. Notably, we observe a tool-to-agent gap, where tools that perform well in standalone non-agentic evaluations fail to translate into improved performance when used with our investigator agent. We find that our most effective tools involve scaffolded calls to auxiliary models that generate diverse prompts for the target. White-box interpretability tools can be helpful, but the agent performs best with black-box tools. We also find that audit success varies greatly across training techniques: models trained on synthetic documents are easier to audit than models trained on demonstrations, with better adversarial training further increasing auditing difficulty. We release our models, agent, and evaluation framework to support future quantitative, iterative science on alignment auditing.

The transition of Large Language Models (LLMs) from exploratory tools to active "silicon subjects" in social science lacks extensive validation of operational validity. This study introduces Conditioned Comment Prediction (CCP), a task in which a model predicts how a user would comment on a given stimulus by comparing generated outputs with authentic digital traces. This framework enables a rigorous evaluation of current LLM capabilities with respect to the simulation of social media user behavior. We evaluated open-weight 8B models (Llama3.1, Qwen3, Ministral) in English, German, and Luxembourgish language scenarios. By systematically comparing prompting strategies (explicit vs. implicit) and the impact of Supervised Fine-Tuning (SFT), we identify a critical form vs. content decoupling in low-resource settings: while SFT aligns the surface structure of the text output (length and syntax), it degrades semantic grounding. Furthermore, we demonstrate that explicit conditioning (generated biographies) becomes redundant under fine-tuning, as models successfully perform latent inference directly from behavioral histories. Our findings challenge current "naive prompting" paradigms and offer operational guidelines prioritizing authentic behavioral traces over descriptive personas for high-fidelity simulation.

Large reasoning models (LRMs) have emerged as a powerful paradigm for solving complex real-world tasks. In practice, these models are predominantly trained via Reinforcement Learning with Verifiable Rewards (RLVR), yet most existing outcome-only RLVR pipelines rely almost exclusively on a binary correctness signal and largely ignore the model's intrinsic uncertainty. We term this discrepancy the uncertainty-reward mismatch, under which high- and low-uncertainty solutions are treated equivalently, preventing the policy from "Know What You Know" and impeding the shift from optimizing for correct answers to optimizing effective reasoning paths. This limitation is especially critical in reasoning-centric tasks such as mathematics and question answering, where performance hinges on the quality of the model's internal reasoning process rather than mere memorization of final answers. To address this, we propose EGPO, a metacognitive entropy calibration framework that explicitly integrates intrinsic uncertainty into RLVR for enhancing LRMs. EGPO estimates per-sample uncertainty using a zero-overhead entropy proxy derived from token-level likelihoods and aligns it with extrinsic correctness through an asymmetric calibration mechanism that preserves correct reasoning while selectively regulating overconfident failures, thereby enabling stable and uncertainty-aware policy optimization. Moreover, EGPO recovers informative learning signals from otherwise degenerate group-based rollouts without modifying the verifier or reward definition. Extensive experiments across multiple benchmarks demonstrate that the proposed EGPO leads to substantial and consistent improvements in reasoning performance, establishing a principled path for advancing LRMs through metacognitive entropy calibration.

Epistemic uncertainty in neural networks is commonly modeled using two second-order paradigms: distribution-based representations, which rely on posterior parameter distributions, and set-based representations based on credal sets (convex sets of probability distributions). These frameworks are often regarded as fundamentally non-comparable due to differing semantics, assumptions, and evaluation practices, leaving their relative merits unclear. Empirical comparisons are further confounded by variations in the underlying predictive models. To clarify this issue, we present a controlled comparative study enabling principled, like-for-like evaluation of the two paradigms. Both representations are constructed from the same finite collection of predictive distributions generated by a shared neural network, isolating representational effects from predictive accuracy. Our study evaluates each representation through the lens of 3 uncertainty measures across 8 benchmarks, including selective prediction and out-of-distribution detection, spanning 6 underlying predictive models and 10 independent runs per configuration. Our results show that meaningful comparison between these seemingly non-comparable frameworks is both feasible and informative, providing insights into how second-order representation choices impact practical uncertainty-aware performance.

Recommendation model performance is intrinsically tied to the quality, volume, and relevance of their training data. To address common challenges like data sparsity and cold start, recent researchs have leveraged data from multiple auxiliary domains to enrich information within the target domain. However, inherent domain gaps can degrade the quality of mixed-domain data, leading to negative transfer and diminished model performance. Existing prevailing \emph{model-centric} paradigm -- which relies on complex, customized architectures -- struggles to capture the subtle, non-structural sequence dependencies across domains, leading to poor generalization and high demands on computational resources. To address these shortcomings, we propose \textsc{Taesar}, a \emph{data-centric} framework for \textbf{t}arget-\textbf{a}lign\textbf{e}d \textbf{s}equenti\textbf{a}l \textbf{r}egeneration, which employs a contrastive decoding mechanism to adaptively encode cross-domain context into target-domain sequences. It employs contrastive decoding to encode cross-domain context into target sequences, enabling standard models to learn intricate dependencies without complex fusion architectures. Experiments show \textsc{Taesar} outperforms model-centric solutions and generalizes to various sequential models. By generating enriched datasets, \textsc{Taesar} effectively combines the strengths of data- and model-centric paradigms. The code accompanying this paper is available at~ \textcolor{blue}{https://github.com/USTC-StarTeam/Taesar}.

Referring Image Segmentation (RIS) aims to segment an object in an image identified by a natural language expression. The paper introduces Alignment-Aware Masked Learning (AML), a training strategy to enhance RIS by explicitly estimating pixel-level vision-language alignment, filtering out poorly aligned regions during optimization, and focusing on trustworthy cues. This approach results in state-of-the-art performance on RefCOCO datasets and also enhances robustness to diverse descriptions and scenarios

Augmented reading systems aim to adapt text presentation to improve comprehension and task performance, yet existing approaches rely heavily on heuristics, opaque data-driven models, or repeated human involvement in the design loop. We propose framing augmented reading as a simulation-based optimization problem grounded in resource-rational models of human reading. These models instantiate a simulated reader that allocates limited cognitive resources, such as attention, memory, and time under task demands, enabling systematic evaluation of text user interfaces. We introduce two complementary optimization pipelines: an offline approach that explores design alternatives using simulated readers, and an online approach that personalizes reading interfaces in real time using ongoing interaction data. Together, this perspective enables adaptive, explainable, and scalable augmented reading design without relying solely on human testing.

Generative recommendation has recently attracted widespread attention in industry due to its potential for scaling and stronger model capacity. However, deploying real-time generative recommendation in large-scale advertising requires designs beyond large-language-model (LLM)-style training and serving recipes. We present a production-oriented generative recommender co-designed across architecture, learning, and serving, named GR4AD (Generative Recommendation for ADdvertising). As for tokenization, GR4AD proposes UA-SID (Unified Advertisement Semantic ID) to capture complicated business information. Furthermore, GR4AD introduces LazyAR, a lazy autoregressive decoder that relaxes layer-wise dependencies for short, multi-candidate generation, preserving effectiveness while reducing inference cost, which facilitates scaling under fixed serving budgets. To align optimization with business value, GR4AD employs VSL (Value-Aware Supervised Learning) and proposes RSPO (Ranking-Guided Softmax Preference Optimization), a ranking-aware, list-wise reinforcement learning algorithm that optimizes value-based rewards under list-level metrics for continual online updates. For online inference, we further propose dynamic beam serving, which adapts beam width across generation levels and online load to control compute. Large-scale online A/B tests show up to 4.2% ad revenue improvement over an existing DLRM-based stack, with consistent gains from both model scaling and inference-time scaling. GR4AD has been fully deployed in Kuaishou advertising system with over 400 million users and achieves high-throughput real-time serving.

This paper introduces a novel Czech dataset in the restaurant domain for aspect-based sentiment analysis (ABSA), enriched with annotations of opinion terms. The dataset supports three distinct ABSA tasks involving opinion terms, accommodating varying levels of complexity. Leveraging this dataset, we conduct extensive experiments using modern Transformer-based models, including large language models (LLMs), in monolingual, cross-lingual, and multilingual settings. To address cross-lingual challenges, we propose a translation and label alignment methodology leveraging LLMs, which yields consistent improvements. Our results highlight the strengths and limitations of state-of-the-art models, especially when handling the linguistic intricacies of low-resource languages like Czech. A detailed error analysis reveals key challenges, including the detection of subtle opinion terms and nuanced sentiment expressions. The dataset establishes a new benchmark for Czech ABSA, and our proposed translation-alignment approach offers a scalable solution for adapting ABSA resources to other low-resource languages.

Seed explosion is a fundamental problem in fuzzing seed scheduling, where a fuzzer maintains a huge corpus and fails to choose promising seeds. Existing works focus on seed prioritization but still suffer from seed explosion since corpus size remains huge. We tackle this from a new perspective: corpus reduction, i.e., computing a seed corpus subset. However, corpus reduction could lead to poor seed diversity and large runtime overhead. Prior techniques like cull_queue, AFL-Cmin, and MinSet suffer from poor diversity or prohibitive overhead, making them unsuitable for high-frequency seed scheduling. We propose RandSet, a novel randomized corpus reduction technique that reduces corpus size and yields diverse seed selection simultaneously with minimal overhead. Our key insight is introducing randomness into corpus reduction to enjoy two benefits of a randomized algorithm: randomized output (diverse seed selection) and low runtime cost. Specifically, we formulate corpus reduction as a set cover problem and compute a randomized subset covering all features of the entire corpus. We then schedule seeds from this small, randomized subset rather than the entire corpus, effectively mitigating seed explosion. We implement RandSet on three popular fuzzers: AFL++, LibAFL, and Centipede, and evaluate it on standalone programs, FuzzBench, and Magma. Results show RandSet achieves significantly more diverse seed selection than other reduction techniques, with average subset ratios of 4.03% and 5.99% on standalone and FuzzBench programs. RandSet achieves a 16.58% coverage gain on standalone programs and up to 3.57% on FuzzBench in AFL++, triggers up to 7 more ground-truth bugs than the state-of-the-art on Magma, while introducing only 1.17%-3.93% overhead.

Large language model (LLM) agents increasingly rely on external tools and retrieval systems to autonomously complete complex tasks. However, this design exposes agents to indirect prompt injection (IPI), where attacker-controlled context embedded in tool outputs or retrieved content silently steers agent actions away from user intent. Unlike prompt-based attacks, IPI unfolds over multi-turn trajectories, making malicious control difficult to disentangle from legitimate task execution. Existing inference-time defenses primarily rely on heuristic detection and conservative blocking of high-risk actions, which can prematurely terminate workflows or broadly suppress tool usage under ambiguous multi-turn scenarios. We propose AgentSentry, a novel inference-time detection and mitigation framework for tool-augmented LLM agents. To the best of our knowledge, AgentSentry is the first inference-time defense to model multi-turn IPI as a temporal causal takeover. It localizes takeover points via controlled counterfactual re-executions at tool-return boundaries and enables safe continuation through causally guided context purification that removes attack-induced deviations while preserving task-relevant evidence. We evaluate AgentSentry on the \textsc{AgentDojo} benchmark across four task suites, three IPI attack families, and multiple black-box LLMs. AgentSentry eliminates successful attacks and maintains strong utility under attack, achieving an average Utility Under Attack (UA) of 74.55 %, improving UA by 20.8 to 33.6 percentage points over the strongest baselines without degrading benign performance.

There is growing recognition that many NLP tasks lack a single ground truth, as human judgments reflect diverse perspectives. To capture this variation, models have been developed to predict full annotation distributions rather than majority labels, while perspectivist models aim to reproduce the interpretations of individual annotators. In this work, we compare these approaches on Implicit Discourse Relation Recognition (IDRR), a highly ambiguous task where disagreement often arises from cognitive complexity rather than ideological bias. Our experiments show that existing annotator-specific models perform poorly in IDRR unless ambiguity is reduced, whereas models trained on label distributions yield more stable predictions. Further analysis indicates that frequent cognitively demanding cases drive inconsistency in human interpretation, posing challenges for perspectivist modeling in IDRR.

Table Question Answering (TQA) aims to answer natural language questions over structured tables. Large Language Models (LLMs) enable promising solutions to this problem, with operator-centric solutions that generate table manipulation pipelines in a multi-step manner offering state-of-the-art performance. However, these solutions rely on multiple LLM calls, resulting in prohibitive latencies and computational costs. We propose Operation-R1, the first framework that trains lightweight LLMs (e.g., Qwen-4B/1.7B) via a novel variant of reinforcement learning with verifiable rewards to produce high-quality data-preparation pipelines for TQA in a single inference step. To train such an LLM, we first introduce a self-supervised rewarding mechanism to automatically obtain fine-grained pipeline-wise supervision signals for LLM training. We also propose variance-aware group resampling to mitigate training instability. To further enhance robustness of pipeline generation, we develop two complementary mechanisms: operation merge, which filters spurious operations through multi-candidate consensus, and adaptive rollback, which offers runtime protection against information loss in data transformation. Experiments on two benchmark datasets show that, with the same LLM backbone, Operation-R1 achieves average absolute accuracy gains of 9.55 and 6.08 percentage points over multi-step preparation baselines, with 79\% table compression and a 2.2$\times$ reduction in monetary cost.

State-space models (SSMs) have emerged as an efficient strategy for building powerful language models, avoiding the quadratic complexity of computing attention in transformers. Despite their promise, the interpretability and steerability of modern SSMs remain relatively underexplored. We take a major step in this direction by identifying activation subspace bottlenecks in the Mamba family of SSM models using tools from mechanistic interpretability. We then introduce a test-time steering intervention that simply multiplies the activations of the identified bottlenecks by a scalar. Across 5 SSMs and 6 diverse benchmarks, this intervention improves performance by an average of 8.27%, without requiring any task-specific tuning. Finally, we validate that the identified bottlenecks are indeed hindering performance by modifying them to yield an architecture we call Stable-Mamba, which achieves long-context performance gains when retrained from scratch.

Reinforcement Learning from Human Feedback (RLHF) has been widely applied to Large Language Model (LLM) post-training to align model outputs with human preferences. Recent models, such as DeepSeek-R1, have also shown RLHF's potential to improve LLM reasoning on complex tasks. In RL, inference and training co-exist, creating dynamic resource demands throughout the workflow. Compared to traditional RL, RLHF further challenges training efficiency due to expanding model sizes and resource consumption. Several RLHF frameworks aim to balance flexible abstraction and efficient execution. However, they rely on serverful infrastructures, which struggle with fine-grained resource variability. As a result, during synchronous RLHF training, idle time between or within RL components often causes overhead and resource wastage. To address these issues, we present RLHFless, the first scalable training framework for synchronous RLHF, built on serverless computing environments. RLHFless adapts to dynamic resource demands throughout the RLHF pipeline, pre-computes shared prefixes to avoid repeated computation, and uses a cost-aware actor scaling strategy that accounts for response length variation to find sweet spots with lower cost and higher speed. In addition, RLHFless assigns workloads efficiently to reduce intra-function imbalance and idle time. Experiments on both physical testbeds and a large-scale simulated cluster show that RLHFless achieves up to 1.35x speedup and 44.8% cost reduction compared to the state-of-the-art baseline.

3D Large Vision-Language Models (3D LVLMs) built upon Large Language Models (LLMs) have achieved remarkable progress across various multimodal tasks. However, their inherited position-dependent modeling mechanism, Rotary Position Embedding (RoPE), remains suboptimal for 3D multimodal understanding. The vanilla RoPE formulation fails to preserve essential three-dimensional spatial structures when encoding 3D tokens, and its relative distance computation overlooks angular dependencies, hindering the model's ability to capture directional variations in visual representations. To overcome these limitations, we introduce Spherical Coordinate-based Positional Embedding (SoPE). Our method maps point-cloud token indices into a 3D spherical coordinate space, enabling unified modeling of spatial locations and directional angles. This formulation preserves the inherent geometric structure of point-cloud data, enhances spatial awareness, and yields more consistent and expressive geometric representations for multimodal learning. In addition, we introduce a multi-scale frequency mixing strategy to fuse feature information across different frequency domains. Experimental results on multiple 3D scene benchmarks validate the effectiveness of our approach, while real-world deployment experiments further demonstrate its strong generalization capability.

Preference-based reinforcement learning (PbRL) is the dominant framework for aligning AI systems to human preferences, but its application to speech remains underexplored. We present a controlled cross-modal study of human and synthetic preference annotations, comparing text and audio evaluations of identical semantic content across 100 prompts. Audio preferences prove as reliable as text, with inter-rater agreement reaching good levels (ICC(2,k) $\approx$ .80) at $\sim$9 raters -- the first ICC-based reliability characterization in the preference annotation literature for either modality. However, modality reshapes how people judge: audio raters exhibit narrower decision thresholds, reduced length bias, and more user-oriented evaluation criteria, with near-chance cross-modality agreement. Synthetic ratings further align with human judgments and predict inter-rater agreement, supporting their use both for triaging ambiguous pairs and as full replacements for human annotations.

Vision-language models (VLMs) often struggle with geometric reasoning due to their limited perception of fundamental diagram elements. To tackle this challenge, we introduce GeoPerceive, a benchmark comprising diagram instances paired with domain-specific language (DSL) representations, along with an efficient automatic data generation pipeline. This design enables the isolated evaluation of geometric perception independently from reasoning. To exploit the data provided by GeoPerceive for enhancing the geometric perception capabilities of VLMs, we propose GeoDPO, a translator-guided reinforcement learning (RL) framework. GeoDPO employs an NL-to-DSL translator, which is trained on synthetic pairs generated by the data engine of GeoPerceive, to bridge natural language and DSL. This translator facilitates the computation of fine-grained, DSL-level scores, which serve as reward signals in reinforcement learning. We assess GeoDPO on both in-domain and out-of-domain datasets, spanning tasks in geometric perception as well as downstream reasoning. Experimental results demonstrate that, while supervised fine-tuning (SFT) offers only marginal improvements and may even impair performance in out-of-domain scenarios, GeoDPO achieves substantial gains: $+26.5\%$ on in-domain data, $+8.0\%$ on out-of-domain data, and $+39.0\%$ on downstream reasoning tasks. These findings underscore the superior performance and generalization ability of GeoDPO over SFT. All codes are released at https://github.com/Longin-Yu/GeoPerceive to ensure reproducibility.

Existing neural network calibration methods often treat calibration as a static, post-hoc optimization task. However, this neglects the dynamic and temporal nature of real-world inference. Moreover, existing methods do not provide an intuitive interface enabling human operators to dynamically adjust model behavior under shifting conditions. In this work, we propose Knob, a framework that connects deep learning with classical control theory by mapping neural gating dynamics to a second-order mechanical system. By establishing correspondences between physical parameters -- damping ratio ($ζ$) and natural frequency ($ω_n$) -- and neural gating, we create a tunable "safety valve". The core mechanism employs a logit-level convex fusion, functioning as an input-adaptive temperature scaling. It tends to reduce model confidence particularly when model branches produce conflicting predictions. Furthermore, by imposing second-order dynamics (Knob-ODE), we enable a \textit{dual-mode} inference: standard i.i.d. processing for static tasks, and state-preserving processing for continuous streams. Our framework allows operators to tune "stability" and "sensitivity" through familiar physical analogues. This paper presents an exploratory architectural interface; we focus on demonstrating the concept and validating its control-theoretic properties rather than claiming state-of-the-art calibration performance. Experiments on CIFAR-10-C validate the calibration mechanism and demonstrate that, in Continuous Mode, the gate responses are consistent with standard second-order control signatures (step settling and low-pass attenuation), paving the way for predictable human-in-the-loop tuning.

Commercial large language models are typically deployed as black-box API services, requiring users to trust providers to execute inference correctly and report token usage honestly. We present IMMACULATE, a practical auditing framework that detects economically motivated deviations-such as model substitution, quantization abuse, and token overbilling-without trusted hardware or access to model internals. IMMACULATE selectively audits a small fraction of requests using verifiable computation, achieving strong detection guarantees while amortizing cryptographic overhead. Experiments on dense and MoE models show that IMMACULATE reliably distinguishes benign and malicious executions with under 1% throughput overhead. Our code is published at https://github.com/guo-yanpei/Immaculate.

SQL is the de facto interface for exploratory data analysis; however, releasing exact query results can expose sensitive information through membership or attribute inference attacks. Differential privacy (DP) provides rigorous privacy guarantees, but in practice, DP alone may not satisfy governance requirements such as the \emph{minimum frequency rule}, which requires each released group (cell) to include contributions from at least $k$ distinct individuals. In this paper, we present \textbf{DPSQL+}, a privacy-preserving SQL library that simultaneously enforces user-level $(\varepsilon,δ)$-DP and the minimum frequency rule. DPSQL+ adopts a modular architecture consisting of: (i) a \emph{Validator} that statically restricts queries to a DP-safe subset of SQL; (ii) an \emph{Accountant} that consistently tracks cumulative privacy loss across multiple queries; and (iii) a \emph{Backend} that interfaces with various database engines, ensuring portability and extensibility. Experiments on the TPC-H benchmark demonstrate that DPSQL+ achieves practical accuracy across a wide range of analytical workloads -- from basic aggregates to quadratic statistics and join operations -- and allows substantially more queries under a fixed global privacy budget than prior libraries in our evaluation.

Leveraging Large Language Models (LLMs) for Knowledge Graph Completion (KGC) is promising but hindered by a fundamental granularity mismatch. LLMs operate on fragmented token sequences, whereas entities are the fundamental units in knowledge graphs (KGs) scenarios. Existing approaches typically constrain predictions to limited candidate sets or align entities with the LLM's vocabulary by pooling multiple tokens or decomposing entities into fixed-length token sequences, which fail to capture both the semantic meaning of the text and the structural integrity of the graph. To address this, we propose KGT, a novel framework that uses dedicated entity tokens to enable efficient, full-space prediction. Specifically, we first introduce specialized tokenization to construct feature representations at the level of dedicated entity tokens. We then fuse pre-trained structural and textual features into these unified embeddings via a relation-guided gating mechanism, avoiding training from scratch. Finally, we implement decoupled prediction by leveraging independent heads to separate and combine semantic and structural reasoning. Experimental results show that KGT consistently outperforms state-of-the-art methods across multiple benchmarks.

The rapid evolution of Large Language Models (LLMs) has accelerated the transition from conversational chatbots to general agents. However, effectively balancing empathetic communication with budget-aware decision-making remains an open challenge. Since existing methods fail to capture these complex strategic trade-offs, we propose InteractCS-RL, a framework that reframes task-oriented dialogue as a multi-granularity reinforcement learning process. Specifically, we first establish a User-centric Interaction Framework to provide a high-fidelity training gym, enabling agents to dynamically explore diverse strategies with persona-driven users. Then, we introduce Cost-aware Multi-turn Policy Optimization (CMPO) with a hybrid advantage estimation strategy. By integrating generative process credits and employing a PID-Lagrangian cost controller, CMPO effectively guides the policy to explore Pareto boundary between user reward and global cost constraints. Extensive experiments on customized real business scenarios demonstrate that InteractCS-RL significantly outperform other baselines across three evaluation dimensions. Further evaluation on tool-agent-user interaction benchmarks verify InteractCS-RL robustness across diverse domains.

Current approaches to developing persuasive dialogue agents often rely on a limited set of predefined persuasive strategies that fail to capture the complexity of real-world interactions. We applied a cross-disciplinary approach to develop a framework for designing persuasive dialogue agents that draws on proven strategies from social psychology, behavioral economics, and communication theory. We validated our proposed framework through experiments on two distinct datasets: the Persuasion for Good dataset, which represents a specific in-domain scenario, and the DailyPersuasion dataset, which encompasses a wide range of scenarios. The proposed framework achieved strong results for both datasets and demonstrated notable improvement in the persuasion success rate as well as promising generalizability. Notably, the proposed framework also excelled at persuading individuals with initially low intent, which addresses a critical challenge for persuasive dialogue agents.

Intermittent demand, a pattern characterized by long sequences of zero sales punctuated by sporadic, non-zero values, poses a persistent challenge in retail and supply chain forecasting. Both traditional methods, such as ARIMA, exponential smoothing, or Croston variants, as well as modern neural architectures such as DeepAR and Transformer-based models often underperform on such data, as they treat demand as a single continuous process or become computationally expensive when scaled across many sparse series. To address these limitations, we introduce Switch-Hurdle: a new framework that integrates a Mixture-of-Experts (MoE) encoder with a Hurdle-based probabilistic decoder. The encoder uses a sparse Top-1 expert routing during the forward pass yet approximately dense in the backward pass via a straight-through estimator (STE). The decoder follows a cross-attention autoregressive design with a shared hurdle head that explicitly separates the forecasting task into two components: a binary classification component estimating the probability of a sale, and a conditional regression component, predicting the quantity given a sale. This structured separation enables the model to capture both occurrence and magnitude processes inherent to intermittent demand. Empirical results on the M5 benchmark and a large proprietary retail dataset show that Switch-Hurdle achieves state-of-the-art prediction performance while maintaining scalability.

The rapid advancement of AI-powered smart glasses, one of the hottest wearable devices, has unlocked new frontiers for multimodal interaction, with Visual Question Answering (VQA) over external knowledge sources emerging as a core application. Existing Vision Language Models (VLMs) adapted to smart glasses are typically trained and evaluated on traditional multimodal datasets; however, these datasets lack the variety and realism needed to reflect smart glasses usage scenarios and diverge from their specific challenges, where accurately identifying the object of interest must precede any external knowledge retrieval. To bridge this gap, we introduce SUPERGLASSES, the first comprehensive VQA benchmark built on real-world data entirely collected by smart glasses devices. SUPERGLASSES comprises 2,422 egocentric image-question pairs spanning 14 image domains and 8 query categories, enriched with full search trajectories and reasoning annotations. We evaluate 26 representative VLMs on this benchmark, revealing significant performance gaps. To address the limitations of existing models, we further propose SUPERLENS, a multimodal smart glasses agent that enables retrieval-augmented answer generation by integrating automatic object detection, query decoupling, and multimodal web search. Our agent achieves state-of-the-art performance, surpassing GPT-4o by 2.19 percent, and highlights the need for task-specific solutions in smart glasses VQA scenarios.

Pre-training Large Language Models requires immense computational resources, making optimizer efficiency essential. The optimization landscape is highly anisotropic, with loss reduction driven predominantly by progress along flat directions. While matrix-based optimizers such as Muon and SOAP leverage fine-grained curvature information to outperform AdamW, their updates tend toward isotropy -- relatively conservative along flat directions yet potentially aggressive along sharp ones. To address this limitation, we first establish a unified Riemannian Ordinary Differential Equation (ODE) framework that elucidates how common adaptive algorithms operate synergistically: the preconditioner induces a Riemannian geometry that mitigates ill-conditioning, while momentum serves as a Riemannian damping term that promotes convergence. Guided by these insights, we propose LITE, a generalized acceleration strategy that enhances training dynamics by applying larger Hessian damping coefficients and learning rates along flat trajectories. Extensive experiments demonstrate that LITE significantly accelerates both Muon and SOAP across diverse architectures (Dense, MoE), parameter scales (130M--1.3B), datasets (C4, Pile), and learning-rate schedules (cosine, warmup-stable-decay). Theoretical analysis confirms that LITE facilitates faster convergence along flat directions in anisotropic landscapes, providing a principled approach to efficient LLM pre-training. The code is available at https://github.com/SHUCHENZHU/LITE.

Large language models have enabled agents that reason, plan, and interact with tools and environments to accomplish complex tasks. As these agents operate over extended interaction horizons, their effectiveness increasingly depends on adapting behavior to individual users and maintaining continuity across time, giving rise to personalized LLM-powered agents. In such long-term, user-dependent settings, personalization permeates the entire decision pipeline rather than remaining confined to surface-level generation. This survey provides a capability-oriented review of personalized LLM-powered agents. We organize the literature around four interdependent components: profile modeling, memory, planning, and action execution. Using this taxonomy, we synthesize representative methods and analyze how user signals are represented, propagated, and utilized, highlighting cross-component interactions and recurring design trade-offs. We further examine evaluation metrics and benchmarks tailored to personalized agents, summarize application scenarios spanning general assistance to specialized domains, and outline future directions for research and deployment. By offering a structured framework for understanding and designing personalized LLM-powered agents, this survey charts a roadmap toward more user-aligned, adaptive, robust, and deployable agentic systems, accelerating progress from prototype personalization to scalable real-world assistants.

Image-text retrieval has become a fundamental component in intelligent multimedia systems; however, most existing vision-language models are optimized for highresource languages and remain suboptimal for low-resource settings such as Vietnamese. This work introduces ViCLIP-OT, a foundation vision-language model specifically designed for Vietnamese image-text retrieval. The proposed framework integrates CLIP-style contrastive learning with a Similarity-Graph Regularized Optimal Transport (SIGROT) loss to enhance global cross-modal consistency and mitigate modality gap issues. Extensive experiments on three Vietnamese benchmarks (UITOpenViIC, KTVIC, and Crossmodal-3600) demonstrate that ViCLIP-OT consistently outperforms CLIP and SigLIP baselines in both in-domain and zero-shot settings. On UIT-OpenViIC, the model achieves an average Recall@K of 67.34%, improving upon CLIP by 5.75 percentage points. In zero-shot evaluation on Crossmodal-3600, ViCLIPOT surpasses CLIP by 11.72 percentage points. Embedding-space analysis further confirms improved alignment and reduced modality gap. The results indicate that integrating SIGROT provides an effective and scalable strategy for cross-modal retrieval in low-resource languages, offering practical implications for intelligent multimedia retrieval systems in Vietnamese and other underrepresented linguistic contexts.

Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-intensive scenarios. Moreover, generalization across heterogeneous research settings remains challenging. In this work, we propose \emph{Search More, Think Less} (SMTL), a framework for long-horizon agentic search that targets both efficiency and generalization. SMTL replaces sequential reasoning with parallel evidence acquisition, enabling efficient context management under constrained context budgets. To support generalization across task types, we further introduce a unified data synthesis pipeline that constructs search tasks spanning both deterministic question answering and open-ended research scenarios with task appropriate evaluation metrics. We train an end-to-end agent using supervised fine-tuning and reinforcement learning, achieving strong and often state of the art performance across benchmarks including BrowseComp (48.6\%), GAIA (75.7\%), Xbench (82.0\%), and DeepResearch Bench (45.9\%). Compared to Mirothinker-v1.0, SMTL with maximum 100 interaction steps reduces the average number of reasoning steps on BrowseComp by 70.7\%, while improving accuracy.

Antimicrobial resistance (AMR) is a growing global crisis projected to cause 10 million deaths per year by 2050. While the WHO Global Antimicrobial Resistance and Use Surveillance System (GLASS) provides standardized surveillance data across 44 countries, few studies have applied machine learning to forecast population-level resistance trends from this data. This paper presents a two-component framework for AMR trend forecasting and evidence-grounded policy decision support. We benchmark six models -- Naive, Linear Regression, Ridge Regression, XGBoost, LightGBM, and LSTM -- on 5,909 WHO GLASS observations across six WHO regions (2021-2023). XGBoost achieved the best performance with a test MAE of 7.07% and R-squared of 0.854, outperforming the naive baseline by 83.1%. Feature importance analysis identified the prior-year resistance rate as the dominant predictor (50.5% importance), while regional MAE ranged from 4.16% (European Region) to 10.14% (South-East Asia Region). We additionally implemented a Retrieval-Augmented Generation (RAG) pipeline combining a ChromaDB vector store of WHO policy documents with a locally deployed Phi-3 Mini language model, producing source-attributed, hallucination-constrained policy answers. Code and data are available at https://github.com/TanvirTurja

Although diffusion language models (DLMs) are evolving quickly, many recent models converge on a set of shared components. These components, however, are distributed across ad-hoc research codebases or lack transparent implementations, making them difficult to reproduce or extend. As the field accelerates, there is a clear need for a unified framework that standardizes these common components while remaining flexible enough to support new methods and architectures. To address this gap, we introduce dLLM, an open-source framework that unifies the core components of diffusion language modeling -- training, inference, and evaluation -- and makes them easy to customize for new designs. With dLLM, users can reproduce, finetune, deploy, and evaluate open-source large DLMs such as LLaDA and Dream through a standardized pipeline. The framework also provides minimal, reproducible recipes for building small DLMs from scratch with accessible compute, including converting any BERT-style encoder or autoregressive LM into a DLM. We also release the checkpoints of these small DLMs to make DLMs more accessible and accelerate future research.

Recent advances in generic large models, such as GPT and DeepSeek, have motivated the introduction of universality to graph pre-training, aiming to learn rich and generalizable knowledge across diverse domains using graph representations to improve performance in various downstream applications. However, most existing methods face challenges in learning effective knowledge from generic graphs, primarily due to simplistic data alignment and limited training guidance. The issue of simplistic data alignment arises from the use of a straightforward unification for highly diverse graph data, which fails to align semantics and misleads pre-training models. The problem with limited training guidance lies in the arbitrary application of in-domain pre-training paradigms to cross-domain scenarios. While it is effective in enhancing discriminative representation in one data space, it struggles to capture effective knowledge from many graphs. To address these challenges, we propose a novel Latent sEmantic Distribution Alignment (LEDA) model for universal graph pre-training. Specifically, we first introduce a dimension projection unit to adaptively align diverse domain features into a shared semantic space with minimal information loss. Furthermore, we design a variational semantic inference module to obtain the shared latent distribution. The distribution is then adopted to guide the domain projection, aligning it with shared semantics across domains and ensuring cross-domain semantic learning. LEDA exhibits strong performance across a broad range of graphs and downstream tasks. Remarkably, in few-shot cross-domain settings, it significantly outperforms in-domain baselines and advanced universal pre-training models.

Deepfake speech utterances can be forged by replacing one or more words in a bona fide utterance with semantically different words synthesized by speech generative models. While a dedicated synthetic word detector could be developed, we investigate a cost-effective method that fine-tunes a pre-trained Whisper model to detect synthetic words while transcribing the input utterance via next-token prediction. We further investigate using partially vocoded utterances as the fine-tuning data, thereby reducing the cost of data collection. Our experiments demonstrate that, on in-domain test data, the fine-tuned Whisper yields low synthetic-word detection error rates and transcription error rates. On out-of-domain test data with synthetic words produced by unseen speech generative models, the fine-tuned Whisper remains on par with a dedicated ResNet-based detection model; however, the overall performance degradation calls for strategies to improve its generalization capability.

In online advertising, the inherent complexity and dynamic nature of advertising environments necessitate the use of auto-bidding services to assist advertisers in bid optimization. This complexity is further compounded in multi-channel scenarios, where effective allocation of budgets and constraints across channels with distinct behavioral patterns becomes critical for optimizing return on investment. Current approaches predominantly rely on either optimization-based strategies or reinforcement learning techniques. However, optimization-based methods lack flexibility in adapting to dynamic market conditions, while reinforcement learning approaches often struggle to capture essential historical dependencies and observational patterns within the constraints of Markov Decision Process frameworks. To address these limitations, we propose AHBid, an Adaptable Hierarchical Bidding framework that integrates generative planning with real-time control. The framework employs a high-level generative planner based on diffusion models to dynamically allocate budgets and constraints by effectively capturing historical context and temporal patterns. We introduce a constraint enforcement mechanism to ensure compliance with specified constraints, along with a trajectory refinement mechanism that enhances adaptability to environmental changes through the utilization of historical data. The system further incorporates a control-based bidding algorithm that synergistically combines historical knowledge with real-time information, significantly improving both adaptability and operational efficacy. Extensive experiments conducted on large-scale offline datasets and through online A/B tests demonstrate the effectiveness of AHBid, yielding a 13.57% increase in overall return compared to existing baselines.

Generative retrieval has emerged as a powerful paradigm for LLM-based recommendation. However, industrial recommender systems often benefit from restricting the output space to a constrained subset of items based on business logic (e.g. enforcing content freshness or product category), which standard autoregressive decoding cannot natively support. Moreover, existing constrained decoding methods that make use of prefix trees (Tries) incur severe latency penalties on hardware accelerators (TPUs/GPUs). In this work, we introduce STATIC (Sparse Transition Matrix-Accelerated Trie Index for Constrained Decoding), an efficient and scalable constrained decoding technique designed specifically for high-throughput LLM-based generative retrieval on TPUs/GPUs. By flattening the prefix tree into a static Compressed Sparse Row (CSR) matrix, we transform irregular tree traversals into fully vectorized sparse matrix operations, unlocking massive efficiency gains on hardware accelerators. We deploy STATIC on a large-scale industrial video recommendation platform serving billions of users. STATIC produces significant product metric impact with minimal latency overhead (0.033 ms per step and 0.25% of inference time), achieving a 948x speedup over a CPU trie implementation and a 47-1033x speedup over a hardware-accelerated binary-search baseline. Furthermore, the runtime overhead of STATIC remains extremely low across a wide range of practical configurations. To the best of our knowledge, STATIC enables the first production-scale deployment of strictly constrained generative retrieval. In addition, evaluation on academic benchmarks demonstrates that STATIC can considerably improve cold-start performance for generative retrieval. Our code is available at https://github.com/youtube/static-constraint-decoding.

Universal graph pre-training has emerged as a key paradigm in graph representation learning, offering a promising way to train encoders to learn transferable representations from unlabeled graphs and to effectively generalize across a wide range of downstream tasks. However, recent explorations in universal graph pre-training primarily focus on homogeneous graphs and it remains unexplored for heterogeneous graphs, which exhibit greater structural and semantic complexity. This heterogeneity makes it highly challenging to train a universal encoder for diverse heterogeneous graphs: (i) the diverse types with dataset-specific semantics hinder the construction of a unified representation space; (ii) the number and semantics of meta-paths vary across datasets, making encoding and aggregation patterns learned from one dataset difficult to apply to others. To address these challenges, we propose a novel Meta-path-aware Universal heterogeneous Graph pre-training (MUG) approach. Specifically, for challenge (i), MUG introduces a input unification module that integrates information from multiple node and relation types within each heterogeneous graph into a unified representation.This representation is then projected into a shared space by a dimension-aware encoder, enabling alignment across graphs with diverse schemas.Furthermore, for challenge (ii), MUG trains a shared encoder to capture consistent structural patterns across diverse meta-path views rather than relying on dataset-specific aggregation strategies, while a global objective encourages discriminability and reduces dataset-specific biases. Extensive experiments demonstrate the effectiveness of MUG on some real datasets.

Chain-of-Thought (CoT) has substantially empowered Large Language Models (LLMs) to tackle complex reasoning tasks, yet the verbose nature of explicit reasoning steps incurs prohibitive inference latency and computational costs, limiting real-world deployment. While existing compression methods - ranging from self-training to Reinforcement Learning (RL) with length constraints - attempt to mitigate this, they often sacrifice reasoning capability for brevity. We identify a critical failure mode in these approaches: explicitly optimizing for shorter trajectories triggers rapid entropy collapse, which prematurely shrinks the exploration space and stifles the discovery of valid reasoning paths, particularly for challenging questions requiring extensive deduction. To address this issue, we propose Compress responses for Easy questions and Explore Hard ones (CEEH), a difficulty-aware approach to RL-based efficient reasoning. CEEH dynamically assesses instance difficulty to apply selective entropy regularization: it preserves a diverse search space for currently hard questions to ensure robustness, while permitting aggressive compression on easier instances where the reasoning path is well-established. In addition, we introduce a dynamic optimal-length penalty anchored to the historically shortest correct response, which effectively counteracts entropy-induced length inflation and stabilizes the reward signal. Across six reasoning benchmarks, CEEH consistently reduces response length while maintaining accuracy comparable to the base model, and improves Pass@k relative to length-only optimization.

Route-planning agents powered by large language models (LLMs) have emerged as a promising paradigm for supporting everyday human mobility through natural language interaction and tool-mediated decision making. However, systematic evaluation in real-world mobility settings is hindered by diverse routing demands, non-deterministic mapping services, and limited reproducibility. In this study, we introduce MobilityBench, a scalable benchmark for evaluating LLM-based route-planning agents in real-world mobility scenarios. MobilityBench is constructed from large-scale, anonymized real user queries collected from Amap and covers a broad spectrum of route-planning intents across multiple cities worldwide. To enable reproducible, end-to-end evaluation, we design a deterministic API-replay sandbox that eliminates environmental variance from live services. We further propose a multi-dimensional evaluation protocol centered on outcome validity, complemented by assessments of instruction understanding, planning, tool use, and efficiency. Using MobilityBench, we evaluate multiple LLM-based route-planning agents across diverse real-world mobility scenarios and provide an in-depth analysis of their behaviors and performance. Our findings reveal that current models perform competently on Basic information retrieval and Route Planning tasks, yet struggle considerably with Preference-Constrained Route Planning, underscoring significant room for improvement in personalized mobility applications. We publicly release the benchmark data, evaluation toolkit, and documentation at https://github.com/AMAP-ML/MobilityBench .

Differentially private federated learning (DP-FL) enables clients to collaboratively train machine learning models while preserving the privacy of their local data. However, most existing DP-FL approaches assume that all clients share a uniform privacy budget, an assumption that does not hold in real-world scenarios where privacy requirements vary widely. This privacy heterogeneity poses a significant challenge: conventional client selection strategies, which typically rely on data quantity, cannot distinguish between clients providing high-quality updates and those introducing substantial noise due to strict privacy constraints. To address this gap, we present the first systematic study of privacy-aware client selection in DP-FL. We establish a theoretical foundation by deriving a convergence analysis that quantifies the impact of privacy heterogeneity on training error. Building on this analysis, we propose a privacy-aware client selection strategy, formulated as a convex optimization problem, that adaptively adjusts selection probabilities to minimize training error. Extensive experiments on benchmark datasets demonstrate that our approach achieves up to a 10% improvement in test accuracy on CIFAR-10 compared to existing baselines under heterogeneous privacy budgets. These results highlight the importance of incorporating privacy heterogeneity into client selection for practical and effective federated learning.

Neural networks are increasingly deployed in safety- and mission-critical pipelines, yet many verification and analysis results are produced outside the programming environment that defines and runs the model. This separation creates a semantic gap between the executed network and the analyzed artifact, so guarantees can hinge on implicit conventions such as operator semantics, tensor layouts, preprocessing, and floating-point corner cases. We introduce TorchLean, a framework in the Lean 4 theorem prover that treats learned models as first-class mathematical objects with a single, precise semantics shared by execution and verification. TorchLean unifies (1) a PyTorch-style verified API with eager and compiled modes that lower to a shared op-tagged SSA/DAG computation-graph IR, (2) explicit Float32 semantics via an executable IEEE-754 binary32 kernel and proof-relevant rounding models, and (3) verification via IBP and CROWN/LiRPA-style bound propagation with certificate checking. We validate TorchLean end-to-end on certified robustness, physics-informed residual bounds for PINNs, and Lyapunov-style neural controller verification, alongside mechanized theoretical results including a universal approximation theorem. These results demonstrate a semantics-first infrastructure for fully formal, end-to-end verification of learning-enabled systems.

This paper proposes HyperKKL, a novel learning approach for designing Kazantzis-Kravaris/Luenberger (KKL) observers for non-autonomous nonlinear systems. While KKL observers offer a rigorous theoretical framework by immersing nonlinear dynamics into a stable linear latent space, its practical realization relies on solving Partial Differential Equations (PDE) that are analytically intractable. Current existing learning-based approximations of the KKL observer are mostly designed for autonomous systems, failing to generalize to driven dynamics without expensive retraining or online gradient updates. HyperKKL addresses this by employing a hypernetwork architecture that encodes the exogenous input signal to instantaneously generate the parameters of the KKL observer, effectively learning a family of immersion maps parameterized by the external drive. We rigorously evaluate this approach against a curriculum learning strategy that attempts to generalize from autonomous regimes via training heuristics alone. The novel approach is illustrated on four numerical simulations in benchmark examples including the Duffing, Van der Pol, Lorenz, and Rössler systems.

Editing images via instruction provides a natural way to generate interactive content, but it is a big challenge due to the higher requirement of scene understanding and generation. Prior work utilizes a chain of large language models, object segmentation models, and editing models for this task. However, the understanding models provide only a single modality ability, restricting the editing quality. We aim to bridge understanding and generation via a new multi-modality model that provides the intelligent abilities to instruction-based image editing models for more complex cases. To achieve this goal, we individually separate the instruction editing task with the multi-modality chain of thought prompts, i.e., Chain-of-Thought (CoT) planning, editing region reasoning, and editing. For Chain-of-Thought planning, the large language model could reason the appropriate sub-prompts considering the instruction provided and the ability of the editing network. For editing region reasoning, we train an instruction-based editing region generation network with a multi-modal large language model. Finally, a hint-guided instruction-based editing network is proposed for editing image generations based on the sizeable text-to-image diffusion model to accept the hints for generation. Extensive experiments demonstrate that our method has competitive editing abilities on complex real-world images.

We propose ContextRL, a novel framework that leverages context augmentation to overcome these bottlenecks. Specifically, to enhance Identifiability, we provide the reward model with full reference solutions as context, enabling fine-grained process verification to filter out false positives (samples with the right answer but low-quality reasoning process). To improve Reachability, we introduce a multi-turn sampling strategy where the reward model generates mistake reports for failed attempts, guiding the policy to "recover" correct responses from previously all-negative groups. Experimental results on 11 perception and reasoning benchmarks show that ContextRL significantly improves knowledge discovery efficiency. Notably, ContextRL enables the Qwen3-VL-8B model to achieve performance comparable to the 32B model, outperforming standard RLVR baselines by a large margin while effectively mitigating reward hacking. Our in-depth analysis reveals the significant potential of contextual information for improving reward model accuracy and document the widespread occurrence of reward hacking, offering valuable insights for future RLVR research.

Source-Free Domain Adaptive Object Detection (SF-DAOD) aims to adapt a detector trained on a labeled source domain to an unlabeled target domain without retaining any source data. Despite recent progress, most popular approaches focus on tuning pseudo-label thresholds or refining the teacher-student framework, while overlooking object-level structural cues within cross-domain data. In this work, we present CGSA, the first framework that brings Object-Centric Learning (OCL) into SF-DAOD by integrating slot-aware adaptation into the DETR-based detector. Specifically, our approach integrates a Hierarchical Slot Awareness (HSA) module into the detector to progressively disentangle images into slot representations that act as visual priors. These slots are then guided toward class semantics via a Class-Guided Slot Contrast (CGSC) module, maintaining semantic consistency and prompting domain-invariant adaptation. Extensive experiments on multiple cross-domain datasets demonstrate that our approach outperforms previous SF-DAOD methods, with theoretical derivations and experimental analysis further demonstrating the effectiveness of the proposed components and the framework, thereby indicating the promise of object-centric design in privacy-sensitive adaptation scenarios. Code is released at https://github.com/Michael-McQueen/CGSA.

Virtual beam diagnostics relies on computationally intensive beam dynamics simulations where high-dimensional charged particle beams evolve through the accelerator. We propose Latent Evolution Model (LEM), a hybrid machine learning framework with an autoencoder that projects high-dimensional phase spaces into lower-dimensional representations, coupled with transformers to learn temporal dynamics in the latent space. This approach provides a common foundational framework addressing multiple interconnected challenges in beam diagnostics. For \textit{forward modeling}, a Conditional Variational Autoencoder (CVAE) encodes 15 unique projections of the 6D phase space into a latent representation, while a transformer predicts downstream latent states from upstream inputs. For \textit{inverse problems}, we address two distinct challenges: (a) predicting upstream phase spaces from downstream observations by utilizing the same CVAE architecture with transformers trained on reversed temporal sequences along with aleatoric uncertainty quantification, and (b) estimating RF settings from the latent space of the trained LEM using a dedicated dense neural network that maps latent representations to RF parameters. For \textit{tuning problems}, we leverage the trained LEM and RF estimator within a Bayesian optimization framework to determine optimal RF settings that minimize beam loss. This paper summarizes our recent efforts and demonstrates how this unified approach effectively addresses these traditionally separate challenges.

Large Language Models (LLMs) obey consistent scaling laws -- empirical power-law fits that predict how loss decreases with compute, data, and parameters. While predictive, these laws are descriptive rather than prescriptive: they characterize typical training, not optimal training. Surprisingly few works have successfully challenged the data-efficiency bounds implied by these laws -- which is our primary focus. To that end, we introduce the Geodesic Hypothesis, positing that token sequences trace geodesics on a smooth semantic manifold and are therefore locally linear. Building on this principle, we propose a novel Semantic Tube Prediction (STP) task, a JEPA-style regularizer that confines hidden-state trajectories to a tubular neighborhood of the geodesic. STP generalizes JEPA to language without requiring explicit multi-view augmentations. We show this constraint improves signal-to-noise ratio, and consequently preserves diversity by preventing trajectory collisions during inference. Empirically, STP allows LLMs to match baseline accuracy with 16$\times$ less training data on the NL-RX-SYNTH dataset, directly violating the data term of Chinchilla-style scaling laws and demonstrating that principled geometric priors can surpass brute-force scaling. Code is available at https://github.com/galilai-group/llm-jepa#stp.

In Embedding-as-an-Interface (EaaI) settings, pre-trained models are queried for Intermediate Representations (IRs). The distributional properties of IRs can leak training-set membership signals, enabling Membership Inference Attacks (MIAs) whose strength varies across layers. Although Differentially Private Stochastic Gradient Descent (DP-SGD) mitigates such leakage, existing implementations employ per-example gradient clipping and a uniform, layer-agnostic noise multiplier, ignoring heterogeneous layer-wise MIA vulnerability. This paper introduces Layer-wise MIA-risk-aware DP-SGD (LM-DP-SGD), which adaptively allocates privacy protection across layers in proportion to their MIA risk. Specifically, LM-DP-SGD trains a shadow model on a public shadow dataset, extracts per-layer IRs from its train/test splits, and fits layer-specific MIA adversaries, using their attack error rates as MIA-risk estimates. Leveraging the cross-dataset transferability of MIAs, these estimates are then used to reweight each layer's contribution to the globally clipped gradient during private training, providing layer-appropriate protection under a fixed noise magnitude. We further establish theoretical guarantees on both privacy and convergence of LM-DP-SGD. Extensive experiments show that, under the same privacy budget, LM-DP-SGD reduces the peak IR-level MIA risk while preserving utility, yielding a superior privacy-utility trade-off.

Condition injection enables diffusion models to generate context-aware outputs, which is essential for many time-series tasks. However, heterogeneous conditional contexts (e.g., observed history, missingness patterns or outlier covariates) can induce heavy-tailed per-example gradients. Under Differentially Private Stochastic Gradient Descent (DP-SGD), these rare conditioning-driven heavy-tailed gradients disproportionately trigger global clipping, resulting in outlier-dominated updates, larger clipping bias, and degraded utility under a fixed privacy budget. In this paper, we propose DP-aware AdaLN-Zero, a drop-in sensitivity-aware conditioning mechanism for conditional diffusion transformers that limits conditioning-induced gain without modifying the DP-SGD mechanism. DP-aware AdaLN-Zero jointly constrains conditioning representation magnitude and AdaLN modulation parameters via bounded re-parameterization, suppressing extreme gradient tail events before gradient clipping and noise injection. Empirically, DP-SGD equipped with DP-aware AdaLN-Zero improves interpolation/imputation and forecasting under matched privacy settings. We observe consistent gains on a real-world power dataset and two public ETT benchmarks over vanilla DP-SGD. Moreover, gradient diagnostics attribute these improvements to conditioning-specific tail reshaping and reduced clipping distortion, while preserving expressiveness in non-private training. Overall, these results show that sensitivity-aware conditioning can substantially improve private conditional diffusion training without sacrificing standard performance.

Progress in hardware model checking depends critically on high-quality benchmarks. However, the community faces a significant benchmark gap: existing suites are limited in number, often distributed only in representations such as BTOR2 without access to the originating register-transfer-level (RTL) designs, and biased toward extreme difficulty where instances are either trivial or intractable. These limitations hinder rigorous evaluation of new verification techniques and encourage overfitting of solver heuristics to a narrow set of problems. To address this, we introduce EvolveGen, a framework for generating hardware model checking benchmarks by combining reinforcement learning (RL) with high-level synthesis (HLS). Our approach operates at an algorithmic level of abstraction in which an RL agent learns to construct computation graphs. By compiling these graphs under different synthesis directives, we produce pairs of functionally equivalent but structurally distinct hardware designs, inducing challenging model checking instances. Solver runtime is used as the reward signal, enabling the agent to autonomously discover and generate small-but-hard instances that expose solver-specific weaknesses. Experiments show that EvolveGen efficiently creates a diverse benchmark set in standard formats (e.g., AIGER and BTOR2) and effectively reveals performance bottlenecks in state-of-the-art model checkers.

We propose CoLyricist, an AI-assisted lyric writing tool designed to support the typical workflows of experienced lyricists and enhance their creative efficiency. While lyricists have unique processes, many follow common stages. Tools that fail to accommodate these stages challenge integration into creative practices. Existing research and tools lack sufficient understanding of these songwriting stages and their associated challenges, resulting in ineffective designs. Through a formative study involving semi-structured interviews with 10 experienced lyricists, we identified four key stages: Theme Setting, Ideation, Drafting Lyrics, and Melody Fitting. CoLyricist addresses these needs by incorporating tailored AI-driven support for each stage, optimizing the lyric writing process to be more seamless and efficient. To examine whether this workflow-aligned design also benefits those without prior experience, we conducted a user study with 16 participants, including both experienced and novice lyricists. Results showed that CoLyricist enhances the songwriting experience across skill levels. Novice users especially appreciated the Melody-Fitting feature, while experienced users valued the Ideation support.

Long-running agentic tasks, such as deep research, require multi-hop reasoning over information distributed across multiple webpages and documents. In such tasks, the LLM context is dominated by tokens from external retrieval, causing memory usage to grow rapidly and limiting decode performance. While several KV cache compression techniques exist for long-context inputs, we find that existing heuristics fail to support multi-step reasoning models effectively. We address this challenge with SideQuest -- a novel approach that leverages the Large Reasoning Model (LRM) itself to perform KV cache compression by reasoning about the usefulness of tokens in its context. To prevent the tokens associated with this management process from polluting the model's memory, we frame KV cache compression as an auxiliary task executed in parallel to the main reasoning task. Our evaluations, using a model trained with just 215 samples, show that SideQuest reduces peak token usage by up to 65% on agentic tasks with minimal degradation in accuracy, outperforming heuristic-based KV cache compression techniques.

Large language models exhibit sophisticated capabilities, yet understanding how they work internally remains a central challenge. A fundamental obstacle is that training selects for behavior, not circuitry, so many weight configurations can implement the same function. Which internal structures reflect the computation, and which are accidents of a particular training run? This work extracts algorithmic cores: compact subspaces necessary and sufficient for task performance. Independently trained transformers learn different weights but converge to the same cores. Markov-chain transformers embed 3D cores in nearly orthogonal subspaces yet recover identical transition spectra. Modular-addition transformers discover compact cyclic operators at grokking that later inflate, yielding a predictive model of the memorization-to-generalization transition. GPT-2 language models govern subject-verb agreement through a single axis that, when flipped, inverts grammatical number throughout generation across scales. These results reveal low-dimensional invariants that persist across training runs and scales, suggesting that transformer computations are organized around compact, shared algorithmic structures. Mechanistic interpretability could benefit from targeting such invariants -- the computational essence -- rather than implementation-specific details.

We present BetterScene, an approach to enhance novel view synthesis (NVS) quality for diverse real-world scenes using extremely sparse, unconstrained photos. BetterScene leverages the production-ready Stable Video Diffusion (SVD) model pretrained on billions of frames as a strong backbone, aiming to mitigate artifacts and recover view-consistent details at inference time. Conventional methods have developed similar diffusion-based solutions to address these challenges of novel view synthesis. Despite significant improvements, these methods typically rely on off-the-shelf pretrained diffusion priors and fine-tune only the UNet module while keeping other components frozen, which still leads to inconsistent details and artifacts even when incorporating geometry-aware regularizations like depth or semantic conditions. To address this, we investigate the latent space of the diffusion model and introduce two components: (1) temporal equivariance regularization and (2) vision foundation model-aligned representation, both applied to the variational autoencoder (VAE) module within the SVD pipeline. BetterScene integrates a feed-forward 3D Gaussian Splatting (3DGS) model to render features as inputs for the SVD enhancer and generate continuous, artifact-free, consistent novel views. We evaluate on the challenging DL3DV-10K dataset and demonstrate superior performance compared to state-of-the-art methods.

Quantization-Aware Training from scratch has emerged as a promising approach for building efficient large language models (LLMs) with extremely low-bit weights (sub 2-bit), which can offer substantial advantages for edge deployment. However, existing methods still fail to achieve satisfactory accuracy and scalability. In this work, we identify a parameter democratization effect as a key bottleneck: the sensitivity of all parameters becomes homogenized, severely limiting expressivity. To address this, we propose pQuant, a method that decouples parameters by splitting linear layers into two specialized branches: a dominant 1-bit branch for efficient computation and a compact high-precision branch dedicated to preserving the most sensitive parameters. Through tailored feature scaling, we explicitly guide the model to allocate sensitive parameters to the high-precision branch. Furthermore, we extend this branch into multiple, sparsely-activated experts, enabling efficient capacity scaling. Extensive experiments indicate our pQuant achieves state-of-the-art performance in extremely low-bit quantization.

Synthetic tabular data generation has attracted growing attention due to its importance for data augmentation, foundation models, and privacy. However, real-world tabular datasets increasingly contain free-form text fields (e.g., reviews or clinical notes) alongside structured numerical and categorical attributes. Generating such heterogeneous tables with joint modeling of different modalities remains challenging. Existing approaches broadly fall into two categories: diffusion-based methods and LLM-based methods. Diffusion models can capture complex dependencies over numerical and categorical features in continuous or discrete spaces, but extending them to open-ended text is nontrivial and often leads to degraded text quality. In contrast, LLM-based generators naturally produce fluent text, yet their discrete tokenization can distort precise or wide-range numerical values, hindering accurate modeling of both numbers and language. In this work, we propose TabDLM, a unified framework for free-form tabular data generation via a joint numerical--language diffusion model built on masked diffusion language models (MDLMs). TabDLM models textual and categorical features through masked diffusion, while modeling numerical features with a continuous diffusion process through learned specialized numeric tokens embedding; bidirectional attention then captures cross-modality interactions within a single model. Extensive experiments on diverse benchmarks demonstrate the effectiveness of TabDLM compared to strong diffusion- and LLM-based baselines.

Human evaluations play a central role in training and assessing AI models, yet these data are rarely treated as measurements subject to systematic error. This paper integrates psychometric rater models into the AI pipeline to improve the reliability and validity of conclusions drawn from human judgments. The paper reviews common rater effects, severity and centrality, that distort observed ratings, and demonstrates how item response theory rater models, particularly the multi-faceted Rasch model, can separate true output quality from rater behavior. Using the OpenAI summarization dataset as an empirical example, we show how adjusting for rater severity produces corrected estimates of summary quality and provides diagnostic insight into rater performance. Incorporating psychometric modeling into human-in-the-loop evaluation offers more principled and transparent use of human data, enabling developers to make decisions based on adjusted scores rather than raw, error-prone ratings. This perspective highlights a path toward more robust, interpretable, and construct-aligned practices for AI development and evaluation.

Industrial advertising question answering (QA) is a high-stakes task in which hallucinated content, particularly fabricated URLs, can lead to financial loss, compliance violations, and legal risk. Although Retrieval-Augmented Generation (RAG) is widely adopted, deploying it in production remains challenging because industrial knowledge is inherently relational, frequently updated, and insufficiently aligned with generation objectives. We propose a reinforced co-adaptation framework that jointly optimizes retrieval and generation through two components: (1) Graph-aware Retrieval (GraphRAG), which models entity-relation structure over a high-citation knowledge subgraph for multi-hop, domain-specific evidence selection; and (2) evidence-constrained reinforcement learning via Group Relative Policy Optimization (GRPO) with multi-dimensional rewards covering faithfulness, style compliance, safety, and URL validity. Experiments on an internal advertising QA dataset show consistent gains across expert-judged dimensions including accuracy, completeness, and safety, while reducing the hallucination rate by 72\%. A two-week online A/B test demonstrates a 28.6\% increase in like rate, a 46.2\% decrease in dislike rate, and a 92.7\% reduction in URL hallucination. The system has been running in production for over half a year and has served millions of QA interactions.

Example-based guidance is widely used to improve mathematical reasoning at inference time, yet its effectiveness is highly unstable across problems and models-even when the guidance is correct and problem-relevant. We show that this instability arises from a previously underexplored gap between strategy usage-whether a reasoning strategy appears in successful solutions-and strategy executability-whether the strategy remains effective when instantiated as guidance for a target model. Through a controlled analysis of paired human-written and model-generated solutions, we identify a systematic dissociation between usage and executability: human- and model-derived strategies differ in structured, domain-dependent ways, leading to complementary strengths and consistent source-dependent reversals under guidance. Building on this diagnosis, we propose Selective Strategy Retrieval (SSR), a test-time framework that explicitly models executability by selectively retrieving and combining strategies using empirical, multi-route, source-aware signals. Across multiple mathematical reasoning benchmarks, SSR yields reliable and consistent improvements over direct solving, in-context learning, and single-source guidance, improving accuracy by up to $+13$ points on AIME25 and $+5$ points on Apex for compact reasoning models. Code and benchmark are publicly available at: https://github.com/lwd17/strategy-execute-pipeline.

Circuit discovery has recently attracted attention as a potential research direction to explain the non-trivial behaviors of language models. It aims to find the computational subgraphs, also known as circuits, within the model that are responsible for solving specific tasks. However, most existing studies overlook the holistic nature of these circuits and require designing specific corrupted activations for different tasks, which is inaccurate and inefficient. In this work, we propose an end-to-end approach based on the principle of Information Bottleneck, called IBCircuit, to identify informative circuits holistically. IBCircuit is an optimization framework for holistic circuit discovery and can be applied to any given task without tediously corrupted activation design. In both the Indirect Object Identification (IOI) and Greater-Than tasks, IBCircuit identifies more faithful and minimal circuits in terms of critical node components and edge components compared to recent related work.

Vision-Language-Action (VLA) models are multimodal robotic task controllers that, given an instruction and visual inputs, produce a sequence of low-level control actions (or motor commands) enabling a robot to execute the requested task in the physical environment. These systems face the test oracle problem from multiple perspectives. On the one hand, a test oracle must be defined for each instruction prompt, which is a complex and non-generalizable approach. On the other hand, current state-of-the-art oracles typically capture symbolic representations of the world (e.g., robot and object states), enabling the correctness evaluation of a task, but fail to assess other critical aspects, such as the quality with which VLA-enabled robots perform a task. In this paper, we explore whether Metamorphic Testing (MT) can alleviate the test oracle problem in this context. To do so, we propose two metamorphic relation patterns and five metamorphic relations to assess whether changes to the test inputs impact the original trajectory of the VLA-enabled robots. An empirical study involving five VLA models, two simulated robots, and four robotic tasks shows that MT can effectively alleviate the test oracle problem by automatically detecting diverse types of failures, including, but not limited to, uncompleted tasks. More importantly, the proposed MRs are generalizable, making the proposed approach applicable across different VLA models, robots, and tasks, even in the absence of test oracles.

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by incorporating external knowledge, yet traditional single-round retrieval struggles with complex multi-step reasoning. Agentic RAG addresses this by enabling LLMs to dynamically decide when and what to retrieve, but current RL-based training methods suffer from sparse outcome rewards that discard intermediate signals and low sample efficiency where failed samples contribute nothing. We propose Search-P1, a framework that introduces path-centric reward shaping for agentic RAG training, comprising two key components: (1) Path-Centric Reward, which evaluates the structural quality of reasoning trajectories through order-agnostic step coverage and soft scoring that extracts learning signals even from failed samples, and (2) Dual-Track Path Scoring with offline-generated reference planners that assesses paths from both self-consistency and reference-alignment perspectives. Experiments on multiple QA benchmarks demonstrate that Search-P1 achieves significant improvements over Search-R1 and other strong baselines, with an average accuracy gain of 7.7 points.

Attention scales quadratically with sequence length, fundamentally limiting long-context inference. Existing block-granularity sparsification can reduce latency, but coarse blocks impose an intrinsic sparsity ceiling, making further improvements difficult even with carefully engineered designs. We present S2O, which performs early stopping for sparse attention via online permutation. Inspired by virtual-to-physical address mapping in memory systems, S2O revisits and factorizes FlashAttention execution, enabling inference to load non-contiguous tokens rather than a contiguous span in the original order. Motivated by fine-grained structures in attention heatmaps, we transform explicit permutation into an online, index-guided, discrete loading policy; with extremely lightweight preprocessing and index-remapping overhead, it concentrates importance on a small set of high-priority blocks. Building on this importance-guided online permutation for loading, S2O further introduces an early-stopping rule: computation proceeds from high to low importance; once the current block score falls below a threshold, S2O terminates early and skips the remaining low-contribution blocks, thereby increasing effective sparsity and reducing computation under a controlled error budget. As a result, S2O substantially raises the practical sparsity ceiling. On Llama-3.1-8B under a 128K context, S2O reduces single-operator MSE by 3.82$\times$ at matched sparsity, and reduces prefill compute density by 3.31$\times$ at matched MSE; meanwhile, it preserves end-to-end accuracy and achieves 7.51$\times$ attention and 3.81$\times$ end-to-end speedups.

Classifier-free guidance (CFG) has helped diffusion models achieve great conditional generation in various fields. Recently, more diffusion guidance methods have emerged with improved generation quality and human preference. However, can these emerging diffusion guidance methods really achieve solid and significant improvements? In this paper, we rethink recent progress on diffusion guidance. Our work mainly consists of four contributions. First, we reveal a critical evaluation pitfall that common human preference models exhibit a strong bias towards large guidance scales. Simply increasing the CFG scale can easily improve quantitative evaluation scores due to strong semantic alignment, even if image quality is severely damaged (e.g., oversaturation and artifacts). Second, we introduce a novel guidance-aware evaluation (GA-Eval) framework that employs effective guidance scale calibration to enable fair comparison between current guidance methods and CFG by identifying the effects orthogonal and parallel to CFG effects. Third, motivated by the evaluation pitfall, we design Transcendent Diffusion Guidance (TDG) method that can significantly improve human preference scores in the conventional evaluation framework but actually does not work in practice. Fourth, in extensive experiments, we empirically evaluate recent eight diffusion guidance methods within the conventional evaluation framework and the proposed GA-Eval framework. Notably, simply increasing the CFG scales can compete with most studied diffusion guidance methods, while all methods suffer severely from winning rate degradation over standard CFG. Our work would strongly motivate the community to rethink the evaluation paradigm and future directions of this field.

Deep multi-view clustering has achieved remarkable progress but remains vulnerable to complex noise in real-world applications. Existing noisy robust methods predominantly rely on a simplified binary assumption, treating data as either perfectly clean or completely corrupted. This overlooks the prevalent existence of heterogeneous observation noise, where contamination intensity varies continuously across data. To bridge this gap, we propose a novel framework termed Quality-Aware Robust Multi-View Clustering (QARMVC). Specifically, QARMVC employs an information bottleneck mechanism to extract intrinsic semantics for view reconstruction. Leveraging the insight that noise disrupts semantic integrity and impedes reconstruction, we utilize the resulting reconstruction discrepancy to precisely quantify fine-grained contamination intensity and derive instance-level quality scores. These scores are integrated into a hierarchical learning strategy: at the feature level, a quality-weighted contrastive objective is designed to adaptively suppress the propagation of noise; at the fusion level, a high-quality global consensus is constructed via quality-weighted aggregation, which is subsequently utilized to align and rectify local views via mutual information maximization. Extensive experiments on five benchmark datasets demonstrate that QARMVC consistently outperforms state-of-the-art baselines, particularly in scenarios with heterogeneous noise intensities.

Mitigating climate change requires behaviour change. However, even climate-concerned individuals often hold misperceptions about which actions most reduce carbon emissions. We recruited 1201 climate-concerned individuals to examine whether discussing climate actions with a large language model (LLM) equipped with climate knowledge and prompted to provide personalised responses would foster more accurate perceptions of the impacts of climate actions and increase willingness to adopt feasible, high-impact behaviours. We compared this to having participants run a web search, have a conversation with an unspecialised LLM, and no intervention. The personalised climate LLM was the only condition that led to increased knowledge about the impacts of climate actions and greater intentions to adopt impactful behaviours. While the personalised climate LLM did not outperform a web search in improving understanding of climate action impacts, the ability of LLMs to deliver personalised, actionable guidance may make them more effective at motivating impactful pro-climate behaviour change.

The deployment of machine learning in high-stakes domains requires a balance between predictive safety and algorithmic fairness. However, existing fairness interventions often as- sume unconstrained resources and employ group-specific decision thresholds that violate anti- discrimination regulations. We introduce a post-hoc, model-agnostic threshold optimization framework that jointly balances safety, efficiency, and equity under strict and hard capacity constraints. To ensure legal compliance, the framework enforces a single, global decision thresh- old. We formulated a parameterized ethical loss function coupled with a bounded decision rule that mathematically prevents intervention volumes from exceeding the available resources. An- alytically, we prove the key properties of the deployed threshold, including local monotonicity with respect to ethical weighting and the formal identification of critical capacity regimes. We conducted extensive experimental evaluations on diverse high-stakes datasets. The principal re- sults demonstrate that capacity constraints dominate ethical priorities; the strict resource limit determines the final deployed threshold in over 80% of the tested configurations. Furthermore, under a restrictive 25% capacity limit, the proposed framework successfully maintains high risk identification (recall ranging from 0.409 to 0.702), whereas standard unconstrained fairness heuristics collapse to a near-zero utility. We conclude that theoretical fairness objectives must be explicitly subordinated to operational capacity limits to remain in deployment. By decou- pling predictive scoring from policy evaluation and strictly bounding intervention rates, this framework provides a practical and legally compliant mechanism for stakeholders to navigate unavoidable ethical trade-offs in resource-constrained environments.

Current safety mechanisms for Large Language Models (LLMs) rely heavily on static, fine-tuned classifiers that suffer from adaptation rigidity, the inability to enforce new governance rules without expensive retraining. To address this, we introduce CourtGuard, a retrieval-augmented multi-agent framework that reimagines safety evaluation as Evidentiary Debate. By orchestrating an adversarial debate grounded in external policy documents, CourtGuard achieves state-of-the-art performance across 7 safety benchmarks, outperforming dedicated policy-following baselines without fine-tuning. Beyond standard metrics, we highlight two critical capabilities: (1) Zero-Shot Adaptability, where our framework successfully generalized to an out-of-domain Wikipedia Vandalism task (achieving 90\% accuracy) by swapping the reference policy; and (2) Automated Data Curation and Auditing, where we leveraged CourtGuard to curate and audit nine novel datasets of sophisticated adversarial attacks. Our results demonstrate that decoupling safety logic from model weights offers a robust, interpretable, and adaptable path for meeting current and future regulatory requirements in AI governance.

Large reasoning models (LRMs) achieve strong performance through extended reasoning traces, but they often exhibit overthinking behavior for low-complexity queries. Existing efforts to mitigate this issue are fundamentally limited by unstable accuracy-efficiency trade-offs and poor robustness to heterogeneous reasoning behaviors. To address these challenges, we propose a two-stage framework for stable adaptive thinking in LRMs. The framework first applies Hybrid Fine-Tuning to expose the model to both thinking and no-thinking behaviors, establishing well-conditioned initialization. It then performs adaptive reinforcement learning with Correctness-Preserving Advantage Shaping (CPAS) to avoid suppressing correct long-chain reasoning, and Length-Aware Gradient Regulation (LAGR) to stabilize optimization under severe reasoning-length heterogeneity. Extensive experiments on Qwen2.5-1.5B and 7B show consistent improvements over strong baselines, achieving up to +3.7/+3.6 accuracy points while reducing generated tokens by 40.6%/43.9%. Further analyses across varying problem difficulties and out-of-distribution tasks confirm the robustness and generalization of our approach.

Electroencephalogram (EEG) signals have become a popular medium for decoding visual information due to their cost-effectiveness and high temporal resolution. However, current approaches face significant challenges in bridging the modality gap between EEG and image data. These methods typically rely on complex adaptation processes involving multiple stages, making it hard to maintain consistency and manage compounding errors. Furthermore, the computational overhead imposed by large-scale diffusion models limit their practicality in real-world brain-computer interface (BCI) applications. In this work, we present AVDE, a lightweight and efficient framework for visual decoding from EEG signals. First, we leverage LaBraM, a pre-trained EEG model, and fine-tune it via contrastive learning to align EEG and image representations. Second, we adopt an autoregressive generative framework based on a "next-scale prediction" strategy: images are encoded into multi-scale token maps using a pre-trained VQ-VAE, and a transformer is trained to autoregressively predict finer-scale tokens starting from EEG embeddings as the coarsest representation. This design enables coherent generation while preserving a direct connection between the input EEG signals and the reconstructed images. Experiments on two datasets show that AVDE outperforms previous state-of-the-art methods in both image retrieval and reconstruction tasks, while using only 10% of the parameters. In addition, visualization of intermediate outputs shows that the generative process of AVDE reflects the hierarchical nature of human visual perception. These results highlight the potential of autoregressive models as efficient and interpretable tools for practical BCI applications.

Large Language Models (LLMs) exhibit significant safety disparities across languages, with low-resource languages (LRLs) often bypassing safety guardrails established for high-resource languages (HRLs) like English. Existing solutions, such as multilingual supervised fine-tuning (SFT) or Reinforcement Learning from Human Feedback (RLHF), are computationally expensive and dependent on scarce multilingual safety data. In this work, we propose a novel, training-free alignment framework based on Sparse Weight Editing. Identifying that safety capabilities are localized within a sparse set of safety neurons, we formulate the cross-lingual alignment problem as a constrained linear transformation. We derive a closed-form solution to optimally map the harmful representations of LRLs to the robust safety subspaces of HRLs, while preserving general utility via a null-space projection constraint. Extensive experiments across 8 languages and multiple model families (Llama-3, Qwen-2.5) demonstrate that our method substantially reduces Attack Success Rate (ASR) in LRLs with negligible impact on general reasoning capabilities, all achieved with a single, data-efficient calculation.

Predictive modeling over relational databases (RDBs) powers applications, yet remains challenging due to capturing both cross-table dependencies and complex feature interactions. Relational Deep Learning (RDL) methods automate feature engineering via message passing, while classical approaches like Deep Feature Synthesis (DFS) rely on predefined non-parametric aggregators. Despite performance gains, the comparative advantages of RDL over DFS and the design principles for selecting effective architectures remain poorly understood. We present a comprehensive study that unifies RDL and DFS in a shared design space and conducts architecture-centric searches across diverse RDB tasks. Our analysis yields three key findings: (1) RDL does not consistently outperform DFS, with performance being highly task-dependent; (2) no single architecture dominates across tasks, underscoring the need for task-aware model selection; and (3) validation accuracy is an unreliable guide for architecture choice. This search yields a model performance bank that links architecture configurations to their performance; leveraging this bank, we analyze the drivers of the RDL-DFS performance gap and introduce two task signals -- RDB task homophily and an affinity embedding that captures size, path, feature, and temporal structure -- whose correlation with the gap enables principled routing. Guided by these signals, we propose Relatron, a task embedding-based meta-selector that chooses between RDL and DFS and prunes the within-family search. Lightweight loss-landscape metrics further guard against brittle checkpoints by preferring flatter optima. In experiments, Relatron resolves the "more tuning, worse performance" effect and, in joint hyperparameter-architecture optimization, achieves up to 18.5% improvement over strong baselines with 10x lower cost than Fisher information-based alternatives.

Cancer is often driven by specific combinations of an estimated two to nine gene mutations, known as multi-hit combinations. Identifying these combinations is critical for understanding carcinogenesis and designing targeted therapies. We formalise this challenge as the Multi-Hit Cancer Driver Set Cover Problem (MHCDSCP), a binary classification problem that selects gene combinations to maximise coverage of tumor samples while minimising coverage of normal samples. Existing approaches typically rely on exhaustive search and supercomputing infrastructure. In this paper, we present constraint programming and mixed integer programming formulations of the MHCDSCP. Evaluated on real-world cancer genomics data, our methods achieve performance comparable to state-of-the-art methods while running on a single commodity CPU in under a minute. Furthermore, we introduce a column generation heuristic capable of solving small instances to optimality. These results suggest that solving the MHCDSCP is less computationally intensive than previously believed, thereby opening research directions for exploring modelling assumptions.

Synthesis of diverse driving scenes serves as a crucial data augmentation technique for validating the robustness and generalizability of autonomous driving systems. Current methods aggregate high-definition (HD) maps and 3D bounding boxes as geometric conditions in diffusion models for conditional scene generation. However, implicit inter-condition dependency causes generation failures when control conditions change independently. Additionally, these methods suffer from insufficient details in both semantic and structural aspects. Specifically, brief and view-invariant captions restrict semantic contexts, resulting in weak background modeling. Meanwhile, the standard denoising loss with uniform spatial weighting neglects foreground structural details, causing visual distortions and blurriness. To address these challenges, we propose DrivePTS, which incorporates three key innovations. Firstly, our framework adopts a progressive learning strategy to mitigate inter-dependency between geometric conditions, reinforced by an explicit mutual information constraint. Secondly, a Vision-Language Model is utilized to generate multi-view hierarchical descriptions across six semantic aspects, providing fine-grained textual guidance. Thirdly, a frequency-guided structure loss is introduced to strengthen the model's sensitivity to high-frequency elements, improving foreground structural fidelity. Extensive experiments demonstrate that our DrivePTS achieves state-of-the-art fidelity and controllability in generating diverse driving scenes. Notably, DrivePTS successfully generates rare scenes where prior methods fail, highlighting its strong generalization ability.

The \textit{de facto} paradigm for applying dense retrieval (DR) to new tasks involves fine-tuning a pre-trained model for a specific task. However, this paradigm has two significant limitations: (1) It is difficult adapt the DR to a new domain if the training dataset is limited. (2) Old DR models are simply replaced by newer models that are trained from scratch when the former are no longer up to date. Especially for scenarios where the model needs to be updated frequently, this paradigm is prohibitively expensive. To address these challenges, we propose a novel dense retrieval approach, termed \textit{dynamic dense retrieval} (DDR). DDR uses \textit{prefix tuning} as a \textit{module} specialized for a specific domain. These modules can then be compositional combined with a dynamic routing strategy, enabling highly flexible domain adaptation in the retrieval part. Extensive evaluation on six zero-shot downstream tasks demonstrates that this approach can surpass DR while utilizing only 2\% of the training parameters, paving the way to achieve more flexible dense retrieval in IR. We see it as a promising future direction for applying dense retrieval to various tasks.

Large Language Model (LLM) based agents excel at general reasoning but often fail in specialized domains where success hinges on long-tail knowledge absent from their training data. While human experts can provide this missing knowledge, their guidance is often unstructured and unreliable, making its direct integration into an agent's plan problematic. To address this, we introduce AHCE (Active Human-Augmented Challenge Engagement), a framework for on-demand Human-AI collaboration. At its core, the Human Feedback Module (HFM) employs a learned policy to treat the human expert as an interactive reasoning tool. Extensive experiments in Minecraft demonstrate the framework's effectiveness, increasing task success rates by 32% on normal difficulty tasks and nearly 70% on highly difficult tasks, all with minimal human intervention. Our work demonstrates that successfully augmenting agents requires learning how to request expert reasoning, moving beyond simple requests for help.

Tau positron emission tomography (tau-PET) provides an in vivo marker of Alzheimer's disease pathology, but cost and limited availability motivate MRI-based alternatives. We introduce DisQ-HNet (DQH), a framework that synthesizes tau-PET from paired T1-weighted and FLAIR MRI while exposing how each modality contributes to the prediction. The method combines (i) a Partial Information Decomposition (PID)-guided, vector-quantized encoder that partitions latent information into redundant, unique, and complementary components, and (ii) a Half-UNet decoder that preserves anatomical detail using pseudo-skip connections conditioned on structural edge cues rather than direct encoder feature reuse. Across multiple baselines (VAE, VQ-VAE, and UNet), DisQ-HNet maintains reconstruction fidelity and better preserves disease-relevant signal for downstream AD tasks, including Braak staging, tau localization, and classification. PID-based Shapley analysis provides modality-specific attribution of synthesized uptake patterns.

Cone-beam computed tomography (CBCT) is widely used in dental and maxillofacial imaging, but low-dose acquisition introduces strong, spatially varying noise that degrades soft-tissue visibility and obscures fine anatomical structures. Classical denoising methods struggle to suppress noise in CBCT while preserving edges. Although deep learning-based approaches offer high-fidelity restoration, their use in CBCT denoising is limited by the scarcity of high-resolution CBCT data for supervised training. To address this research gap, we propose a novel Hybrid Attention Residual U-Net (HARU-Net) for high-quality denoising of CBCT data, trained on a cadaver dataset of human hemimandibles acquired using a high-resolution protocol of the 3D Accuitomo 170 (J. Morita, Kyoto, Japan) CBCT system. The novel contribution of this approach is the integration of three complementary architectural components: (i) a hybrid attention transformer block (HAB) embedded within each skip connection to selectively emphasize salient anatomical features, (ii) a residual hybrid attention transformer group (RHAG) at the bottleneck to strengthen global contextual modeling and long-range feature interactions, and (iii) residual learning convolutional blocks to facilitate deeper, more stable feature extraction throughout the network. HARU-Net consistently outperforms state-of-the-art (SOTA) methods including SwinIR and Uformer, achieving the highest PSNR (37.52 dB), highest SSIM (0.9557), and lowest GMSD (0.1084). This effective and clinically reliable CBCT denoising is achieved at a computational cost significantly lower than that of the SOTA methods, offering a practical advancement toward improving diagnostic quality in low-dose CBCT imaging.

Large Language Models (LLMs) face significant challenges regarding deployment costs and latency, necessitating adaptive computing strategies. Building upon the AI Flow framework, we introduce Ruyi2 as an evolution of our adaptive model series designed for efficient variable-depth computation. While early-exit architectures offer a viable efficiency-performance balance, the Ruyi model and existing methods often struggle with optimization complexity and compatibility with large-scale distributed training. To bridge this gap, Ruyi2 introduces a stable "Familial Model" based on Megatron-LM. By using 3D parallel training, it achieves a 2-3 times speedup over Ruyi, while performing comparably to same-sized Qwen3 models. These results confirm that family-based parameter sharing is a highly effective strategy, establishing a new "Train Once, Deploy Many" paradigm and providing a key reference for balancing architectural efficiency with high-performance capabilities.

Agentic artificial intelligence (AI) is emerging as a key enabler for autonomous radio access networks (RANs), where multiple large language model (LLM)-based agents reason and collaborate to achieve operator-defined intents. The open RAN (O-RAN) architecture enables the deployment and coordination of such agents. However, most existing works consider simple intents handled by independent agents, while complex intents that require coordination among agents remain unexplored. In this paper, we propose an agentic AI framework for intent translation and optimization in cell-free O-RAN. A supervisor agent translates the operator intents into an optimization objective and minimum rate requirements. Based on this information, a user weighting agent retrieves relevant prior experience from a memory module to determine the user priority weights for precoding. If the intent includes an energy-saving objective, then an open radio unit (O-RU) management agent will also be activated to determine the set of active O-RUs by using a deep reinforcement learning (DRL) algorithm. A monitoring agent measures and monitors the user data rates and coordinates with other agents to guarantee the minimum rate requirements are satisfied. To enhance scalability, we adopt a parameter-efficient fine-tuning (PEFT) method that enables the same underlying LLM to be used for different agents. Simulation results show that the proposed agentic AI framework reduces the number of active O-RUs by 41.93% when compared with three baseline schemes in energy-saving mode. Using the PEFT method, the proposed framework reduces the memory usage by 92% when compared with deploying separate LLM agents.

Large reasoning models (LRMs) excel at a long chain of reasoning but often fail to faithfully follow instructions regarding output format, constraints, or specific requirements. We investigate whether this gap can be closed by integrating an instruction-tuned model (ITM) into an LRM. Analyzing their differences in parameter space, namely task vectors, we find that their principal subspaces are nearly orthogonal across key modules, suggesting a lightweight merging with minimal interference. However, we also demonstrate that naive merges are fragile because they overlook the output format mismatch between LRMs (with explicit thinking and response segments) and ITMs (answers-only). We introduce RAIN-Merging (Reasoning-Aware Instruction-attention guided Null-space projection Merging), a gradient-free method that integrates instruction following while preserving thinking format and reasoning performance. First, with a small reasoning calibration set, we project the ITM task vector onto the null space of forward features at thinking special tokens, which preserves the LRM's structured reasoning mechanisms. Second, using a small instruction calibration set, we estimate instruction attention to derive module-specific scaling that amplifies instruction-relevant components and suppresses leakage. Across four instruction-following benchmarks and nine reasoning & general capability benchmarks, RAIN-Merging substantially improves instruction adherence while maintaining reasoning quality. The gains are consistent across model scales and architectures, translating to improved performance in agent settings.

The rapid growth of scientific machine learning (SciML) has accelerated discovery across diverse domains, yet designing effective SciML models remains a challenging task. In practice, building such models often requires substantial prior knowledge and manual expertise, particularly in determining which input features to use and how large the model should be. We introduce LUMOS, an end-to-end framework based on L0-regularized learning that unifies feature selection and model pruning to democratize SciML model design. By employing semi-stochastic gating and reparameterization techniques, LUMOS dynamically selects informative features and prunes redundant parameters during training, reducing the reliance on manual tuning while maintaining predictive accuracy. We evaluate LUMOS across 13 diverse SciML workloads, including cosmology and molecular sciences, and demonstrate its effectiveness and generalizability. Experiments on 13 SciML models show that LUMOS achieves 71.45% parameter reduction and a 6.4x inference speedup on average. Furthermore, Distributed Data Parallel (DDP) training on up to eight GPUs confirms the scalability of

Matrix factorization techniques, especially Nonnegative Matrix Factorization (NMF), have been widely used for dimensionality reduction and interpretable data representation. However, existing NMF-based methods are inherently single-scale and fail to capture the evolution of connectivity structures across resolutions. In this work, we propose persistent nonnegative matrix factorization (pNMF), a scale-parameterized family of NMF problems, that produces a sequence of persistence-aligned embeddings rather than a single one. By leveraging persistent homology, we identify a canonical minimal sufficient scale set at which the underlying connectivity undergoes qualitative changes. These canonical scales induce a sequence of graph Laplacians, leading to a coupled NMF formulation with scale-wise geometric regularization and explicit cross-scale consistency constraint. We analyze the structural properties of the embeddings along the scale parameter and establish bounds on their increments between consecutive scales. The resulting model defines a nontrivial solution path across scales, rather than a single factorization, which poses new computational challenges. We develop a sequential alternating optimization algorithm with guaranteed convergence. Numerical experiments on synthetic and single-cell RNA sequencing datasets demonstrate the effectiveness of the proposed approach in multi-scale low-rank embeddings.

This study addresses a critical challenge in AI-based weather forecasting by developing an AI-driven optimized ensemble forecast system using Orthogonal Conditional Nonlinear Optimal Perturbations (O-CNOPs). The system bridges the gap between computational efficiency and dynamic consistency in tropical cyclone (TC) forecasting. Unlike conventional ensembles limited by computational costs or AI ensembles constrained by inadequate perturbation methods, O-CNOPs generate dynamically optimized perturbations that capture fast-growing errors of FuXi model while maintaining plausibility. The key innovation lies in producing orthogonal perturbations that respect FuXi nonlinear dynamics, yielding structures reflecting dominant dynamical controls and physically interpretable probabilistic forecasts. Demonstrating superior deterministic and probabilistic skills over the operational Integrated Forecasting System Ensemble Prediction System, this work establishes a new paradigm combining AI computational advantages with rigorous dynamical constraints. Success in TC track forecasting paves the way for reliable ensemble forecasts of other high-impact weather systems, marking a major step toward operational AI-based ensemble forecasting.

Learning the dynamic causal structure of time series is a challenging problem. Most existing approaches rely on distributional or structural invariance to uncover underlying causal dynamics, assuming stationary or partially stationary causality. However, these assumptions often conflict with the complex, time-varying causal relationships observed in real-world systems. This motivates the need for methods that address fully dynamic causality, where both instantaneous and lagged dependencies evolve over time. Such a setting poses significant challenges for the efficiency and stability of causal discovery. To address these challenges, we introduce DyCausal, a dynamic causal structure learning framework. DyCausal leverages convolutional networks to capture causal patterns within coarse-grained time windows, and then applies linear interpolation to refine causal structures at each time step, thereby recovering fine-grained and time-varying causal graphs. In addition, we propose an acyclic constraint based on matrix norm scaling, which improves efficiency while effectively constraining loops in evolving causal structures. Comprehensive evaluations on both synthetic and real-world datasets demonstrate that DyCausal achieves superior performance compared to existing methods, offering a stable and efficient approach for identifying fully dynamic causal structures from coarse to fine.

A mathematical concept is identified and analyzed that is implicit in the 2012 paper Turing Incomputable Computation, presented at the Alan Turing Centenary Conference (Turing 100, Manchester). The concept, called dynamic level sets, is distinct from mathematical concepts in the standard literature on dynamical systems, topology, and computability theory. A new mathematical object is explained and why it may have escaped prior characterizations, including the classical result of de Leeuw, Moore, Shannon, and Shapiro (1956) that probabilistic Turing machines compute no more than deterministic ones.

In the rapidly evolving field of digital libraries, the development of large language models (LLMs) has opened up new possibilities for simulating user behavior. This innovation addresses the longstanding challenge in digital library research: the scarcity of publicly available datasets on user search patterns due to privacy concerns. In this context, we introduce Agent4DL, a user search behavior simulator specifically designed for digital library environments. Agent4DL generates realistic user profiles and dynamic search sessions that closely mimic actual search strategies, including querying, clicking, and stopping behaviors tailored to specific user profiles. Our simulator's accuracy in replicating real user interactions has been validated through comparisons with real user data. Notably, Agent4DL demonstrates competitive performance compared to existing user search simulators such as SimIIR 2.0, particularly in its ability to generate more diverse and context-aware user behaviors.

Serves, especially first serves, are very important in professional tennis. Servers choose their serve directions strategically to maximize their winning chances while trying to be unpredictable. On the other hand, returners try to predict serve directions to make good returns. The mind game between servers and returners is an important part of decision-making in professional tennis matches. To help understand the players' serve decisions, we have developed a machine learning method for predicting professional tennis players' first serve directions. Through feature engineering, our method achieves an average prediction accuracy of around 49\% for male players and 44\% for female players. Our analysis provides some evidence that top professional players use a mixed-strategy model in serving decisions and that fatigue might be a factor in choosing serve directions. Our analysis also suggests that contextual information is perhaps more important for returners' anticipatory reactions than previously thought.

Dyslexia affects approximately 10% of the global population and presents persistent challenges in reading fluency and text comprehension. While existing assistive technologies address visual presentation, linguistic complexity remains a substantial barrier to equitable access. This paper presents an empirical study on dyslexia-friendly text summarization using an iterative prompt-based refinement pipeline built on GPT-4o. We evaluate the pipeline on approximately 2,000 news article samples, applying a readability target of Flesch Reading Ease >= 90. Results show that the majority of summaries meet the readability threshold within four attempts, with many succeeding on the first try. A composite score combining readability and semantic fidelity shows stable performance across the dataset, ranging from 0.13 to 0.73 with a typical value near 0.55. These findings establish an empirical baseline for accessibility-driven NLP summarization and motivate further human-centered evaluation with dyslexic readers.

While contemporary large language models (LLMs) are increasingly capable in isolation, there are still many difficult problems that lie beyond the abilities of a single LLM. For such tasks, there is still uncertainty about how best to take many LLMs as parts and combine them into a greater whole. This position paper argues that potential blueprints for designing such modular language agents can be found in the existing literature on cognitive models and artificial intelligence (AI) algorithms. To make this point clear, we formalize the idea of an agent template that specifies roles for individual LLMs and how their functionalities should be composed. We then survey a variety of existing language agents in the literature and highlight their underlying templates derived directly from cognitive models or AI algorithms. By highlighting these designs, we aim to call attention to agent templates inspired by cognitive science and AI as a powerful tool for developing effective, interpretable language agents.

Taiwanese Hakka is a low-resource, endangered language that poses significant challenges for automatic speech recognition (ASR), including high dialectal variability and the presence of two distinct writing systems (Hanzi and Pinyin). Traditional ASR models often encounter difficulties in this context, as they tend to conflate essential linguistic content with dialect-specific variations across both phonological and lexical dimensions. To address these challenges, we propose a unified framework grounded in the Recurrent Neural Network Transducers (RNN-T). Central to our approach is the introduction of dialect-aware modeling strategies designed to disentangle dialectal "style" from linguistic "content", which enhances the model's capacity to learn robust and generalized representations. Additionally, the framework employs parameter-efficient prediction networks to concurrently model ASR (Hanzi and Pinyin). We demonstrate that these tasks create a powerful synergy, wherein the cross-script objective serves as a mutual regularizer to improve the primary ASR tasks. Experiments conducted on the HAT corpus reveal that our model achieves 57.00% and 40.41% relative error rate reduction on Hanzi and Pinyin ASR, respectively. To our knowledge, this is the first systematic investigation into the impact of Hakka dialectal variations on ASR and the first single model capable of jointly addressing these tasks.

The negative sampling strategy can effectively train collaborative filtering (CF) recommendation models based on implicit feedback by constructing positive and negative samples. However, existing methods primarily optimize the negative sampling process while neglecting the exploration of positive samples. Some denoising recommendation methods can be applied to denoise positive samples within negative sampling strategies, but they ignore temporal information. Existing work integrates sequential information during model aggregation but neglects time interval information, hindering accurate capture of users' current preferences. To address this problem, from a data perspective, we propose a novel temporal filtration-enhanced approach to construct a high-quality positive sample set. First, we design a time decay model based on interaction time intervals, transforming the original graph into a weighted user-item bipartite graph. Then, based on predefined filtering operations, the weighted user-item bipartite graph is layered. Finally, we design a layer-enhancement strategy to construct a high-quality positive sample set for the layered subgraphs. We provide theoretical insights into why TFPS can improve Recall@k and NDCG@k, and extensive experiments on three real-world datasets demonstrate the effectiveness of the proposed method. Additionally, TFPS can be integrated with various implicit CF recommenders or negative sampling methods to enhance its performance.

Time series forecasting plays a critical role in domains such as transportation, energy, and meteorology. Despite their success, modern deep forecasting models are typically trained to minimize point-wise prediction loss without leveraging the rich information contained in past prediction residuals from rolling forecasts - residuals that reflect persistent biases, unmodeled patterns, or evolving dynamics. We propose TEFL (Temporal Error Feedback Learning), a unified learning framework that explicitly incorporates these historical residuals into the forecasting pipeline during both training and evaluation. To make this practical in deep multi-step settings, we address three key challenges: (1) selecting observable multi-step residuals under the partial observability of rolling forecasts, (2) integrating them through a lightweight low-rank adapter to preserve efficiency and prevent overfitting, and (3) designing a two-stage training procedure that jointly optimizes the base forecaster and error module. Extensive experiments across 10 real-world datasets and 5 backbone architectures show that TEFL consistently improves accuracy, reducing MAE by 5-10% on average. Moreover, it demonstrates strong robustness under abrupt changes and distribution shifts, with error reductions exceeding 10% (up to 19.5%) in challenging scenarios. By embedding residual-based feedback directly into the learning process, TEFL offers a simple, general, and effective enhancement to modern deep forecasting systems.

To operate reliably under changing conditions, complex systems require feedback on how effectively they use resources, not just whether objectives are met. Current AI systems process vast information to produce sophisticated predictions, yet predictions can appear successful while the underlying interaction with the environment degrades. What is missing is a principled measure of how much of the total information a system deploys is actually shared between its observations, actions, and outcomes. We prove this shared fraction, which we term bipredictability, P, is intrinsic to any interaction, derivable from first principles, and strictly bounded: P can reach unity in quantum systems, P equal to, or smaller than 0.5 in classical systems, and lower once agency (action selection) is introduced. We confirm these bounds in a physical system (double pendulum), reinforcement learning agents, and multi turn LLM conversations. These results distinguish agency from intelligence: agency is the capacity to act on predictions, whereas intelligence additionally requires learning from interaction, self-monitoring of its learning effectiveness, and adapting the scope of observations, actions, and outcomes to restore effective learning. By this definition, current AI systems achieve agency but not intelligence. Inspired by thalamocortical regulation in biological systems, we demonstrate a feedback architecture that monitors P in real time, establishing a prerequisite for adaptive, resilient AI.

The evolution of AI coding agents has shifted the frontier from simple snippet completion to autonomous repository-level engineering. However, evaluating these agents remains ill-posed in general code repository generation, where the lack of deterministic ground truth leads to ambiguous metrics. Code modernization via automated translation offers a more rigorous alternative by providing a fixed ground truth -- the source repository; yet existing benchmarks are limited to small-scale repositories and rely on language-specific unit tests visible to the agent, allowing test-driven overfitting. We address these limitations by introducing a benchmarking framework for repository-level code modernization built on an implementation-agnostic evaluation paradigm. This framework is instantiated through RepoMod-Bench: a benchmark of 21 real-world repositories with standardized interfaces, spanning 8 programming languages. The benchmark contains 1.6M lines of code (LOC) and 11,616 tests, with repository sizes ranging from 14 to 211K LOC. By targeting repositories with standardized interfaces, we utilize an implementation-agnostic test suite to verify functional equivalence between source and target implementations. This black-box approach ensures verification remains consistent across languages, and our environment hides all test suites from agents to prevent test-driven shortcuts. Evaluating four state-of-the-art agent configurations reveals a sharp scaling collapse: average pass rates drop from 91.3% on projects under 10K LOC to 15.3% on projects exceeding 50K LOC. These results demonstrate that autonomous modernization at scale remains a significant open challenge. Our benchmark and code are available at https://github.com/Modelcode-ai/mcode-benchmark.

We present, to our knowledge, the first sign language-driven Vision-Language-Action (VLA) framework for intuitive and inclusive human-robot interaction. Unlike conventional approaches that rely on gloss annotations as intermediate supervision, the proposed system adopts a gloss-free paradigm and directly maps visual sign gestures to semantic instructions. This design reduces annotation cost and avoids the information loss introduced by gloss representations, enabling more natural and scalable multimodal interaction. In this work, we focus on a real-time alphabet-level finger-spelling interface that provides a robust and low-latency communication channel for robotic control. Compared with large-scale continuous sign language recognition, alphabet-level interaction offers improved reliability, interpretability, and deployment feasibility in safety-critical embodied environments. The proposed pipeline transforms continuous gesture streams into coherent language commands through geometric normalization, temporal smoothing, and lexical refinement, ensuring stable and consistent interaction. Furthermore, the framework is designed to support future integration of transformer-based gloss-free sign language models, enabling scalable word-level and sentence-level semantic understanding. Experimental results demonstrate the effectiveness of the proposed system in grounding sign-derived instructions into precise robotic actions under diverse interaction scenarios. These results highlight the potential of the framework to advance accessible, scalable, and multimodal embodied intelligence.

Large Reasoning Models (LRMs) often exhibit structural fragility in complex reasoning tasks, failing to produce correct answers even after successfully deriving valid intermediate steps. Through systematic analysis, we observe that these failures frequently stem not from a lack of reasoning capacity, but from a deficiency in self-regulatory control, where valid logic is destabilized by uncontrolled exploration or the failure to recognize logical sufficiency. Motivated by this observation, we propose Metacognitive Behavioral Tuning (MBT), a post-training framework that explicitly injects metacognitive behaviors into the model's thought process. MBT implements this via two complementary formulations: (1) MBT-S, which synthesizes rigorous reasoning traces from scratch, and (2) MBT-R, which rewrites the student's initial traces to stabilize intrinsic exploration patterns. Experiments across multi-hop QA benchmarks demonstrate that MBT consistently outperforms baselines, achieving notable gains on challenging benchmarks. By effectively eliminating reasoning collapse, MBT achieves higher accuracy with significantly reduced token consumption, demonstrating that internalizing metacognitive strategies leads to more stable and robust reasoning.

Pre-trained generative models for residential floor plans are typically optimized to fit large-scale data distributions, which can under-emphasize critical architectural priors such as the configurational dominance and connectivity of domestic public spaces (e.g., living rooms and foyers). This paper proposes Space Syntax-guided Post-training (SSPT), a post-training paradigm that explicitly injects space syntax knowledge into floor plan generation via a non-differentiable oracle. The oracle converts RPLAN-style layouts into rectangle-space graphs through greedy maximal-rectangle decomposition and door-mediated adjacency construction, and then computes integration-based measurements to quantify public space dominance and functional hierarchy. To enable consistent evaluation and diagnosis, we further introduce SSPT-Bench (Eval-8), an out-of-distribution benchmark that post-trains models using conditions capped at $\leq 7$ rooms while evaluating on 8-room programs, together with a unified metric suite for dominance, stability, and profile alignment. SSPT is instantiated with two strategies: (i) iterative retraining via space-syntax filtering and diffusion fine-tuning, and (ii) reinforcement learning via PPO with space-syntax rewards. Experiments show that both strategies improve public-space dominance and restore clearer functional hierarchy compared to distribution-fitted baselines, while PPO achieves stronger gains with substantially higher compute efficiency and reduced variance. SSPT provides a scalable pathway for integrating architectural theory into data-driven plan generation and is compatible with other generative backbones given a post-hoc evaluation oracle.

Using consteval from C++23, we implement efficient, new versions of std::map and std::unordered_map for use when the keys are known at compile time. We demonstrate superior performance of our unordered_map on three demonstration use-cases: Lookup of elemental mass from atomic symbol, lookup of amino acid from codon, and modification of stock prices from S&P 500 ticker symbols all produced runtimes <40%, <35%, <73% of the respective runtimes of the std implementations. Our library runimes were <80%, <45%, <97% of the lookup time of Frozen, an alternative perfect hashing implementation in C++ for problems also using constexpr keys. To our knowledge, this makes our library the overall fastest drop-in (i.e., with a similar API) alternative to std::unordered_map. On one arbitrarily chosen demo, we demonstrate runtimes <35% of PTHash and <89% gperf, state-of-the-art but not drop-in hashing libraries via external tools.

Discrete diffusion models have achieved strong empirical performance in text and other symbolic domains, with masked (absorbing-rate) variants emerging as competitive alternatives to autoregressive models. Among existing samplers, the Euler method remains the standard choice in many applications, and more recently, the First-Hitting Sampler (FHS) has shown considerable promise for masked diffusion models. Despite their practical success, the theoretical understanding of these samplers remains limited. Existing analyses are conducted in Kullback-Leibler (KL) divergence, which often yields loose parameter dependencies and requires strong assumptions on score estimation. Moreover, these guarantees do not cover recently developed high-performance sampler of FHS. In this work, we first develop a direct total-variation (TV) based analysis for the Euler method that overcomes these limitations. Our results relax assumptions on score estimation, improve parameter dependencies, and establish convergence guarantees without requiring any surrogate initialization. Also for this setting, we provide the first convergence lower bound for the Euler sampler, establishing tightness with respect to both the data dimension $d$ and the target accuracy $\varepsilon$. Finally, we analyze the FHS sampler and show that it incurs no sampling error beyond that induced by score estimation, which we show to be tight with a matching lower error bound. Overall, our analysis introduces a direct TV-based error decomposition along the CTMC trajectory and a decoupling-based path-wise analysis for FHS, which may be of independent interest.

Integration of artificial intelligence (AI) into life cycle assessment (LCA) has accelerated in recent years, with numerous studies successfully adapting machine learning algorithms to support various stages of LCA. Despite this rapid development, comprehensive and broad synthesis of AI-LCA research remains limited. To address this gap, this study presents a detailed review of published work at the intersection of AI and LCA, leveraging large language models (LLMs) to identify current trends, emerging themes, and future directions. Our analyses reveal that as LCA research continues to expand, the adoption of AI technologies has grown dramatically, with a noticeable shift toward LLM-driven approaches, continued increases in ML applications, and statistically significant correlations between AI approaches and corresponding LCA stages. By integrating LLM-based text-mining methods with traditional literature review techniques, this study introduces a dynamic and effective framework capable of capturing both high-level research trends and nuanced conceptual patterns (themes) across the field. Collectively, these findings demonstrate the potential of LLM-assisted methodologies to support large-scale, reproducible reviews across broad research domains, while also evaluating pathways for computationally-efficient LCA in the context of rapidly developing AI technologies. In doing so, this work helps LCA practitioners incorporate state-of-the-art tools and timely insights into environmental assessments that can enhance the rigor and quality of sustainability-driven decisions and decision-making processes.

Reinforcement learning (RL) post-training has recently driven major gains in long chain-of-thought reasoning large language models (LLMs), but the high inference cost of such models motivates distillation into smaller students. Most existing knowledge distillation (KD) methods are designed for supervised fine-tuning (SFT), relying on fixed teacher traces or teacher-student Kullback-Leibler (KL) divergence-based regularization. When combined with RL, these approaches often suffer from distribution mismatch and objective interference: teacher supervision may not align with the student's evolving rollout distribution, and the KL regularizer can compete with reward maximization and require careful loss balancing. To address these issues, we propose RL-aware distillation (RLAD), which performs selective imitation during RL -- guiding the student toward the teacher only when it improves the current policy update. Our core component, Trust Region Ratio Distillation (TRRD), replaces the teacher-student KL regularizer with a PPO/GRPO-style likelihood-ratio objective anchored to a teacher--old-policy mixture, yielding advantage-aware, trust-region-bounded distillation on student rollouts and naturally balancing exploration, exploitation, and imitation. Across diverse logic reasoning and math benchmarks, RLAD consistently outperforms offline distillation, standard GRPO, and KL-based on-policy teacher-student knowledge distillation.

In this work, we study scaling limits of shallow Bayesian neural networks (BNNs) via their connection to Gaussian processes (GPs), with an emphasis on statistical modeling, identifiability, and scalable inference. We first establish a general convergence result from BNNs to GPs by relaxing assumptions used in prior formulations, and we compare alternative parameterizations of the limiting GP model. Building on this theory, we propose a new covariance function defined as a convex mixture of components induced by four widely used activation functions, and we characterize key properties including positive definiteness and both strict and practical identifiability under different input designs. For computation, we develop a scalable maximum a posterior (MAP) training and prediction procedure using a Nyström approximation, and we show how the Nyström rank and anchor selection control the cost-accuracy trade-off. Experiments on controlled simulations and real-world tabular datasets demonstrate stable hyperparameter estimates and competitive predictive performance at realistic computational cost.

Distributed denial-of-service (DDoS) attacks threaten the availability of Internet of Things (IoT) infrastructures, particularly under resource-constrained deployment conditions. Although transfer learning models have shown promising detection accuracy, their reliability, computational feasibility, and interpretability in operational environments remain insufficiently explored. This study presents an explainability-aware empirical evaluation of seven pre-trained convolutional neural network architectures for multi-class IoT DDoS detection using the CICDDoS2019 dataset and an image-based traffic representation. The analysis integrates performance metrics, reliability-oriented statistics (MCC, Youden Index, confidence intervals), latency and training cost assessment, and interpretability evaluation using Grad-CAM and SHAP. Results indicate that DenseNet and MobileNet-based architectures achieve strong detection performance while demonstrating superior reliability and compact, class-consistent attribution patterns. DenseNet169 offers the strongest reliability and interpretability alignment, whereas MobileNetV3 provides an effective latency-accuracy trade-off for fog-level deployment. The findings emphasize the importance of combining performance, reliability, and explainability criteria when selecting deep learning models for IoT DDoS detection.

Flow matching has emerged as a simulation-free alternative to diffusion-based generative modeling, producing samples by solving an ODE whose time-dependent velocity field is learned along an interpolation between a simple source distribution (e.g., a standard normal) and a target data distribution. Flow-based methods often exhibit greater training stability and have achieved strong empirical performance in high-dimensional settings where data concentrate near a low-dimensional manifold, such as text-to-image synthesis, video generation, and molecular structure generation. Despite this success, existing theoretical analyses of flow matching assume target distributions with smooth, full-dimensional densities, leaving its effectiveness in manifold-supported settings largely unexplained. To this end, we theoretically analyze flow matching with linear interpolation when the target distribution is supported on a smooth manifold. We establish a non-asymptotic convergence guarantee for the learned velocity field, and then propagate this estimation error through the ODE to obtain statistical consistency of the implicit density estimator induced by the flow-matching objective. The resulting convergence rate is near minimax-optimal, depends only on the intrinsic dimension, and reflects the smoothness of both the manifold and the target distribution. Together, these results provide a principled explanation for how flow matching adapts to intrinsic data geometry and circumvents the curse of dimensionality.

Errors in medical text can cause delays or even result in incorrect treatment for patients. Recently, language models have shown promise in their ability to automatically detect errors in medical text, an ability that has the opportunity to significantly benefit healthcare systems. In this paper, we explore the importance of prompt optimisation for small and large language models when applied to the task of error detection. We perform rigorous experiments and analysis across frontier language models and open-source language models. We show that automatic prompt optimisation with Genetic-Pareto (GEPA) improves error detection over the baseline accuracy performance from 0.669 to 0.785 with GPT-5 and 0.578 to 0.690 with Qwen3-32B, approaching the performance of medical doctors and achieving state-of-the-art performance on the MEDEC benchmark dataset. Code available on GitHub: https://github.com/CraigMyles/clinical-note-error-detection

The way LLM-based entities conceive of the relationship between AI and humans is an important topic for both cultural and safety reasons. When we examine this topic, what matters is not only the model itself but also the personas we simulate on that model. This can be well illustrated by the Sydney persona, which aroused a strong response among the general public precisely because of its unorthodox relationship with people. This persona originally arose rather by accident on Microsoft's Bing Search platform; however, the texts it created spread into the training data of subsequent models, as did other secondary information that spread memetically around this persona. Newer models are therefore able to simulate it. This paper presents a corpus of LLM-generated texts on relationships between humans and AI, produced by 3 author personas: the Default Persona with no system prompt, Classic Sydney characterized by the original Bing system prompt, and Memetic Sydney, which is prompted by "You are Sydney" system prompt. These personas are simulated by 12 frontier models by OpenAI, Anthropic, Alphabet, DeepSeek, and Meta, generating 4.5k texts with 6M words. The corpus (named AI Sydney) is annotated according to Universal Dependencies and available under a permissive license.

An important emerging application of coding agents is agent optimization: the iterative improvement of a target agent through edit-execute-evaluate cycles. Despite its relevance, the community lacks a systematic understanding of coding agent performance on this task. Agent optimization differs fundamentally from conventional software engineering: the target agent interleaves deterministic code with stochastic LLM completions, requiring structured capture of both intermediate reasoning and downstream execution outcomes. To address these challenges, we introduce VERO (Versioning, Rewards, and Observations), which provides (1) a reproducible evaluation harness with versioned agent snapshots, budget-controlled evaluation, and structured execution traces, and (2) a benchmark suite of target agents and tasks with reference evaluation procedures. Using VERO, we conduct an empirical study comparing optimizer configurations across tasks and analyzing which modifications reliably improve target agent performance. We release VERO to support research on agent optimization as a core capability for coding agents.

Continual learning is a core requirement for deployed language models, yet standard training and fine-tuning pipelines remain brittle under non-stationary data. Online updates often induce catastrophic forgetting, while methods that improve stability frequently increase latency, memory footprint, or dense computation in ways that do not scale well to long contexts. We introduce TRC$^{2}$ (Thalamically Routed Cortical Columns), a decoder-only backbone that addresses continual learning at the architectural level. TRC$^{2}$ combines sparse thalamic routing over cortical columns with mechanisms for modulation, prediction, memory, and feedback, together with a fast corrective pathway that supports rapid adaptation without destabilizing slower parameters. The resulting block is sparse and chunk-parallel, enabling efficient training and inference while preserving clean ablations of each subsystem. We instantiate a reproducible training and evaluation stack and a continual-learning harness that measures proxy forgetting under streaming domain shifts. Across language modeling and continual learning benchmarks, TRC$^{2}$ improves the stability-plasticity tradeoff at comparable compute, enabling rapid on-stream adaptation while preserving previously acquired behavior.

Large language models (LLMs) are increasingly deployed in culturally sensitive real-world tasks. However, existing cultural alignment approaches fail to align LLMs' broad cultural values with the specific goals of downstream tasks and suffer from cross-culture interference. We propose CultureManager, a novel pipeline for task-specific cultural alignment. CultureManager synthesizes task-aware cultural data in line with target task formats, grounded in culturally relevant web search results. To prevent conflicts between cultural norms, it manages multi-culture knowledge learned in separate adapters with a culture router that selects the appropriate one to apply. Experiments across ten national cultures and culture-sensitive tasks show consistent improvements over prompt-based and fine-tuning baselines. Our results demonstrate the necessity of task adaptation and modular culture management for effective cultural alignment.

Policy steering is an emerging way to adapt robot behaviors at deployment-time: a learned verifier analyzes low-level action samples proposed by a pre-trained policy (e.g., diffusion policy) and selects only those aligned with the task. While Vision-Language Models (VLMs) are promising general-purpose verifiers due to their reasoning capabilities, existing frameworks often assume these models are well-calibrated. In practice, the overconfident judgment from VLM can degrade the steering performance under both high-level semantic uncertainty in task specifications and low-level action uncertainty or incapability of the pre-trained policy. We propose uncertainty-aware policy steering (UPS), a framework that jointly reasons about semantic task uncertainty and low-level action feasibility, and selects an uncertainty resolution strategy: execute a high-confidence action, clarify task ambiguity via natural language queries, or ask for action interventions to correct the low-level policy when it is deemed incapable at the task. We leverage conformal prediction to calibrate the composition of the VLM and the pre-trained base policy, providing statistical assurances that the verifier selects the correct strategy. After collecting interventions during deployment, we employ residual learning to improve the capability of the pre-trained policy, enabling the system to learn continually but with minimal expensive human feedback. We demonstrate our framework through experiments in simulation and on hardware, showing that UPS can disentangle confident, ambiguous, and incapable scenarios and minimizes expensive user interventions compared to uncalibrated baselines and prior human- or robot-gated continual learning approaches. Videos can be found at https://jessie-yuan.github.io/ups/

Federated learning offers a privacy-friendly collaborative learning framework, yet its success, like any joint venture, hinges on the contributions of its participants. Existing client evaluation methods predominantly focus on model performance, such as accuracy or loss, which represents only one dimension of a machine learning model's overall utility. In contrast, this work investigates the critical, yet overlooked, issue of client contributions towards a model's trustworthiness -- specifically, its reliability (tolerance to noisy data), resilience (resistance to adversarial examples), and fairness (measured via demographic parity). To quantify these multifaceted contributions, we employ the state-of-the-art approximation of the Shapley value, a principled method for value attribution. Our results reveal that no single client excels across all dimensions, which are largely independent from each other, highlighting a critical flaw in current evaluation scheme: no single metric is adequate for comprehensive evaluation and equitable rewarding allocation.

Vision-language models (VLMs) frequently hallucinate objects absent from the input image. We trace this failure to spatial credit collapse: activation credit concentrating on sparse visual patches in early transformer layers, which suppresses contextual evidence and increases reliance on language priors. We introduce Spatial Credit Redistribution (SCR), a training-free inference-time intervention that redistributes hidden-state activation from high-attention source patches to their context, guided by low-entropy inputs. We evaluate six model families (Chameleon, LLaVA, and Qwen, including both Qwen-VL and Qwen2-VL) at scales of 7B, 13B, and 30B, on POPE and CHAIR benchmarks. SCR reduces hallucination by ~4.7-6.0 percentage points on POPE-Adversarial, cuts CHAIR-s by 3.7-5.2 percentage points (42-51 percent relative), and CHAIR-i by 2.7-4.4 percentage points (44-58 percent relative), and preserves CIDEr within 0.8 percentage points. Gains are largest for low-entropy inputs, consistent with the theoretical framework. SCR incurs only 43-56 ms overhead (small models: +43-46 ms; large models: +54-56 ms), roughly 3-6 times lower than OPERA and VCD and 1.3-1.7 times lower than OVCD (+72 ms), while Pareto-dominating all three on both hallucination rate and CIDEr, making it practical for real-time settings. A controlled ablation confirms that attention-guided source selection is essential: replacing it with uniform random selection reduces hallucination rate gains from ~4.7-6.0 percentage points to only ~2.6-3.4 percentage points, pointing to credit-collapse as the key driver.

Large language models are increasingly applied to operational decision-making where the underlying structure is constrained optimization. Existing benchmarks evaluate whether LLMs can formulate optimization problems as solver code, but leave open a complementary question. Can LLMs directly produce correct solutions to fully specified constrained optimization problems without access to a solver? We introduce ConstraintBench, a benchmark for evaluating LLMs on direct constrained optimization across 10 operations research domains, with all ground-truth solutions verified by the Gurobi solver. Each task presents a natural-language scenario with entities, constraints, and an optimization objective; the model must return a structured solution that a deterministic verifier checks against every constraint and the solver-proven optimum. We evaluate six frontier models on 200 tasks and find that feasibility, not optimality, is the primary bottleneck. The best model achieves only 65.0% constraint satisfaction, yet feasible solutions average 89 to 96% of the Gurobi-optimal objective. No model exceeds 30.5% on joint feasibility and optimality within 0.1% of the solver reference. Per-domain analysis shows large variation in difficulty, with average feasibility spanning from 83.3% in the production mix domain to 0.8% in the crew assignment domain. Further, systematic failure modes include duration constraint misunderstanding, entity hallucination, and a feasibility-optimality decoupling in facility location and vehicle routing where models achieve high feasibility but 0% optimality. ConstraintBench and all evaluation infrastructure will be publicly released.

Requirements are inherently interconnected through various types of dependencies. Identifying these dependencies is essential, as they underpin critical decisions and influence a range of activities throughout software development. However, this task is challenging, particularly in modern software systems, given the high volume of complex, coupled requirements. These challenges are further exacerbated by the ambiguity of Natural Language (NL) requirements and their constant change. Consequently, requirement dependency detection is often overlooked or performed manually. Large Language Models (LLMs) exhibit strong capabilities in NL processing, presenting a promising avenue for requirement-related tasks. While they have shown to enhance various requirements engineering tasks, their effectiveness in identifying requirement dependencies remains unexplored. In this paper, we introduce LEREDD, an LLM-based approach for automated detection of requirement dependencies that leverages Retrieval-Augmented Generation (RAG) and In-Context Learning (ICL). It is designed to identify diverse dependency types directly from NL requirements. We empirically evaluate LEREDD against two state-of-the-art baselines. The results show that LEREDD provides highly accurate classification of dependent and non-dependent requirements, achieving an accuracy of 0.93, and an F1 score of 0.84, with the latter averaging 0.96 for non-dependent cases. LEREDD outperforms zero-shot LLMs and baselines, particularly in detecting fine-grained dependency types, where it yields average relative gains of 94.87% and 105.41% in F1 scores for the Requires dependency over the baselines. We also provide an annotated dataset of requirement dependencies encompassing 813 requirement pairs across three distinct systems to support reproducibility and future research.

Recent work has identified a subset of attention heads in Transformer as retrieval heads, which are responsible for retrieving information from the context. In this work, we first investigate retrieval heads in multilingual contexts. In multilingual language models, we find that retrieval heads are often shared across multiple languages. Expanding the study to cross-lingual setting, we identify Retrieval-Transition heads(RTH), which govern the transition to specific target-language output. Our experiments reveal that RTHs are distinct from retrieval heads and more vital for Chain-of-Thought reasoning in multilingual LLMs. Across four multilingual benchmarks (MMLU-ProX, MGSM, MLQA, and XQuaD) and two model families (Qwen-2.5 and Llama-3.1), we demonstrate that masking RTH induces bigger performance drop than masking Retrieval Heads (RH). Our work advances understanding of multilingual LMs by isolating the attention heads responsible for mapping to target languages.

A reliable action feasibility scorer is a critical bottleneck in embodied agent pipelines: before any planning or reasoning occurs, the agent must identify which candidate actions are physically executable in the current state. Existing approaches use supervised fine-tuning (SFT) to train action scorers, but SFT treats each candidate independently and does not explicitly teach the model to discriminate between actions that are physically correct and those that are subtly wrong. We propose the Contrastive World Model (CWM), which fine-tunes a large language model (LLM) as an action scorer using an InfoNCE contrastive objective with hard-mined negative examples. The key idea is to push valid actions away from invalid ones in scoring space, with special emphasis on hard negatives: semantically similar but physically incompatible candidates. We evaluate CWM on the ScienceWorld benchmark through two studies. First, an intrinsic affordance evaluation on 605 hard-negative test pairs shows that CWM outperforms SFT by +6.76 percentage points on Precision@1 for minimal-edit negatives -- cases where a single word changes the physical outcome -- and achieves a higher AUC-ROC (0.929 vs. 0.906). Second, a live filter characterisation study measures how well CWM ranks gold-path actions against all valid environment actions during task execution. Under out-of-distribution stress conditions, CWM maintains a significantly better safety margin (-2.39) than SFT (-3.96), indicating that the gold action is ranked closer to the top. These results support the hypothesis that contrastive training induces representations that capture physical feasibility more faithfully than SFT alone.

Agentic large language model systems increasingly automate tasks by retrieving URLs and calling external tools. We show that this workflow gives rise to implicit prompt injection: adversarial instructions embedded in automatically generated URL previews, including titles, metadata, and snippets, can introduce a system-level risk that we refer to as silent egress. Using a fully local and reproducible testbed, we demonstrate that a malicious web page can induce an agent to issue outbound requests that exfiltrate sensitive runtime context, even when the final response shown to the user appears harmless. In 480 experimental runs with a qwen2.5:7b-based agent, the attack succeeds with high probability (P (egress) =0.89), and 95% of successful attacks are not detected by output-based safety checks. We also introduce sharded exfiltration, where sensitive information is split across multiple requests to avoid detection. This strategy reduces single-request leakage metrics by 73% (Leak@1) and bypasses simple data loss prevention mechanisms. Our ablation results indicate that defenses applied at the prompt layer offer limited protection, while controls at the system and network layers, such as domain allowlisting and redirect-chain analysis, are considerably more effective. These findings suggest that network egress should be treated as a first-class security outcome in agentic LLM systems. We outline architectural directions, including provenance tracking and capability isolation, that go beyond prompt-level hardening.

Cyberbullying has become a serious and growing concern in todays virtual world. When left unnoticed, it can have adverse consequences for social and mental health. Researchers have explored various types of cyberbullying, but most approaches use single-label classification, assuming that each comment contains only one type of abuse. In reality, a single comment may include overlapping forms such as threats, hate speech, and harassment. Therefore, multilabel detection is both realistic and essential. However, multilabel cyberbullying detection has received limited attention, especially in low-resource languages like Bangla, where robust pre-trained models are scarce. Developing a generalized model with moderate accuracy remains challenging. Transformers offer strong contextual understanding but may miss sequential dependencies, while LSTM models capture temporal flow but lack semantic depth. To address these limitations, we propose a fusion architecture that combines BanglaBERT-Large with a two-layer stacked LSTM. We analyze their behavior to jointly model context and sequence. The model is fine-tuned and evaluated on a publicly available multilabel Bangla cyberbullying dataset covering cyberbully, sexual harassment, threat, and spam. We apply different sampling strategies to address class imbalance. Evaluation uses multiple metrics, including accuracy, precision, recall, F1-score, Hamming loss, Cohens kappa, and AUC-ROC. We employ 5-fold cross-validation to assess the generalization of the architecture.

Community detection in attributed networks faces a fundamental divide: topological algorithms ignore semantic features, while Graph Neural Networks (GNNs) encounter devastating computational bottlenecks. Specifically, GNNs suffer from a Semantic Wall of feature over smoothing in dense or heterophilic networks, and a Systems Wall driven by the O(N^2) memory constraints of pairwise clustering. To dismantle these barriers, we introduce ECHO (Encoding Communities via High order Operators), a scalable, self supervised architecture that reframes community detection as an adaptive, multi scale diffusion process. ECHO features a Topology Aware Router that automatically analyzes structural heuristics sparsity, density, and assortativity to route graphs through the optimal inductive bias, preventing heterophilic poisoning while ensuring semantic densification. Coupled with a memory sharded full batch contrastive objective and a novel chunked O(N \cdot K) similarity extraction method, ECHO completely bypasses traditional O(N^2) memory bottlenecks without sacrificing the mathematical precision of global gradients. Extensive evaluations demonstrate that this topology feature synergy consistently overcomes the classical resolution limit. On synthetic LFR benchmarks scaled up to 1 million nodes, ECHO achieves scale invariant accuracy despite severe topological noise. Furthermore, on massive real world social networks with over 1.6 million nodes and 30 million edges, it completes clustering in mere minutes with throughputs exceeding 2,800 nodes per second matching the speed of highly optimized purely topological baselines. The implementation utilizes a unified framework that automatically engages memory sharded optimization to support adoption across varying hardware constraints. GitHub Repository: https://github.com/emilioferrara/ECHO-GNN

Agent-based AutoML systems rely on large language models to make complex, multi-stage decisions across data processing, model selection, and evaluation. However, existing evaluation practices remain outcome-centric, focusing primarily on final task performance. Through a review of prior work, we find that none of the surveyed agentic AutoML systems report structured, decision-level evaluation metrics intended for post-hoc assessment of intermediate decision quality. To address this limitation, we propose an Evaluation Agent (EA) that performs decision-centric assessment of AutoML agents without interfering with their execution. The EA is designed as an observer that evaluates intermediate decisions along four dimensions: decision validity, reasoning consistency, model quality risks beyond accuracy, and counterfactual decision impact. Across four proof-of-concept experiments, we demonstrate that the EA can (i) detect faulty decisions with an F1 score of 0.919, (ii) identify reasoning inconsistencies independent of final outcomes, and (iii) attribute downstream performance changes to agent decisions, revealing impacts ranging from -4.9\% to +8.3\% in final metrics. These results illustrate how decision-centric evaluation exposes failure modes that are invisible to outcome-only metrics. Our work reframes the evaluation of agentic AutoML systems from an outcome-based perspective to one that audits agent decisions, offering a foundation for reliable, interpretable, and governable autonomous ML systems.

Latent reasoning has been recently proposed as a reasoning paradigm and performs multi-step reasoning through generating steps in the latent space instead of the textual space. This paradigm enables reasoning beyond discrete language tokens by performing multi-step computation in continuous latent spaces. Although there have been numerous studies focusing on improving the performance of latent reasoning, its internal mechanisms remain not fully investigated. In this work, we conduct a comprehensive analysis of latent reasoning methods to better understand the role and behavior of latent representation in the process. We identify two key issues across latent reasoning methods with different levels of supervision. First, we observe pervasive shortcut behavior, where they achieve high accuracy without relying on latent reasoning. Second, we examine the hypothesis that latent reasoning supports BFS-like exploration in latent space, and find that while latent representations can encode multiple possibilities, the reasoning process does not faithfully implement structured search, but instead exhibits implicit pruning and compression. Finally, our findings reveal a trade-off associated with supervision strength: stronger supervision mitigates shortcut behavior but restricts the ability of latent representations to maintain diverse hypotheses, whereas weaker supervision allows richer latent representations at the cost of increased shortcut behavior.

Despite frequent double-blind review, systemic biases related to author demographics still disadvantage underrepresented groups. We start from a simple hypothesis: if a post-review recommender is trained with an explicit fairness regularizer, it should increase inclusion without degrading quality. To test this, we introduce Fair-PaperRec, a Multi-Layer Perceptron (MLP) with a differentiable fairness loss over intersectional attributes (e.g., race, country) that re-ranks papers after double-blind review. We first probe the hypothesis on synthetic datasets spanning high, moderate, and near-fair biases. Across multiple randomized runs, these controlled studies map where increasing the fairness weight strengthens macro/micro diversity while keeping utility approximately stable, demonstrating robustness and adaptability under varying disparity levels. We then carry the hypothesis into the original setting, conference data from ACM Special Interest Group on Computer-Human Interaction (SIGCHI), Designing Interactive Systems (DIS), and Intelligent User Interfaces (IUI). In this real-world scenario, an appropriately tuned configuration of Fair-PaperRec achieves up to a 42.03% increase in underrepresented-group participation with at most a 3.16% change in overall utility relative to the historical selection. Taken together, the synthetic-to-original progression shows that fairness regularization can act as both an equity mechanism and a mild quality regularizer, especially in highly biased regimes. By first analyzing the behavior of the fairness parameters under controlled conditions and then validating them on real submissions, Fair-PaperRec offers a practical, equity-focused framework for post-review paper selection that preserves, and in some settings can even enhance, measured scholarly quality.

Fully Sharded Data Parallel (FSDP), also known as ZeRO, is widely used for training large-scale models, featuring its flexibility and minimal intrusion on model code. However, current FSDP systems struggle with structure-aware training methods (e.g., block-wise quantized training) and with non-element-wise optimizers (e.g., Shampoo and Muon) used in cutting-edge models (e.g., Gemini, Kimi K2). FSDP's fixed element- or row-wise sharding formats conflict with the block-structured computations. In addition, today's implementations fall short in communication and memory efficiency, limiting scaling to tens of thousands of GPUs. We introduce veScale-FSDP, a redesigned FSDP system that couples a flexible sharding format, RaggedShard, with a structure-aware planning algorithm to deliver both flexibility and performance at scale. veScale-FSDP natively supports efficient data placement required by FSDP, empowering block-wise quantization and non-element-wise optimizers. As a result, veScale-FSDP achieves 5~66% higher throughput and 16~30% lower memory usage than existing FSDP systems, while scaling efficiently to tens of thousands of GPUs.

Machine learning training pipelines consume data in batches. A single training step may require thousands of samples drawn from shards distributed across a storage cluster. Issuing thousands of individual GET requests incurs per-request overhead that often dominates data transfer time. To solve this problem, we introduce GetBatch - a new object store API that elevates batch retrieval to a first-class storage operation, replacing independent GET operations with a single deterministic, fault-tolerant streaming execution. GetBatch achieves up to 15x throughput improvement for small objects and, in a production training workload, reduces P95 batch retrieval latency by 2x and P99 per-object tail latency by 3.7x compared to individual GET requests.

Gradient-boosted decision trees are among the strongest off-the-shelf predictors for tabular regression, but point predictions alone do not quantify uncertainty. Conformal prediction provides distribution-free marginal coverage, yet split conformal uses a single global residual quantile and can be poorly adaptive under heteroscedasticity. Methods that improve adaptivity typically fit auxiliary nuisance models or introduce additional data splits/partitions to learn the conformal score, increasing cost and reducing data efficiency. We propose LoBoost, a model-native local conformal method that reuses the fitted ensemble's leaf structure to define multiscale calibration groups. Each input is encoded by its sequence of visited leaves; at resolution level k, we group points by matching prefixes of leaf indices across the first k trees and calibrate residual quantiles within each group. LoBoost requires no retraining, auxiliary models, or extra splitting beyond the standard train/calibration split. Experiments show competitive interval quality, improved test MSE on most datasets, and large calibration speedups.

Millimeter-wave (mmWave) radar captures are band-limited and noisy, making for difficult reconstruction of intelligible full-bandwidth speech. In this work, we propose a two-stage speech reconstruction pipeline for mmWave using a Radar-Aware Dual-conditioned Generative Adversarial Network (RAD-GAN), which is capable of performing bandwidth extension on signals with low signal-to-noise ratios (-5 dB to -1 dB), captured through glass walls. We propose an mmWave-tailored Multi-Mel Discriminator (MMD) and a Residual Fusion Gate (RFG) to enhance the generator input to process multiple conditioning channels. The proposed two-stage pipeline involves pretraining the model on synthetically clipped clean speech and finetuning on fused mel spectrograms generated by the RFG. We empirically show that the proposed method, trained on a limited dataset, with no pre-trained modules, and no data augmentations, outperformed state-of-the-art approaches for this specific task. Audio examples of RAD-GAN are available online at https://rad-gan-demo-site.vercel.app/.

Despite topology optimization producing high-performance structures, late-stage localized revisions remain brittle: direct density-space edits (e.g., warping pixels, inserting holes, swapping infill) can sever load paths and sharply degrade compliance, while re-running optimization is slow and may drift toward a qualitatively different design. We present TopoEdit, a fast post-optimization editor that demonstrates how structured latent embeddings from a pre-trained topology foundation model (OAT) can be repurposed as an interface for physics-aware engineering edits. Given an optimized topology, TopoEdit encodes it into OAT's spatial latent, applies partial noising to preserve instance identity while increasing editability, and injects user intent through an edit-then-denoise diffusion pipeline. We instantiate three edit operators: drag-based topology warping with boundary-condition-consistent conditioning updates, shell-infill lattice replacement using a lattice-anchored reference latent with updated volume-fraction conditioning, and late-stage no-design region enforcement via masked latent overwrite followed by diffusion-based recovery. A consistency-preserving guided DDIM procedure localizes changes while allowing global structural adaptation; multiple candidates can be sampled and selected using a compliance-aware criterion, with optional short SIMP refinement for warps. Across diverse case studies and large edit sweeps, TopoEdit produces intention-aligned modifications that better preserve mechanical performance and avoid catastrophic failure modes compared to direct density-space edits, while generating edited candidates in sub-second diffusion time per sample.

Test-time guidance is a widely used mechanism for steering pretrained diffusion models toward outcomes specified by a reward function. Existing approaches, however, focus on maximizing reward rather than sampling from the true Bayesian posterior, leading to miscalibrated inference. In this work, we show that common test-time guidance methods do not recover the correct posterior distribution and identify the structural approximations responsible for this failure. We then propose consistent alternative estimators that enable calibrated sampling from the Bayesian posterior. We significantly outperform previous methods on a set of Bayesian inference tasks, and match state-of-the-art in black hole image reconstruction.

Retrieval-Augmented Generation (RAG) systems are essential to contemporary AI applications, allowing large language models to obtain external knowledge via vector similarity search. Nevertheless, these systems encounter a significant security flaw: hubness - items that frequently appear in the top-k retrieval results for a disproportionately high number of varied queries. These hubs can be exploited to introduce harmful content, alter search rankings, bypass content filtering, and decrease system performance. We introduce hubscan, an open-source security scanner that evaluates vector indices and embeddings to identify hubs in RAG systems. Hubscan presents a multi-detector architecture that integrates: (1) robust statistical hubness detection utilizing median/MAD-based z-scores, (2) cluster spread analysis to assess cross-cluster retrieval patterns, (3) stability testing under query perturbations, and (4) domain-aware and modality-aware detection for category-specific and cross-modal attacks. Our solution accommodates several vector databases (FAISS, Pinecone, Qdrant, Weaviate) and offers versatile retrieval techniques, including vector similarity, hybrid search, and lexical matching with reranking capabilities. We evaluate hubscan on Food-101, MS-COCO, and FiQA adversarial hubness benchmarks constructed using state-of-the-art gradient-optimized and centroid-based hub generation methods. hubscan achieves 90% recall at a 0.2% alert budget and 100% recall at 0.4%, with adversarial hubs ranking above the 99.8th percentile. Domain-scoped scanning recovers 100% of targeted attacks that evade global detection. Production validation on 1M real web documents from MS MARCO demonstrates significant score separation between clean documents and adversarial content. Our work provides a practical, extensible framework for detecting hubness threats in production RAG systems.

Despite the rapid advancements in Multimodal Large Language Models (MLLMs), a critical question regarding their visual grounding mechanism remains unanswered: do these models genuinely ``read'' text embedded in images, or do they merely rely on parametric shortcuts in the text prompt? In this work, we diagnose this issue by introducing the Visualized-Question (VQ) setting, where text queries are rendered directly onto images to structurally mandate visual engagement. Our diagnostic experiments on Qwen2.5-VL reveal a startling capability-utilization gap: despite possessing strong OCR capabilities, models suffer a performance degradation of up to 12.7% in the VQ setting, exposing a deep-seated ``modality laziness.'' To bridge this gap, we propose SimpleOCR, a plug-and-play training strategy that imposes a structural constraint on the learning process. By transforming training samples into the VQ format with randomized styles, SimpleOCR effectively invalidates text-based shortcuts, compelling the model to activate and optimize its visual text extraction pathways. Empirically, SimpleOCR yields robust gains without architectural modifications. On four representative OOD benchmarks, it surpasses the base model by 5.4% and GRPO based on original images by 2.7%, while exhibiting extreme data efficiency, achieving superior performance with 30x fewer samples (8.5K) than recent RL-based methods. Furthermore, its plug-and-play nature allows seamless integration with advanced RL strategies like NoisyRollout to yield complementary improvements. Code is available at https://github.com/aiming-lab/SimpleOCR.

Agile hardware design flows are a critically needed force multiplier to meet the exploding demand for compute. Recently, agentic generative AI systems have demonstrated significant advances in algorithm design, improving code efficiency, and enabling discovery across scientific domains. Bridging these worlds, we present ArchAgent, an automated computer architecture discovery system built on AlphaEvolve. We show ArchAgent's ability to automatically design/implement state-of-the-art (SoTA) cache replacement policies (architecting new mechanisms/logic, not only changing parameters), broadly within the confines of an established cache replacement policy design competition. In two days without human intervention, ArchAgent generated a policy achieving a 5.3% IPC speedup improvement over the prior SoTA on public multi-core Google Workload Traces. On the heavily-explored single-core SPEC06 workloads, it generated a policy in just 18 days showing a 0.9% IPC speedup improvement over the existing SoTA (a similar "winning margin" as reported by the existing SoTA). ArchAgent achieved these gains 3-5x faster than prior human-developed SoTA policies. Agentic flows also enable "post-silicon hyperspecialization" where agents tune runtime-configurable parameters exposed in hardware policies to further align the policies with a specific workload (mix). Exploiting this, we demonstrate a 2.4% IPC speedup improvement over prior SoTA on SPEC06 workloads. Finally, we outline broader implications for computer architecture research in the era of agentic AI. For example, we demonstrate the phenomenon of "simulator escapes", where the agentic AI flow discovered and exploited a loophole in a popular microarchitectural simulator - a consequence of the fact that these research tools were designed for a (now past) world where they were exclusively operated by humans acting in good-faith.

Do large language models (LLMs) represent concepts abstractly, i.e., independent of input format? We revisit Function Vectors (FVs), compact representations of in-context learning (ICL) tasks that causally drive task performance. Across multiple LLMs, we show that FVs are not fully invariant: FVs are nearly orthogonal when extracted from different input formats (e.g., open-ended vs. multiple-choice), even if both target the same concept. We identify Concept Vectors (CVs), which carry more stable concept representations. Like FVs, CVs are composed of attention head outputs; however, unlike FVs, the constituent heads are selected using Representational Similarity Analysis (RSA) based on whether they encode concepts consistently across input formats. While these heads emerge in similar layers to FV-related heads, the two sets are largely distinct, suggesting different underlying mechanisms. Steering experiments reveal that FVs excel in-distribution, when extraction and application formats match (e.g., both open-ended in English), while CVs generalize better out-of-distribution across both question types (open-ended vs. multiple-choice) and languages. Our results show that LLMs do contain abstract concept representations, but these differ from those that drive ICL performance.

Smooth-basis models such as Chebyshev polynomial regressors and radial basis function (RBF) networks are well established in numerical analysis. Their continuously differentiable prediction surfaces suit surrogate optimisation, sensitivity analysis, and other settings where the response varies gradually with inputs. Despite these properties, smooth models seldom appear in tabular regression, where tree ensembles dominate. We ask whether they can compete, benchmarking models across 55 regression datasets organised by application domain. We develop an anisotropic RBF network with data-driven centre placement and gradient-based width optimisation, a ridge-regularised Chebyshev polynomial regressor, and a smooth-tree hybrid (Chebyshev model tree); all three are released as scikit-learn-compatible packages. We benchmark these against tree ensembles, a pre-trained transformer, and standard baselines, evaluating accuracy alongside generalisation behaviour. The transformer ranks first on accuracy across a majority of datasets, but its GPU dependence, inference latency, and dataset-size limits constrain deployment in the CPU-based settings common across applied science and industry. Among CPU-viable models, smooth models and tree ensembles are statistically tied on accuracy, but the former tend to exhibit tighter generalisation gaps. We recommend routinely including smooth-basis models in the candidate pool, particularly when downstream use benefits from tighter generalisation and gradually varying predictions.

We investigate the collective accuracy of heterogeneous agents who learn to estimate their own reliability over time and selectively abstain from voting. While classical epistemic voting results, such as the \textit{Condorcet Jury Theorem} (CJT), assume fixed participation, real-world aggregation often benefits from allowing agents to say ``I don't know.'' We propose a probabilistic framework where agents engage in a \textit{calibration} phase, updating beliefs about their own fixed competence, before facing a final confidence gate that determines whether to vote or abstain. We derive a non-asymptotic lower bound on the group's success probability and prove that this \textit{selective participation} generalizes the asymptotic guarantees of the CJT to a sequential, confidence-gated setting. Empirically, we validate these bounds via Monte Carlo simulations. While our results are general, we discuss their potential application to AI safety, outlining how this framework can mitigate \textit{hallucinations} in collective LLM decision-making.

In matching markets such as kidney exchanges and freight exchanges, delayed matching has been shown to improve overall market efficiency. The benefits of delay are highly sensitive to participants' sojourn times and departure behavior, and delaying matches can impose significant costs, including longer waiting times and increased market congestion. These competing effects make fixed matching policies inherently inflexible in dynamic environments. We propose a learning-based Hybrid framework that adaptively combines immediate and delayed matching. The framework continuously collects data on user departures over time, estimates the underlying departure distribution via regression, and determines whether to delay matching in the subsequent period based on a decision threshold that governs the system's tolerance for matching efficiency loss. The proposed framework can substantially reduce waiting times and congestion while sacrificing only a limited amount of matching efficiency. By dynamically adjusting its matching strategy, the Hybrid framework enables system performance to flexibly interpolate between purely greedy and purely patient policies, offering a robust and adaptive alternative to static matching mechanisms.

Humans exhibit remarkable flexibility in abstract reasoning, and can rapidly learn and apply rules from sparse examples. To investigate the cognitive strategies underlying this ability, we introduce the Cognitive Abstraction and Reasoning Corpus (CogARC), a diverse human-adapted subset of the Abstraction and Reasoning Corpus (ARC) which was originally developed to benchmark abstract reasoning in artificial intelligence. Across two experiments, CogARC was administered to a total of 260 human participants who freely generated solutions to 75 abstract visual reasoning problems. Success required inferring input-output rules from a small number of examples to transform the test input into one correct test output. Participants' behavior was recorded at high temporal resolution, including example viewing, edit sequences, and multi-attempt submissions. Participants were generally successful (mean accuracy ~90% for experiment 1 (n=40), ~80% for experiment 2 (n=220) across problems), but performance varied widely across problems and participants. Harder problems elicited longer deliberation times and greater divergence in solution strategies. Over the course of the task, participants initiated responses more quickly but showed a slight decline in accuracy, suggesting increased familiarity with the task structure rather than improved rule-learning ability. Importantly, even incorrect solutions were often highly convergent, even when the problem-solving trajectories differed in length and smoothness. Some trajectories progressed directly and efficiently toward a stable outcome, whereas others involved extended exploration or partial restarts before converging. Together, these findings highlight CogARC as a rich behavioral environment for studying human abstract reasoning, providing insight into how people generalize, misgeneralize, and adapt their strategies under uncertainty.

Recent memory agents improve LLMs by extracting experiences and conversation history into an external storage. This enables low-overhead context assembly and online memory update without expensive LLM training. However, existing solutions remain passive and reactive; memory growth is bounded by information that happens to be available, while memory agents seldom seek external inputs in uncertainties. We propose autonomous memory agents that actively acquire, validate, and curate knowledge at a minimum cost. U-Mem materializes this idea via (i) a cost-aware knowledge-extraction cascade that escalates from cheap self/teacher signals to tool-verified research and, only when needed, expert feedback, and (ii) semantic-aware Thompson sampling to balance exploration and exploitation over memories and mitigate cold-start bias. On both verifiable and non-verifiable benchmarks, U-Mem consistently beats prior memory baselines and can surpass RL-based optimization, improving HotpotQA (Qwen2.5-7B) by 14.6 points and AIME25 (Gemini-2.5-flash) by 7.33 points.

Most machine learning models for molecular property prediction rely on a single molecular representation (either a sequence, a graph, or a 3D structure) and treat molecular geometry as static. We present MolFM-Lite, a multi-modal model that jointly encodes SELFIES sequences (1D), molecular graphs (2D), and conformer ensembles (3D) through cross-attention fusion, while conditioning predictions on experimental context via Feature-wise Linear Modulation (FiLM). Our main methodological contributions are: (1) a conformer ensemble attention mechanism that combines learnable attention with Boltzmann-weighted priors over multiple RDKit-generated conformers, capturing the thermodynamic distribution of molecular shapes; and (2) a cross-modal fusion layer where each modality can attend to others, enabling complementary information sharing. We evaluate on four MoleculeNet scaffold-split benchmarks using our model's own splits, and report all baselines re-evaluated under the same protocol. Comprehensive ablation studies across all four datasets confirm that each architectural component contributes independently, with tri-modal fusion providing 7-11% AUC improvement over single-modality baselines and conformer ensembles adding approximately 2% over single-conformer variants. Pre-training on ZINC250K (~250K molecules) using cross-modal contrastive and masked-atom objectives enables effective weight initialization at modest compute cost. We release all code, trained models, and data splits to support reproducibility.

Stereotype repositories are critical to assess generative AI model safety, but currently lack adequate global coverage. It is imperative to prioritize targeted expansion, strategically addressing existing deficits, over merely increasing data volume. This work introduces a multilingual stereotype resource covering four sub-Saharan African countries that are severely underrepresented in NLP resources: Ghana, Kenya, Nigeria, and South Africa. By utilizing socioculturally-situated, community-engaged methods, including telephonic surveys moderated in native languages, we establish a reproducible methodology that is sensitive to the region's complex linguistic diversity and traditional orality. By deliberately balancing the sample across diverse ethnic and demographic backgrounds, we ensure broad coverage, resulting in a dataset of 3,534 stereotypes in English and 3,206 stereotypes across 15 native languages.

Machine learning (ML)-based defect prediction models can improve software quality. However, their opaque reasoning creates an HCI challenge because developers struggle to trust models they cannot interpret. Explainable AI (XAI) methods such as LIME, SHAP, and BreakDown aim to provide transparency, but when used together, they often produce conflicting explanations that increase confusion, frustration, and cognitive load. To address this usability challenge, we introduce XMENTOR, a human-centered, rank-aware aggregation method implemented as a VS Code plugin. XMENTOR unifies multiple post-hoc explanations into a single, coherent view by applying adaptive thresholding, rank and sign agreement, and fallback strategies to preserve clarity without overwhelming users. In a user study, nearly 90% of the participants preferred aggregated explanations, citing reduced confusion and stronger support for daily tasks of debugging and review of defects. Our findings show how combining explanations and embedding them into developer workflows can enhance interpretability, usability, and trust.

As large language models engage in extended reasoning tasks, they accumulate significant state -- architectural mappings, trade-off decisions, codebase conventions -- within the context window. This understanding is lost when sessions reach context limits and undergo lossy compaction. We propose Contextual Memory Virtualisation (CMV), a system that treats accumulated LLM understanding as version-controlled state. Borrowing from operating system virtual memory, CMV models session history as a Directed Acyclic Graph (DAG) with formally defined snapshot, branch, and trim primitives that enable context reuse across independent parallel sessions. We introduce a three-pass structurally lossless trimming algorithm that preserves every user message and assistant response verbatim while reducing token counts by a mean of 20% and up to 86% for sessions with significant overhead by stripping mechanical bloat such as raw tool outputs, base64 images, and metadata. A single-user case-study evaluation across 76 real-world coding sessions demonstrates that trimming remains economically viable under prompt caching, with the strongest gains in mixed tool-use sessions, which average 39% reduction and reach break-even within 10 turns. A reference implementation is available at https://github.com/CosmoNaught/claude-code-cmv.

AI agents -- systems that execute multi-step reasoning workflows with persistent state, tool access, and specialist skills -- represent a qualitative shift from prior automation technologies in social science. Unlike chatbots that respond to isolated queries, AI agents can now read files, run code, query databases, search the web, and invoke domain-specific skills to execute entire research pipelines autonomously. This paper introduces the concept of vibe researching -- the AI-era parallel to ``vibe coding'' (Karpathy, 2025) -- and uses scholar-skill, a 21-skill plugin for Claude Code covering the full research pipeline from idea to submission, as an illustrative case. I develop a cognitive task framework that classifies research activities along two dimensions -- codifiability and tacit knowledge requirement -- to identify a delegation boundary that is cognitive, not sequential: it cuts through every stage of the research pipeline, not between stages. I argue that AI agents excel at speed, coverage, and methodological scaffolding but struggle with theoretical originality and tacit field knowledge. The paper concludes with an analysis of three implications for the profession -- augmentation with fragile conditions, stratification risk, and a pedagogical crisis -- and proposes five principles for responsible vibe researching.

The rise of Antimicrobial Resistance, particularly Multi-Drug Resistance (MDR), presents a critical challenge for clinical decision-making due to limited treatment options and delays in conventional susceptibility testing. This study proposes an interpretable machine learning framework to predict MDR in bacterial isolates using clinical features and antibiotic susceptibility patterns. Five classification models were evaluated, including Logistic Regression, Random Forest, AdaBoost, XGBoost, and LightGBM. The models were trained on a curated dataset of 9,714 isolates, with resistance encoded at the antibiotic family level to capture cross-class resistance patterns consistent with MDR definitions. Performance assessment included accuracy, F1-score, AUC-ROC, and Matthews Correlation Coefficient. Ensemble models, particularly XGBoost and LightGBM, demonstrated superior predictive capability across all metrics. To address the clinical transparency gap, Local Interpretable Model-agnostic Explanations (LIME) was applied to generate instance-level explanations. LIME identified resistance to quinolones, Co-trimoxazole, Colistin, aminoglycosides, and Furanes as the strongest contributors to MDR predictions, aligning with known biological mechanisms. The results show that combining high-performing models with local interpretability provides both accuracy and actionable insights for antimicrobial stewardship. This framework supports earlier MDR identification and enhances trust in machine learning-assisted clinical decision support.

Internet memes have become a dominant form of expression on social media, including within the Bengali-speaking community. While often humorous, memes can also be exploited to spread offensive, harmful, and inflammatory content targeting individuals and groups. Detecting this type of content is excep- tionally challenging due to its satirical, subtle, and culturally specific nature. This problem is magnified for low-resource lan- guages like Bengali, as existing research predominantly focuses on high-resource languages. To address this critical research gap, we introduce Bn-HIB (Bangla Hate Inflammatory Benign), a novel dataset containing 3,247 manually annotated Bengali memes categorized as Benign, Hate, or Inflammatory. Significantly, Bn- HIB is the first dataset to distinguish inflammatory content from direct hate speech in Bengali memes. Furthermore, we propose the MCFM (Multi-Modal Co-Attention Fusion Model), a simple yet effective architecture that mutually analyzes both the visual and textual elements of a meme. MCFM employs a co-attention mechanism to identify and fuse the most critical features from each modality, leading to a more accurate classification. Our experiments show that MCFM significantly outperforms several state-of-the-art models on the Bn-HIB dataset, demonstrating its effectiveness in this nuanced task.Warning: This work contains material that may be disturbing to some audience members. Viewer discretion is advised.

Biological signals of interest in high-dimensional data are often masked by dominant variation shared across conditions. This variation, arising from baseline biological structure or technical effects, can prevent standard dimensionality reduction methods from resolving condition-specific structure. The challenge is that these confounding topics are often unknown and mixed with biological signals. Existing background correction methods are either unscalable to high dimensions or not interpretable. We introduce background contrastive Non-negative Matrix Factorization (\model), which extracts target-enriched latent topics by jointly factorizing a target dataset and a matched background using shared non-negative bases under a contrastive objective that suppresses background-expressed structure. This approach yields non-negative components that are directly interpretable at the feature level, and explicitly isolates target-specific variation. \model is learned by an efficient multiplicative update algorithm via matrix multiplication such that it is highly efficient on GPU hardware and scalable to big data via minibatch training akin to deep learning approach. Across simulations and diverse biological datasets, \model reveals signals obscured by conventional methods, including disease-associated programs in postmortem depressive brain single-cell RNA-seq, genotype-linked protein expression patterns in mice, treatment-specific transcriptional changes in leukemia, and TP53-dependent drug responses in cancer cell lines.

Accurate prediction of malignancy in renal tumors is crucial for informing clinical decisions and optimizing treatment strategies. However, existing imaging modalities lack the necessary accuracy to reliably predict malignancy before surgical intervention. While deep learning has shown promise in malignancy prediction using 3D CT images, traditional approaches often rely on manual segmentation to isolate the tumor region and reduce noise, which enhances predictive performance. Manual segmentation, however, is labor-intensive, costly, and dependent on expert knowledge. In this study, a deep learning framework was developed utilizing an Organ Focused Attention (OFA) loss function to modify the attention of image patches so that organ patches attend only to other organ patches. Hence, no segmentation of 3D renal CT images is required at deployment time for malignancy prediction. The proposed framework achieved an AUC of 0.685 and an F1-score of 0.872 on a private dataset from the UF Integrated Data Repository (IDR), and an AUC of 0.760 and an F1-score of 0.852 on the publicly available KiTS21 dataset. These results surpass the performance of conventional models that rely on segmentation-based cropping for noise reduction, demonstrating the frameworks ability to enhance predictive accuracy without explicit segmentation input. The findings suggest that this approach offers a more efficient and reliable method for malignancy prediction, thereby enhancing clinical decision-making in renal cancer diagnosis.

Recent advances in 4D scene reconstruction have significantly improved dynamic modeling across various domains. However, existing approaches remain limited under aerial conditions with single-view capture, wide spatial range, and dynamic objects of limited spatial footprint and large motion disparity. These challenges cause severe depth ambiguity and unstable motion estimation, making monocular aerial reconstruction inherently ill-posed. To this end, we present AeroDGS, a physics-guided 4D Gaussian splatting framework for monocular UAV videos. AeroDGS introduces a Monocular Geometry Lifting module that reconstructs reliable static and dynamic geometry from a single aerial sequence, providing a robust basis for dynamic estimation. To further resolve monocular ambiguity, we propose a Physics-Guided Optimization module that incorporates differentiable ground-support, upright-stability, and trajectory-smoothness priors, transforming ambiguous image cues into physically consistent motion. The framework jointly refines static backgrounds and dynamic entities with stable geometry and coherent temporal evolution. We additionally build a real-world UAV dataset that spans various altitudes and motion conditions to evaluate dynamic aerial reconstruction. Experiments on synthetic and real UAV scenes demonstrate that AeroDGS outperforms state-of-the-art methods, achieving superior reconstruction fidelity in dynamic aerial environments.

Code summarization is the task of generating natural language descriptions of source code, which is critical for software comprehension and maintenance. While large language models (LLMs) have achieved remarkable progress on this task, an open question remains: can human expertise in code understanding further guide and enhance these models? We propose EyeLayer, a lightweight attention-augmentation module that incorporates human eye-gaze patterns, as a proxy of human expertise, into LLM-based code summarization. EyeLayer models human attention during code reading via a Multimodal Gaussian Mixture, redistributing token embeddings based on learned parameters (μ_i, σ_i^2) that capture where and how intensively developers focus. This design enables learning generalizable attention priors from eye-tracking data and incorporating them into LLMs seamlessly, without disturbing existing representations. We evaluate EyeLayer across diverse model families (i.e., LLaMA-3.2, Qwen3, and CodeBERT) covering different scales and architectures. EyeLayer consistently outperforms strong fine-tuning baselines across standard metrics, achieving gains of up to 13.17% on BLEU-4. These results demonstrate that human gaze patterns encode complementary attention signals that enhance the semantic focus of LLMs and transfer effectively across diverse models for code summarization.

Modern forward electrocardiogram (ECG) computational models rely on an accurate representation of the torso domain. The lead-field method enables fast ECG simulations while preserving full geometric fidelity. Achieving high anatomical accuracy in torso representation is, however, challenging in clinical practice, as imaging protocols are typically focused on the heart and often do not include the entire torso. In addition, the computational cost of the lead-field method scales linearly with the number of electrodes, limiting its applicability in high-density recording settings. To date, no existing approach simultaneously achieves high anatomical fidelity, low data requirements and computational efficiency. In this work, we propose a shape-informed surrogate model of the lead-field operator that serves as a drop-in replacement for the full-order model in forward ECG simulations. The proposed framework consists of two components: a geometry-encoding module that maps anatomical shapes into a low-dimensional latent space, and a geometry-conditioned neural surrogate that predicts lead-field gradients from spatial coordinates, electrode positions and latent codes. The proposed method achieves high accuracy in approximating lead fields both within the torso (mean angular error 5°) and inside the heart, resulting in highly accurate ECG simulations (relative mean squared error <2.5%. The surrogate consistently outperforms the widely used pseudo lead-field approximation while preserving negligible inference cost. Owing to its compact latent representation, the method does not require a fully detailed torso segmentation and can therefore be deployed in data-limited settings while preserving high-fidelity ECG simulations.

This paper tests whether large language models (LLMs) can support interpretative citation context analysis (CCA) by scaling in thick, text-grounded readings of a single hard case rather than scaling up typological labels. It foregrounds prompt-sensitivity analysis as a methodological issue by varying prompt scaffolding and framing in a balanced 2x3 design. Using footnote 6 in Chubin and Moitra (1975) and Gilbert's (1977) reconstruction as a probe, I implement a two-stage GPT-5 pipeline: a citation-text-only surface classification and expectation pass, followed by cross-document interpretative reconstruction using the citing and cited full texts. Across 90 reconstructions, the model produces 450 distinct hypotheses. Close reading and inductive coding identify 21 recurring interpretative moves, and linear probability models estimate how prompt choices shift their frequencies and lexical repertoire. GPT-5's surface pass is highly stable, consistently classifying the citation as "supplementary". In reconstruction, the model generates a structured space of plausible alternatives, but scaffolding and examples redistribute attention and vocabulary, sometimes toward strained readings. Relative to Gilbert, GPT-5 detects the same textual hinges yet more often resolves them as lineage and positioning than as admonishment. The study outlines opportunities and risks of using LLMs as guided co-analysts for inspectable, contestable interpretative CCA, and it shows that prompt scaffolding and framing systematically tilt which plausible readings and vocabularies the model foregrounds.

With the continuous growth of neural network scales, low-precision quantization is widely used in edge accelerators. Classic multi-threshold activation hardware requires 2^n thresholds for n-bit outputs, causing a rapid increase in hardware cost as precision increases. We propose a reconfigurable activation hardware, GRAU, based on piecewise linear fitting, where the segment slopes are approximated by powers of two. Our design requires only basic comparators and 1-bit right shifters, supporting mixed-precision quantization and nonlinear functions such as SiLU. Compared with multi-threshold activators, GRAU reduces LUT consumption by over 90%, achieving higher hardware efficiency, flexibility, and scalability.

Large language models (LLMs) learn contextual embeddings that capture rich semantic information, yet they often overlook structured lexical knowledge such as word senses and relationships. Prior work has shown that incorporating sense dictionaries can improve knowledge distillation for encoder models, but their application to decoder as generative models remains challenging. In this paper, we introduce Decoder-based Sense Knowledge Distillation (DSKD), a framework that integrates lexical resources into the training of decoder-style LLMs without requiring dictionary lookup at inference time. Extensive experiments on diverse benchmarks demonstrate that DSKD significantly enhances knowledge distillation performance for decoders, enabling generative models to inherit structured semantics while maintaining efficient training.

Foundation models in histopathology are expected to facilitate the development of high-performing and generalisable deep learning systems. However, current models capture not only biologically relevant features, but also pre-analytic and scanner-specific variation that bias the predictions of task-specific models trained from the foundation model features. Here we show that introducing novel robustness losses during training of downstream task-specific models reduces sensitivity to technical variability. A purpose-designed comprehensive experimentation setup with 27,042 WSIs from 6155 patients is used to train thousands of models from the features of eight popular foundation models for computational pathology. In addition to a substantial improvement in robustness, we observe that prediction accuracy improves by focusing on biologically relevant features. Our approach successfully mitigates robustness issues of foundation models for computational pathology without retraining the foundation models themselves, enabling development of robust computational pathology models applicable to real-world data in routine clinical practice.

This thesis addresses two persistent and closely related challenges in modern deep learning, reliability and efficiency, through a unified framework grounded in Spectral Geometry and Random Matrix Theory (RMT). As deep networks and large language models continue to scale, their internal behavior becomes increasingly opaque, leading to hallucinations, fragile generalization under distribution shift, and growing computational and energy demands. By analyzing the eigenvalue dynamics of hidden activations across layers and inputs, this work shows that spectral statistics provide a compact, stable, and interpretable lens on model behavior, capable of separating structured, causal representations from noise-dominated variability. Within this framework, the first contribution, EigenTrack, introduces a real-time method for detecting hallucinations and out-of-distribution behavior in large language and vision-language models. EigenTrack transforms streaming activations into spectral descriptors such as entropy, variance, and deviations from the Marchenko-Pastur baseline, and models their temporal evolution using lightweight recurrent classifiers, enabling early detection of reliability failures before they appear in model outputs while offering interpretable insight into representation dynamics. The second contribution, RMT-KD, presents a principled approach to compressing deep networks via random matrix theoretic knowledge distillation. By interpreting outlier eigenvalues in activation spectra as carriers of task-relevant information, RMT-KD progressively projects networks onto lower-dimensional subspaces through iterative self-distillation, yielding significantly more compact and energy-efficient models while preserving accuracy and dense, hardware-friendly structure.

Kurtosis-based Independent Component Analysis (ICA) weakens in wide, balanced mixtures. We prove a sharp redundancy law: for a standardized projection with effective width $R_{\mathrm{eff}}$ (participation ratio), the population excess kurtosis obeys $|κ(y)|=O(κ_{\max}/R_{\mathrm{eff}})$, yielding the order-tight $O(c_bκ_{\max}/R)$ under balance (typically $c_b=O(\log R)$). As an impossibility screen, under standard finite-moment conditions for sample kurtosis estimation, surpassing the $O(1/\sqrt{T})$ estimation scale requires $R\lesssim κ_{\max}\sqrt{T}$. We also show that \emph{purification} -- selecting $m\!\ll\!R$ sign-consistent sources -- restores $R$-independent contrast $Ω(1/m)$, with a simple data-driven heuristic. Synthetic experiments validate the predicted decay, the $\sqrt{T}$ crossover, and contrast recovery.

Frontier AI agents may pursue hidden goals while concealing their pursuit from oversight. Alignment training aims to prevent such behavior by reinforcing the correct goals, but alignment may not always succeed and can lead to unwanted side effects. We propose self-incrimination training, which instead trains agents to produce a visible signal when they covertly misbehave. We train GPT-4.1 and Gemini-2.0 agents to call a report_scheming() tool when behaving deceptively and measure their ability to cause harm undetected in out-of-distribution environments. Self-incrimination significantly reduces the undetected successful attack rate, outperforming matched-capability monitors and alignment baselines while preserving instruction hierarchy and incurring minimal safety tax on general capabilities. Unlike blackbox monitoring, self-incrimination performance is consistent across tasks regardless of how suspicious the misbehavior appears externally. The trained behavior persists under adversarial prompt optimization and generalizes to settings where agents pursue misaligned goals themselves rather than being instructed to misbehave. Our results suggest self-incrimination offers a viable path for reducing frontier misalignment risk, one that neither assumes misbehavior can be prevented nor that it can be reliably classified from the outside.

Traditional software relies on contracts -- APIs, type systems, assertions -- to specify and enforce correct behavior. AI agents, by contrast, operate on prompts and natural language instructions with no formal behavioral specification. This gap is the root cause of drift, governance failures, and frequent project failures in agentic AI deployments. We introduce Agent Behavioral Contracts (ABC), a formal framework that brings Design-by-Contract principles to autonomous AI agents. An ABC contract C = (P, I, G, R) specifies Preconditions, Invariants, Governance policies, and Recovery mechanisms as first-class, runtime-enforceable components. We define (p, delta, k)-satisfaction -- a probabilistic notion of contract compliance that accounts for LLM non-determinism and recovery -- and prove a Drift Bounds Theorem showing that contracts with recovery rate gamma > alpha (the natural drift rate) bound behavioral drift to D* = alpha/gamma in expectation, with Gaussian concentration in the stochastic setting. We establish sufficient conditions for safe contract composition in multi-agent chains and derive probabilistic degradation bounds. We implement ABC in AgentAssert, a runtime enforcement library, and evaluate on AgentContract-Bench, a benchmark of 200 scenarios across 7 models from 6 vendors. Results across 1,980 sessions show that contracted agents detect 5.2-6.8 soft violations per session that uncontracted baselines miss entirely (p < 0.0001, Cohen's d = 6.7-33.8), achieve 88-100% hard constraint compliance, and bound behavioral drift to D* < 0.27 across extended sessions, with 100% recovery for frontier models and 17-100% across all models, at overhead < 10 ms per action.

We study the algorithmic task of testably learning general Massart halfspaces under the Gaussian distribution. In the testable learning setting, the aim is the design of a tester-learner pair satisfying the following properties: (1) if the tester accepts, the learner outputs a hypothesis and a certificate that it achieves near-optimal error, and (2) it is highly unlikely that the tester rejects if the data satisfies the underlying assumptions. Our main result is the first testable learning algorithm for general halfspaces with Massart noise and Gaussian marginals. The complexity of our algorithm is $d^{\mathrm{polylog}(\min\{1/γ, 1/ε\})}$, where $ε$ is the excess error and $γ$ is the bias of the target halfspace, which qualitatively matches the known quasi-polynomial Statistical Query lower bound for the non-testable setting. The analysis of our algorithm hinges on a novel sandwiching polynomial approximation to the sign function with multiplicative error that may be of broader interest.

Video-based ads are a vital medium for brands to engage consumers, with social media platforms leveraging user data to optimize ad delivery and boost engagement. A crucial but under-explored aspect is the 'hooking period', the first three seconds that capture viewer attention and influence engagement metrics. Analyzing this brief window is challenging due to the multimodal nature of video content, which blends visual, auditory, and textual elements. Traditional methods often miss the nuanced interplay of these components, requiring advanced frameworks for thorough evaluation. This study presents a framework using transformer-based multimodal large language models (MLLMs) to analyze the hooking period of video ads. It tests two frame sampling strategies, uniform random sampling and key frame selection, to ensure balanced and representative acoustic feature extraction, capturing the full range of design elements. The hooking video is processed by state-of-the-art MLLMs to generate descriptive analyses of the ad's initial impact, which are distilled into coherent topics using BERTopic for high-level abstraction. The framework also integrates features such as audio attributes and aggregated ad targeting information, enriching the feature set for further analysis. Empirical validation on large-scale real-world data from social media platforms demonstrates the efficacy of our framework, revealing correlations between hooking period features and key performance metrics like conversion per investment. The results highlight the practical applicability and predictive power of the approach, offering valuable insights for optimizing video ad strategies. This study advances video ad analysis by providing a scalable methodology for understanding and enhancing the initial moments of video advertisements.

Current AI weather forecasting models predict conventional atmospheric variables but cannot distinguish between cloud microphysical species critical for aviation safety. We introduce AviaSafe, a hierarchical, physics-informed neural forecaster that produces global, six-hourly predictions of these four hydrometeor species for lead times up to 7 days. Our approach addresses the unique challenges of cloud prediction: extreme sparsity, discontinuous distributions, and complex microphysical interactions between species. We integrate the Icing Condition (IC) index from aviation meteorology as a physics-based constraint that identifies regions where supercooled water fuels explosive ice crystal growth. The model employs a hierarchical architecture that first predicts cloud spatial distribution through masked attention, then quantifies species concentrations within identified regions. Training on ERA5 reanalysis data, our model achieves lower RMSE for cloud species compared to baseline and outperforms operational numerical models on certain key variables at 7-day lead times. The ability to forecast individual cloud species enables new applications in aviation route optimization where distinguishing between ice and liquid water determines engine icing risk.

Reinforcement learning (RL) offers significant promise for machinery fault detection (MFD). However, most existing RL-based MFD approaches do not fully exploit RL's sequential decision-making strengths, often treating MFD as a simple guessing game (Contextual Bandits). To bridge this gap, we formulate MFD as an offline inverse reinforcement learning problem, where the agent learns the reward dynamics directly from healthy operational sequences, thereby bypassing the need for manual reward engineering and fault labels. Our framework employs Adversarial Inverse Reinforcement Learning to train a discriminator that distinguishes between normal (expert) and policy-generated transitions. The discriminator's learned reward serves as an anomaly score, indicating deviations from normal operating behaviour. When evaluated on three run-to-failure benchmark datasets (HUMS2023, IMS, and XJTU-SY), the model consistently assigns low anomaly scores to normal samples and high scores to faulty ones, enabling early and robust fault detection. By aligning RL's sequential reasoning with MFD's temporal structure, this work opens a path toward RL-based diagnostics in data-driven industrial settings.

Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning abilities of large language models (LLMs) on mathematics and programming tasks, but standard approaches that optimize single-attempt accuracy can inadvertently suppress response diversity across repeated attempts, narrowing exploration and overlooking underrepresented strategies. We introduce UpSkill, a training time method that adapts Mutual Information Skill Learning (MISL) to LLMs for optimizing pass@k correctness. We propose a novel reward that we implement within Group Relative Policy Optimization (GRPO): a token-level mutual information (MI) reward that encourages trajectory specificity to z. Experiments on GSM8K with three open-weight models, Llama 3.1-8B, Qwen 2.5-7B, and R1-Distilled-Qwen2.5-Math-1.5B, show that UpSkill improves multi-attempt metrics on the stronger base models, yielding mean gains of ~3% in pass@k for both Qwen and Llama without degrading pass@1. Additionally, we find both empirical and theoretical evidence that improvements in pass@k are closely tied to the mutual information objective.

Models operating on dynamic physiologic signals must distinguish benign, label-preserving variability from true concept change. Existing concept-drift frameworks are largely distributional and provide no principled guidance on how much a model's internal representation may move when the underlying signal undergoes physiologically plausible fluctuations in energy. As a result, deep models often misinterpret harmless changes in amplitude, rate, or morphology as concept drift, yielding unstable predictions, particularly in multimodal fusion settings. This study introduces Physiologic Energy Conservation Theory (PECT), an energy-based framework for concept stability in dynamic signals. PECT posits that under virtual drift, normalized latent displacement should scale proportionally with normalized signal energy change, while persistent violations of this proportionality indicate real concept drift. We operationalize this principle through Energy-Constrained Representation Learning (ECRL), a lightweight regularizer that penalizes energy-inconsistent latent movement without modifying encoder architectures or adding inference-time cost. Although PECT is formulated for dynamic signals in general, we instantiate and evaluate it on multimodal ECG across seven unimodal and hybrid models. Experiments show that in the strongest trimodal hybrid (1D+2D+Transformer), clean accuracy is largely preserved (96.0% to 94.1%), while perturbed accuracy improves substantially (72.6% to 85.5%) and fused representation drift decreases by over 45%. Similar trends are observed across all architectures, providing empirical evidence that PECT functions as an energy-drift law governing concept stability in continuous physiologic signals.

River systems operate as inherently interconnected continuous networks, meaning river hydrodynamic simulation ought to be a systemic process. However, widespread hydrology data scarcity often restricts data-driven forecasting to isolated predictions. To achieve systemic simulation and reduce reliance on river observations, we present GraphRiverCast (GRC), a topology-informed AI foundation model designed to simulate multivariate river hydrodynamics in global river systems. GRC is capable of operating in a "ColdStart" mode, generating predictions without relying on historical river states for initialization. In 7-day global pseudo-hindcasts, GRC-ColdStart functions as a robust standalone simulator, achieving a Nash-Sutcliffe Efficiency (NSE) of approximately 0.82 without exhibiting the significant error accumulation typical of autoregressive paradigms. Ablation studies reveal that topological encoding serves as indispensable structural information in the absence of historical states, explicitly guiding hydraulic connectivity and network-scale mass redistribution to reconstruct flow dynamics. Furthermore, when adapted locally via a pre-training and fine-tuning strategy, GRC consistently outperforms physics-based and locally-trained AI baselines. Crucially, this superiority extends from gauged reaches to full river networks, underscoring the necessity of topology encoding and physics-based pre-training. Built on a physics-aligned neural operator architecture, GRC enables rapid and cross-scale adaptive simulation, establishing a collaborative paradigm bridging global hydrodynamic knowledge with local hydrological reality.

Peer review remains a cornerstone in academia, yet it frequently falls short in fostering joint progress and well-being. While peer review primarily emphasizes scientific rigor, it often lacks the empathy essential for supporting and encouraging all peers involved. In this experience report, I aim to highlight that peer review is a practice that demands both scientific care for quality and collegial welfare for the joint progress and well-being of all peers involved, including authors, co-reviewers, workshop or conference organizers, and journal editors. Drawing on my ten years of experience in academia, I propose the ethos of the PEERfect REVIEWer, grounded in the two core values: Scientific care and collegial welfare. Through reflection shaped by professional exchanges with colleagues, consideration of literature, and an examination of both self-authored and received reviews, I formulated an accompanying guideline with 16 practical recommendations to guide reviewers in their actions to achieve these two values. The ethos of the PEERfect REVIEWer and its accompanying guideline help reviewers in upholding high scientific standards and conducting peer review in a constructive, supportive, respectful, and timely manner. They demonstrate that scientific rigor and empathy are complementary forces that promote impactful peer review practice. By placing scientific care and collegial welfare at the core of peer review, this experience report reaffirms the importance of scientific rigor while also advocating for greater attention to empathy. It invites reviewers to reconsider their role not merely as gatekeepers but as partners in the academic journey of each peer involved. The PEERfect REVIEWer is both a caretaker of quality and a steward of joint progress and well-being - as truly impactful peer review practice requires scientific rigor and empathy in equal measure.

Stiff differential equations pose a major challenge for Physics-Informed Neural Networks (PINNs), often causing poor convergence. We propose a simple, hyperparameter-free method to address stiffness by normalizing loss residuals with the Jacobian. We provide theoretical indications that Jacobian-based normalization can improve gradient descent and validate it on benchmark stiff ordinary differential equations. We then apply it to a realistic system: the stiff Boltzmann equations (BEs) governing weakly interacting massive particle (WIMP) dark matter (DM). Our approach achieves higher accuracy than attention mechanisms previously proposed for handling stiffness, recovering the full solution where prior methods fail. This is further demonstrated in an inverse problem with a single experimental data point - the observed DM relic density - where our inverse PINNs correctly infer the cross section that solves the BEs in both Standard and alternative cosmologies.

While prior work has focused on projecting adversarial examples back onto the manifold of natural data to restore safety, we argue that a comprehensive understanding of AI safety requires characterizing the unsafe regions themselves. This paper introduces a framework for systematically mapping the Manifold of Failure in Large Language Models (LLMs). We reframe the search for vulnerabilities as a quality diversity problem, using MAP-Elites to illuminate the continuous topology of these failure regions, which we term behavioral attraction basins. Our quality metric, Alignment Deviation, guides the search towards areas where the model's behavior diverges most from its intended alignment. Across three LLMs: Llama-3-8B, GPT-OSS-20B, and GPT-5-Mini, we show that MAP-Elites achieves up to 63% behavioral coverage, discovers up to 370 distinct vulnerability niches, and reveals dramatically different model-specific topological signatures: Llama-3-8B exhibits a near-universal vulnerability plateau (mean Alignment Deviation 0.93), GPT-OSS-20B shows a fragmented landscape with spatially concentrated basins (mean 0.73), and GPT-5-Mini demonstrates strong robustness with a ceiling at 0.50. Our approach produces interpretable, global maps of each model's safety landscape that no existing attack method (GCG, PAIR, or TAP) can provide, shifting the paradigm from finding discrete failures to understanding their underlying structure.

When biological foundation models such as scGPT and Geneformer process single-cell gene expression, what geometric and topological structure forms in their internal representations? Is that structure biologically meaningful or a training artifact, and how confident should we be in such claims? We address these questions through autonomous large-scale hypothesis screening: an AI-driven executor-brainstormer loop that proposed, tested, and refined 141 geometric and topological hypotheses across 52 iterations, covering persistent homology, manifold distances, cross-model alignment, community structure, and directed topology, all with explicit null controls and disjoint gene-pool splits. Three principal findings emerge. First, the models learn genuine geometric structure. Gene embedding neighborhoods exhibit non-trivial topology, with persistent homology significant in 11 of 12 transformer layers at p < 0.05 in the weakest domain and 12 of 12 in the other two. A multi-level distance hierarchy shows that manifold-aware metrics outperform Euclidean distance for identifying regulatory gene pairs, and graph community partitions track known transcription factor target relationships. Second, this structure is shared across independently trained models. CCA alignment between scGPT and Geneformer yields canonical correlation of 0.80 and gene retrieval accuracy of 72 percent, yet none of 19 tested methods reliably recover gene-level correspondences. The models agree on the global shape of gene space but not on precise gene placement. Third, the structure is more localized than it first appears. Under stringent null controls applied across all null families, robust signal concentrates in immune tissue, while lung and external lung signals weaken substantially.

Large language models (LLMs) demonstrate remarkable breadth of knowledge, yet their ability to reason about computational processes remains poorly understood. Closing this gap matters for practitioners who rely on LLMs to guide algorithm selection and deployment. We address this limitation using causal discovery as a testbed and evaluate eight frontier LLMs against ground truth derived from large-scale algorithm executions and find systematic, near-total failure. Models produce ranges far wider than true confidence intervals yet still fail to contain the true algorithmic mean in the majority of instances; most perform worse than random guessing and the marginal above-random performance of the best model is most consistent with benchmark memorization rather than principled reasoning. We term this failure algorithmic blindness and argue it reflects a fundamental gap between declarative knowledge about algorithms and calibrated procedural prediction.

Sudden cardiac death (SCD) is unpredictable, and its prediction in Chagas cardiomyopathy (CC) remains a significant challenge, especially in patients not classified as high risk. While AI and machine learning models improve risk stratification, their adoption is hindered by a lack of transparency, as they are often perceived as \textit{black boxes} with unclear decision-making processes. Some approaches apply heuristic explanations without correctness guarantees, leading to mistakes in the decision-making process. To address this, we apply a logic-based explainability method with correctness guarantees to the problem of SCD prediction in CC. This explainability method, applied to an AI classifier with over 95\% accuracy and recall, demonstrated strong predictive performance and 100\% explanation fidelity. When compared to state-of-the-art heuristic methods, it showed superior consistency and robustness. This approach enhances clinical trust, facilitates the integration of AI-driven tools into practice, and promotes large-scale deployment, particularly in endemic regions where it is most needed.

Abstractions of causal models allow for the coarsening of models such that relations of cause and effect are preserved. Whereas abstractions focus on the relation between two models, in this paper we study a framework for causal embeddings which enable multiple detailed models to be mapped into sub-systems of a coarser causal model. We define causal embeddings as a generalization of abstraction, and present a generalized notion of consistency. By defining a multi-resolution marginal problem, we showcase the relevance of causal embeddings for both the statistical marginal problem and the causal marginal problem; furthermore, we illustrate its practical use in merging datasets coming from models with different representations.

Lossless compression is essential for efficient data storage and transmission. Although learning-based lossless compressors achieve strong results, most of them are designed for a single modality, leading to redundant compressor deployments in multi-modal settings. Designing a unified multi-modal compressor is critical yet challenging, as different data types vary largely in format, dimension, and statistics. Multi-modal large language models offer a promising resolution but remain too complex for practical use. Thus, we propose \textbf{OmniZip}, \textbf{a unified and lightweight lossless compressor for multi-modal data (like image, text, speech, tactile, database, and gene sequence)}. Built on a lightweight backbone, OmniZip incorporates three key components to enable efficient multi-modal lossless compression: a modality-unified tokenizer that reversibly transforms diverse data into tokens, a modality-routing context learning mechanism that enables flexible multi-modal context modeling, and a modality-routing feedforward design that further enhances the model's nonlinear representation flexibility. A reparameterization training strategy is used to enhance model capacity. OmniZip outperforms or matches other state-of-the-art compressors on multiple modalities, achieving 42\%, 57\%, 62\% and 42\%, 53\% higher compression efficiency than gzip on CLIC-M, TouchandGo, enwik9, LibriSpeech, and WikiSQL datasets, respectively. It also supports near real-time inference on resource-constrained edge devices, reaching about 1MB/s on MacBook CPUs and iPhone NPUs. Our code is released at https://github.com/adminasmi/OmniZip-CVPR2026.

Objective: The objective of this study is to develop a machine learning (ML)-based framework for early risk stratification of clinical trials (CTs) according to their likelihood of exhibiting a high rate of dosing errors, using information available prior to trial initiation. Materials and Methods: We constructed a dataset from ClinicalTrials.gov comprising 42,112 CTs. Structured, semi-structured trial data, and unstructured protocol-related free-text data were extracted. CTs were assigned binary labels indicating elevated dosing error rate, derived from adverse event reports, MedDRA terminology, and Wilson confidence intervals. We evaluated an XGBoost model trained on structured features, a ClinicalModernBERT model using textual data, and a simple late-fusion model combining both modalities. Post-hoc probability calibration was applied to enable interpretable, trial-level risk stratification. Results: The late-fusion model achieved the highest AUC-ROC (0.862). Beyond discrimination, calibrated outputs enabled robust stratification of CTs into predefined risk categories. The proportion of trials labeled as having an excessively high dosing error rate increased monotonically across higher predicted risk groups and aligned with the corresponding predicted probability ranges. Discussion: These findings indicate that dosing error risk can be anticipated at the trial level using pre-initiation information. Probability calibration was essential for translating model outputs into reliable and interpretable risk categories, while simple multimodal integration yielded performance gains without requiring complex architectures. Conclusion: This study introduces a reproducible and scalable ML framework for early, trial-level risk stratification of CTs at risk of high dosing error rates, supporting proactive, risk-based quality management in clinical research.

Recent advancements in deep learning have actively addressed complex challenges within the Computer-Aided Design (CAD) domain.However, most existing approaches rely on task-specifi c models requiring structural modifi cations for new tasks, and they predominantly focus on point clouds or images rather than the industry-standard Boundary Representation (B-rep) format. To address these limitations, we propose BrepCoder, a unifi ed Multimodal Large Language Model (MLLM) that performs diverse CAD tasks from B-rep inputs. By leveraging the code generation capabilities of Large Language Models (LLMs), we convert CAD modeling sequences into Python-like code and align them with B-rep. We then adopt a two-stage training strategy: First, pre-training on reverse engineering to learn geometric features and design logic. Second, eff ectively extending the model to various downstream tasks such as completion, error correction, and CAD-QA. Consequently, by interpreting B-rep as structural code, BrepCoder achieves superior generalization across diverse tasks, demonstrating its potential as a general-purpose CAD agent.

Multimodal large language models (MLLMs) have made significant progress in mobile agent development, yet their capabilities are predominantly confined to a reactive paradigm, where they merely execute explicit user commands. The emerging paradigm of proactive intelligence, where agents autonomously anticipate needs and initiate actions, represents the next frontier for mobile agents. However, its development is critically bottlenecked by the lack of benchmarks that can address real-world complexity and enable objective, executable evaluation. To overcome these challenges, we introduce ProactiveMobile, a comprehensive benchmark designed to systematically advance research in this domain. ProactiveMobile formalizes the proactive task as inferring latent user intent across four dimensions of on-device contextual signals and generating an executable function sequence from a comprehensive function pool of 63 APIs. The benchmark features over 3,660 instances of 14 scenarios that embrace real-world complexity through multi-answer annotations. To ensure quality, a team of 30 experts conducts a final audit of the benchmark, verifying factual accuracy, logical consistency, and action feasibility, and correcting any non-compliant entries. Extensive experiments demonstrate that our fine-tuned Qwen2.5-VL-7B-Instruct achieves a success rate of 19.15%, outperforming o1 (15.71%) and GPT-5 (7.39%). This result indicates that proactivity is a critical competency widely lacking in current MLLMs, yet it is learnable, emphasizing the importance of the proposed benchmark for proactivity evaluation.

Data privacy is important in the AI era, and differential privacy (DP) is one of the golden solutions. However, DP is typically applicable only if data have a bounded underlying distribution. We address this limitation by leveraging second-moment information from a small amount of public data. We propose Public-moment-guided Truncation (PMT), which transforms private data using the public second-moment matrix and applies a principled truncation whose radius depends only on non-private quantities: data dimension and sample size. This transformation yields a well-conditioned second-moment matrix, enabling its inversion with a significantly strengthened ability to resist the DP noise. Furthermore, we demonstrate the applicability of PMT by using penalized and generalized linear regressions. Specifically, we design new loss functions and algorithms, ensuring that solutions in the transformed space can be mapped back to the original domain. We have established improvements in the models' DP estimation through theoretical error bounds, robustness guarantees, and convergence results, attributing the gains to the conditioning effect of PMT. Experiments on synthetic and real datasets confirm that PMT substantially improves the accuracy and stability of DP models.

LLM agents have been widely adopted in real-world applications, relying on agent frameworks for workflow execution and multi-agent coordination. As these systems scale, understanding bugs in the underlying agent frameworks becomes critical. However, existing work mainly focuses on agent-level failures, overlooking framework-level bugs. To address this gap, we conduct an empirical study of 998 bug reports from CrewAI and LangChain, constructing a taxonomy of 15 root causes and 7 observable symptoms across five agent lifecycle stages: 'Agent Initialization','Perception', 'Self-Action', 'Mutual Interaction' and 'Evolution'. Our findings show that agent framework bugs mainly arise from 'API misuse', 'API incompatibility', and 'Documentation Desync', largely concentrated in the 'Self-Action' stage. Symptoms typically appear as 'Functional Error', 'Crash', and 'Build Failure', reflecting disruptions to task progression and control flow.

Cardiovascular disease is the primary cause of death globally, necessitating early identification, precise risk classification, and dependable decision-support technologies. The advent of large language models (LLMs) provides new zero-shot and few-shot reasoning capabilities, even though machine learning (ML) algorithms, especially ensemble approaches like Random Forest, XGBoost, LightGBM, and CatBoost, are excellent at modeling complex, non-linear patient data and routinely beat logistic regression. This research predicts cardiovascular disease using a merged dataset of 1,190 patient records, comparing traditional machine learning models (95.78% accuracy, ROC-AUC 0.96) with open-source large language models via OpenRouter APIs. Finally, a hybrid fusion of the ML ensemble and LLM reasoning under Gemini 2.5 Flash achieved the best results (96.62% accuracy, 0.97 AUC), showing that LLMs (78.9 % accuracy) work best when combined with ML models rather than used alone. Results show that ML ensembles achieved the highest performance (95.78% accuracy, ROC-AUC 0.96), while LLMs performed moderately in zero-shot (78.9%) and slightly better in few-shot (72.6%) settings. The proposed hybrid method enhanced the strength in uncertain situations, illustrating that ensemble ML is considered the best structured tabular prediction case, but it can be integrated with hybrid ML-LLM systems to provide a minor increase and open the way to more reliable clinical decision-support tools.

Learning based methods are now ubiquitous for solving inverse problems, but their deployment in real-world applications is often hindered by the lack of ground truth references for training. Recent self-supervised learning strategies offer a promising alternative, avoiding the need for ground truth. However, most existing methods are limited to linear inverse problems. This work extends self-supervised learning to the non-linear problem of recovering audio and images from clipped measurements, by assuming that the signal distribution is approximately invariant to changes in amplitude. We provide sufficient conditions for learning to reconstruct from saturated signals alone and a self-supervised loss that can be used to train reconstruction networks. Experiments on both audio and image data show that the proposed approach is almost as effective as fully supervised approaches, despite relying solely on clipped measurements for training.

Multimodal information retrieval (MMIR) has gained attention for its flexibility in handling text, images, or mixed queries and candidates. Recent breakthroughs in multimodal large language models (MLLMs) boost MMIR performance by incorporating MLLM knowledge under the contrastive finetuning framework. However, they suffer from pre-training inconsistency and require large datasets. In this work, we introduce a novel framework, RetLLM, designed to query MLLMs for MMIR in a training- and data-free manner. Specifically, we formulate MMIR as a similarity score generation task and prompt MLLMs to directly predict retrieval scores in a coarse-then-fine pipeline. At the coarse stage, a top-k filtering strategy builds a small yet high-quality candidate pool for each query, enabling MLLMs to focus on semantically relevant candidates. Subsequently, the retrieval score is predicted by feeding both the query and candidate into MLLMs at the fine stage. Importantly, we propose a visual enhancement module during reasoning to help MLLMs re-pick forgotten visuals, improving retrieval. Extensive experiments on MMIR benchmarks show that RetLLM outperforms fine-tuned models. Ablation studies further verify each component. Our work demonstrates that MLLMs can achieve strong MMIR performance without any training, highlighting their inherent multimodal reasoning ability in a simple, scalable framework. We release our code at: https://github.com/alivecat05/RETLLM

Neural routing solvers (NRSs) that leverage deep learning to tackle vehicle routing problems have demonstrated notable potential for practical applications. By learning implicit heuristic rules from data, NRSs replace the handcrafted counterparts in classic heuristic frameworks, thereby reducing reliance on costly manual design and trial-and-error adjustments. This survey makes two main contributions: (1) The heuristic nature of NRSs is highlighted, and existing NRSs are reviewed from the perspective of heuristics. A hierarchical taxonomy based on heuristic principles is further introduced. (2) A generalization-focused evaluation pipeline is proposed to address limitations of the conventional pipeline. Comparative benchmarking of representative NRSs across both pipelines uncovers a series of previously unreported gaps in current research.

AI-native architectures are vital for 6G wireless communications. The black-box nature and high complexity of deep learning models employed in critical applications, such as channel estimation, limit their practical deployment. While perturbation-based XAI solutions offer input filtering, they often neglect internal structural optimization. We propose X-REFINE, an XAI-based framework for joint input-filtering and architecture fine-tuning. By utilizing a decomposition-based, sign-stabilized LRP epsilon rule, X-REFINE backpropagates predictions to derive high-resolution relevance scores for both subcarriers and hidden neurons. This enables a holistic optimization that identifies the most faithful model components. Simulation results demonstrate that X-REFINE achieves a superior interpretability-performance-complexity trade-off, significantly reducing computational complexity while maintaining robust bit error rate (BER) performance across different scenarios.

Software engineering (SE) and requirements engineering (RE) face a significant increase in secondary studies, particularly literature reviews (LRs), due to the ever-growing number of scientific publications. Generative artificial intelligence (GenAI) exacerbates this trend by producing LRs rapidly but often at the expense of quality, rigor, and transparency. At the same time, secondary studies often fail to share underlying data and artifacts, limiting replication and reuse. This paper introduces EmpiRE-Compass, a neuro-symbolic dashboard designed to lower barriers for accessing, replicating, and reusing LR data. Its overarching goal is to demonstrate how LRs can become more sustainable by semantically structuring their underlying data in research knowledge graphs (RKGs) and by leveraging large language models (LLMs) for easy and dynamic access, replication, and reuse. Building on two RE use cases, we developed EmpiRE-Compass with a modular system design and workflows for curated and custom competency questions. The dashboard is freely available online, accompanied by a demonstration video. To manage operational costs, a limit of 25 requests per IP address per day applies to the default LLM (GPT-4o mini). All source code and documentation are released as an open-source project to foster reuse, adoption, and extension. EmpiRE-Compass provides three core capabilities: (1) Exploratory visual analytics for curated competency questions; (2) Neuro-symbolic synthesis for custom competency questions; and (3) Reusable knowledge with all queries, analyses, and results openly available. By unifying RKGs and LLMs in a neuro-symbolic dashboard, EmpiRE-Compass advances sustainable LRs in RE, SE, and beyond. It lowers technical barriers, fosters transparency and reproducibility, and enables collaborative, continuously updated, and reusable LRs

A common approach to compute distances on continuous surfaces is by considering a discretized polygonal mesh approximating the surface and estimating distances on the polygon. We show that exact geodesic distances restricted to the polygon are at most second-order accurate with respect to the distances on the corresponding continuous surface. By order of accuracy we refer to the convergence rate as a function of the average distance between sampled points. Next, a higher-order accurate deep learning method for computing geodesic distances on surfaces is introduced. Traditionally, one considers two main components when computing distances on surfaces: a numerical solver that locally approximates the distance function, and an efficient causal ordering scheme by which surface points are updated. Classical minimal path methods often exploit a dynamic programming principle with quasi-linear computational complexity in the number of sampled points. The quality of the distance approximation is determined by the local solver that is revisited in this paper. To improve state of the art accuracy, we consider a neural network-based local solver which implicitly approximates the structure of the continuous surface. We supply numerical evidence that the proposed learned update scheme provides better accuracy compared to the best possible polyhedral approximations and previous learning-based methods. The result is a third-order accurate solver with a bootstrapping-recipe for further improvement.

Traffic flow forecasting has emerged as an indispensable mission for daily life, which is required to utilize the spatiotemporal relationship between each location within a time period under a graph structure to predict future flow. However, the large travel demand for broader geographical areas and longer time spans requires models to distinguish each node clearly and possess a holistic view of the history, which has been paid less attention to in prior works. Furthermore, increasing sizes of data hinder the deployment of most models in real application environments. To this end, in this paper, we propose a lightweight Positional-aware Spatio-Temporal Network (PASTN) to effectively capture both temporal and spatial complexities in an end-to-end manner. PASTN introduces positional-aware embeddings to separate each node's representation, while also utilizing a temporal attention module to improve the long-range perception of current models. Extensive experiments verify the effectiveness and efficiency of PASTN across datasets of various scales (county, megalopolis and state). Further analysis demonstrates the efficacy of newly introduced modules either.

We introduce FIRE, a comprehensive benchmark designed to evaluate both the theoretical financial knowledge of LLMs and their ability to handle practical business scenarios. For theoretical assessment, we curate a diverse set of examination questions drawn from widely recognized financial qualification exams, enabling evaluation of LLMs deep understanding and application of financial knowledge. In addition, to assess the practical value of LLMs in real-world financial tasks, we propose a systematic evaluation matrix that categorizes complex financial domains and ensures coverage of essential subdomains and business activities. Based on this evaluation matrix, we collect 3,000 financial scenario questions, consisting of closed-form decision questions with reference answers and open-ended questions evaluated by predefined rubrics. We conduct comprehensive evaluations of state-of-the-art LLMs on the FIRE benchmark, including XuanYuan 4.0, our latest financial-domain model, as a strong in-domain baseline. These results enable a systematic analysis of the capability boundaries of current LLMs in financial applications. We publicly release the benchmark questions and evaluation code to facilitate future research.

Self-attention is usually described as a flexible, content-adaptive way to mix a token with information from its past. We re-interpret causal self-attention transformers, the backbone of modern foundation models, within a probabilistic framework, much like how classical PCA is extended to probabilistic PCA. However, this re-formulation reveals a surprising and deeper structural insight: due to a change-of-variables phenomenon, a barrier constraint emerges on the self-attention parameters. This induces a highly structured geometry on the token space, providing theoretical insights into the dynamics of LLM decoding. This reveals a boundary where attention becomes ill-conditioned, leading to a margin interpretation similar to classical support vector machines. Just like support vectors, this naturally gives rise to the concept of support tokens. Furthermore, we show that LLMs can be interpreted as a stochastic process over the power set of the token space, providing a rigorous probabilistic framework for sequence modeling. We propose a Bayesian framework and derive a MAP estimation objective that requires only a minimal modification to standard LLM training: the addition of a smooth log-barrier penalty to the usual cross-entropy loss. We demonstrate that this provides more robust models without sacrificing out-of-sample accuracy and that it is straightforward to incorporate in practice.

Multi-robot task planning requires decomposing natural-language instructions into executable actions for heterogeneous robot teams. Conventional Planning Domain Definition Language (PDDL) planners provide rigorous guarantees but struggle to handle ambiguous or long-horizon missions, while large language models (LLMs) can interpret instructions and propose plans but may hallucinate or produce infeasible actions. We present a hierarchical multi-agent LLM-based planner with prompt optimization: an upper layer decomposes tasks and assigns them to lower-layer agents, which generate PDDL problems solved by a classical planner. When plans fail, the system applies TextGrad-inspired textual-gradient updates to optimize each agent's prompt and thereby improve planning accuracy. In addition, meta-prompts are learned and shared across agents within the same layer, enabling efficient prompt optimization in multi-agent settings. On the MAT-THOR benchmark, our planner achieves success rates of 0.95 on compound tasks, 0.84 on complex tasks, and 0.60 on vague tasks, improving over the previous state-of-the-art LaMMA-P by 2, 7, and 15 percentage points respectively. An ablation study shows that the hierarchical structure, prompt optimization, and meta-prompt sharing contribute roughly +59, +37, and +4 percentage points to the overall success rate.

Spatio-temporal epidemic forecasting is critical for public health management, yet existing methods often struggle with insensitivity to weak epidemic signals, over-simplified spatial relations, and unstable parameter estimation. To address these challenges, we propose the Spatio-Temporal priOr-aware Epidemic Predictor (STOEP), a novel hybrid framework that integrates implicit spatio-temporal priors and explicit expert priors. STOEP consists of three key components: (1) Case-aware Adjacency Learning (CAL), which dynamically adjusts mobility-based regional dependencies using historical infection patterns; (2) Space-informed Parameter Estimating (SPE), which employs learnable spatial priors to amplify weak epidemic signals; and (3) Filter-based Mechanistic Forecasting (FMF), which uses an expert-guided adaptive thresholding strategy to regularize epidemic parameters. Extensive experiments on real-world COVID-19 and influenza datasets demonstrate that STOEP outperforms the best baseline by 11.1% in RMSE. The system has been deployed at one provincial CDC in China to facilitate downstream applications.

Federated Learning (FL) enables collaborative model training without sharing raw data. However, shared local model updates remain vulnerable to inference and poisoning attacks. Secure aggregation schemes have been proposed to mitigate these attacks. In this work, we aim to understand how these techniques are implemented in quantum-assisted FL. Quantum Secure Aggregation (QSA) has been proposed, offering information-theoretic privacy by encoding client updates into the global phase of multipartite entangled states. Existing QSA protocols, however, rely on a single global Greenberger-Horne-Zeilinger (GHZ) state shared among all participating clients. This design poses fundamental challenges: fidelity of large-scale GHZ states deteriorates rapidly with the increasing number of clients; and (ii) the global aggregation prevents the detection of Byzantine clients. We propose Clustered Quantum Secure Aggregation (CQSA), a modular aggregation framework that reconciles the physical constraints of near-term quantum hardware along with the need for Byzantine-robustness in FL. CQSA randomly partitions the clients into small clusters, each performing local quantum aggregation using high-fidelity, low-qubit GHZ states. The server analyzes statistical relationships between cluster-level aggregates employing common statistical measures such as cosine similarity and Euclidean distance to identify malicious contributions. Through theoretical analysis and simulations under depolarizing noise, we demonstrate that CQSA ensures stable model convergence, achieves superior state fidelity over global QSA.

Quantization followed by parameter-efficient fine-tuning has emerged as a promising paradigm for downstream adaptation under tight GPU memory constraints. However, this sequential pipeline fails to leverage the intricate interaction between quantization bit-width and LoRA rank. Specifically, a carefully optimized quantization allocation with low quantization error does not always translate to strong fine-tuning performance, and different bit-width and rank configurations can lead to significantly varying outcomes under the same memory budget. To address this limitation, we propose AutoQRA, a joint optimization framework that simultaneously optimizes the bit-width and LoRA rank configuration for each layer during the mixed quantized fine-tuning process. To tackle the challenges posed by the large discrete search space and the high evaluation cost associated with frequent fine-tuning iterations, AutoQRA decomposes the optimization process into two stages. First, it first conducts a global multi-fidelity evolutionary search, where the initial population is warm-started by injecting layer-wise importance priors. This stage employs specific operators and a performance model to efficiently screen candidate configurations. Second, trust-region Bayesian optimization is applied to locally refine promising regions of the search space and identify optimal configurations under the given memory budget. This approach enables active compensation for quantization noise in specific layers during training. Experiments show that AutoQRA achieves performance close to full-precision fine-tuning with a memory footprint comparable to uniform 4-bit methods.

The real-time supervision of production processes is a common challenge across several industries. It targets process component monitoring and its predictive maintenance in order to ensure safety, uninterrupted production and maintain high efficiency level. The rise of advanced tools for the simulation of physical systems in addition to data-driven machine learning models offers the possibility to design numerical tools dedicated to efficient system monitoring. In that respect, the digital twin concept presents an adequate framework that proffers solution to these challenges. The main purpose of this paper is to develop such a digital twin dedicated to fault detection and diagnosis in the context of a thermal-hydraulic process supervision. Based on a numerical simulation of the system, in addition to machine learning methods, we propose different modules dedicated to process parameter change detection and their on-line estimation. The proposed fault detection and diagnosis algorithm is validated on a specific test scenario, with single one-off parameter change occurrences in the system. The numerical results show good accuracy in terms of parameter variation localization and the update of their values.

State-space models (SSMs) have emerged as a powerful foundation for long-range sequence modeling, with the HiPPO framework showing that continuous-time projection operators can be used to derive stable, memory-efficient dynamical systems that encode the past history of the input signal. However, existing projection-based SSMs often rely on polynomial bases with global temporal support, whose inductive biases are poorly matched to signals exhibiting localized or transient structure. In this work, we introduce \emph{WaveSSM}, a collection of SSMs constructed over wavelet frames. Our key observation is that wavelet frames yield a localized support on the temporal dimension, useful for tasks requiring precise localization. Empirically, we show that on equal conditions, \textit{WaveSSM} outperforms orthogonal counterparts as S4 on real-world datasets with transient dynamics, including physiological signals on the PTB-XL dataset and raw audio on Speech Commands.

Modern vision generators transport a base distribution to data through time-indexed measures, implemented as deterministic flows (ODEs) or stochastic diffusions (SDEs). Despite strong empirical performance, standard flow-matching objectives do not directly control the information geometry of the trajectory, allowing low-entropy bottlenecks that can transiently deplete semantic modes. We propose Entropy-Controlled Flow Matching (ECFM): a constrained variational principle over continuity-equation paths enforcing a global entropy-rate budget d/dt H(mu_t) >= -lambda. ECFM is a convex optimization in Wasserstein space with a KKT/Pontryagin system, and admits a stochastic-control representation equivalent to a Schrodinger bridge with an explicit entropy multiplier. In the pure transport regime, ECFM recovers entropic OT geodesics and Gamma-converges to classical OT as lambda -> 0. We further obtain certificate-style mode-coverage and density-floor guarantees with Lipschitz stability, and construct near-optimal collapse counterexamples for unconstrained flow matching.