The Inference Report

Accurately upscaling terrestrial carbon fluxes is central to estimating the global carbon budget, yet remains challenging due to the sparse and regionally biased distribution of ground measurements. Existing data-driven upscaling products often fail to generalize beyond observed domains, leading to systematic regional biases and high predictive uncertainty. We introduce Task-Aware Modulation with Representation Learning (TAM-RL), a framework that couples spatio-temporal representation learning with knowledge-guided encoder-decoder architecture and loss function derived from the carbon balance equation. Across 150+ flux tower sites representing diverse biomes and climate regimes, TAM-RL improves predictive performance relative to existing state-of-the-art datasets, reducing RMSE by 8-9.6% and increasing explained variance ($R^2$) from 19.4% to 43.8%, depending on the target flux. These results demonstrate that integrating physically grounded constraints with adaptive representation learning can substantially enhance the robustness and transferability of global carbon flux estimates.

A central idea in mechanistic interpretability is that neural networks represent more features than they have dimensions, arranging them in superposition to form an over-complete basis. This framing has been influential, motivating dictionary learning approaches such as sparse autoencoders. However, superposition has mostly been studied in idealized settings where features are sparse and uncorrelated. In these settings, superposition is typically understood as introducing interference that must be minimized geometrically and filtered out by non-linearities such as ReLUs, yielding local structures like regular polytopes. We show that this account is incomplete for realistic data by introducing Bag-of-Words Superposition (BOWS), a controlled setting to encode binary bag-of-words representations of internet text in superposition. Using BOWS, we find that when features are correlated, interference can be constructive rather than just noise to be filtered out. This is achieved by arranging features according to their co-activation patterns, making interference between active features constructive, while still using ReLUs to avoid false positives. We show that this kind of arrangement is more prevalent in models trained with weight decay and naturally gives rise to semantic clusters and cyclical structures which have been observed in real language models yet were not explained by the standard picture of superposition. Code for this paper can be found at https://github.com/LucasPrietoAl/correlations-feature-geometry.

A key component of creativity is associative reasoning: the ability to draw novel yet meaningful connections between concepts. We introduce CREATE, a benchmark designed to evaluate models' capacity for creative associative reasoning. CREATE requires models to generate sets of paths connecting concepts in a model's parametric knowledge. Paths should have high specificity (distinctiveness and closeness of the concept connection) and high diversity (dissimilarity from other paths), and models are scored more highly if they produce a larger set of strong, diverse paths. This task shares demands of real creativity tasks like hypothesis generation, including an extremely large search space, but enables collection of a sizable benchmark with objective answer grading. Evaluation of frontier models shows that the strongest models achieve higher creative utility than others, with the high multiplicity of answers and complexity of the search making benchmark saturation difficult to achieve. Furthermore, our results illustrate that thinking models are not always more effective on our task, even with high token budgets. Recent approaches for creative prompting give some but limited additional improvement. CREATE provides a sandbox for developing new methods to improve models' capacity for associative creativity.

As social virtual reality (VR) grows more popular, addressing accessibility for blind and low vision (BLV) users is increasingly critical. Researchers have proposed an AI "sighted guide" to help users navigate VR and answer their questions, but it has not been studied with users. To address this gap, we developed a large language model (LLM)-powered guide and studied its use with 16 BLV participants in virtual environments with confederates posing as other users. We found that when alone, participants treated the guide as a tool, but treated it companionably around others, giving it nicknames, rationalizing its mistakes with its appearance, and encouraging confederate-guide interaction. Our work furthers understanding of guides as a versatile method for VR accessibility and presents design recommendations for future guides.

Collective decision-making in biological and human groups often emerges from simple interaction rules that amplify minor differences into consensus. The bee equation, developed initially to describe nest-site selection in honeybee swarms, captures this dynamic through recruitment and inhibition processes. Here, we extend the bee equation into an agent-based model in which emotional valence (positive-negative) and arousal (low-high) act as modulators of interaction rates, effectively altering the recruitment and cross-inhibition parameters. Agents display simulated facial expressions mapped from their valence-arousal states, allowing the study of emotional contagion in consensus formation. Three scenarios are explored: (1) the joint effect of valence and arousal on consensus outcomes and speed, (2) the role of arousal in breaking ties when valence is matched, and (3) the "snowball effect" in which consensus accelerates after surpassing intermediate support thresholds. Results show that emotional modulation can bias decision outcomes and alter convergence times by shifting effective recruitment and inhibition rates. At the same time, intrinsic non-linear amplification can produce decisive wins even in fully symmetric emotional conditions. These findings link classical swarm decision theory with affective and social modelling, highlighting how both emotional asymmetries and structural tipping points shape collective outcomes. The proposed framework offers a flexible tool for studying the emotional dimensions of collective choice in both natural and artificial systems.

Language-conditioned local navigation requires a robot to infer a nearby traversable target location from its current observation and an open-vocabulary, relational instruction. Existing vision-language spatial grounding methods usually rely on vision-language models (VLMs) to reason in image space, producing 2D predictions tied to visible pixels. As a result, they struggle to infer target locations in occluded regions, typically caused by furniture or moving humans. To address this issue, we propose BEACON, which predicts an ego-centric Bird's-Eye View (BEV) affordance heatmap over a bounded local region including occluded areas. Given an instruction and surround-view RGB-D observations from four directions around the robot, BEACON predicts the BEV heatmap by injecting spatial cues into a VLM and fusing the VLM's output with depth-derived BEV features. Using an occlusion-aware dataset built in the Habitat simulator, we conduct detailed experimental analysis to validate both our BEV space formulation and the design choices of each module. Our method improves the accuracy averaged across geodesic thresholds by 22.74 percentage points over the state-of-the-art image-space baseline on the validation subset with occluded target locations. Our project page is: https://xin-yu-gao.github.io/beacon.

While existing evaluations of large language models (LLMs) measure deception rates, the underlying conditions that give rise to deceptive behavior are poorly understood. We investigate this question using a novel dataset of realistic moral trade-offs where honesty incurs variable costs. Contrary to humans, who tend to become less honest given time to deliberate (Capraro, 2017; Capraro et al., 2019), we find that reasoning consistently increases honesty across scales and for several LLM families. This effect is not only a function of the reasoning content, as reasoning traces are often poor predictors of final behaviors. Rather, we show that the underlying geometry of the representational space itself contributes to the effect. Namely, we observe that deceptive regions within this space are metastable: deceptive answers are more easily destabilized by input paraphrasing, output resampling, and activation noise than honest ones. We interpret the effect of reasoning in this vein: generating deliberative tokens as part of moral reasoning entails the traversal of a biased representational space, ultimately nudging the model toward its more stable, honest defaults.

Self-supervised visual pre-training methods face an inherent tension: contrastive learning (CL) captures global semantics but loses fine-grained detail, while masked image modeling (MIM) preserves local textures but suffers from "attention drift" due to semantically-agnostic random masking. We propose C2FMAE, a coarse-to-fine masked autoencoder that resolves this tension by explicitly learning hierarchical visual representations across three data granularities: semantic masks (scene-level), instance masks (object-level), and RGB images (pixel-level). Two synergistic innovations enforce a strict top-down learning principle. First, a cascaded decoder sequentially reconstructs from scene semantics to object instances to pixel details, establishing explicit cross-granularity dependencies that parallel decoders cannot capture. Second, a progressive masking curriculum dynamically shifts the training focus from semantic-guided to instance-guided and finally to random masking, creating a structured learning path from global context to local features. To support this framework, we construct a large-scale multi-granular dataset with high-quality pseudo-labels for all 1.28M ImageNet-1K images. Extensive experiments show that C2FMAE achieves significant performance gains on image classification, object detection, and semantic segmentation, validating the effectiveness of our hierarchical design in learning more robust and generalizable representations.

A central question in modern deep learning is how to design optimizers whose behavior remains stable as the network width $w$ increases. We address this question by interpreting several widely used neural-network optimizers, including \textrm{AdamW} and \textrm{Muon}, as instances of steepest descent under matrix operator norms. This perspective links optimizer geometry with the Lipschitz structure of the network forward map, and enables width-independent control of both Lipschitz and smoothness constants. However, steepest-descent rules induced by standard $p \to q$ operator norms lack layerwise composability and therefore cannot provide width-independent bounds in deep architectures. We overcome this limitation by introducing a family of mean-normalized operator norms, denoted $\pmean \to \qmean$, that admit layerwise composability, yield width-independent smoothness bounds, and give rise to practical optimizers such as \emph{rescaled} \textrm{AdamW}, row normalization, and column normalization. The resulting learning rate width-aware scaling rules recover $μ$P scaling~\cite{yang2021tensor} as a special case and provide a principled mechanism for cross-width learning-rate transfer across a broad class of optimizers. We further show that \textrm{Muon} can suffer an $\mathcal{O}(\sqrt{w})$ worst-case growth in the smoothness constant, whereas a new family of row-normalized optimizers we propose achieves width-independent smoothness guarantees. Based on the observations, we propose MOGA (Matrix Operator Geometry Aware), a width-aware optimizer based only on row/column-wise normalization that enables stable learning-rate transfer across model widths. Large-scale pre-training on GPT-2 and LLaMA shows that MOGA, especially with row normalization, is competitive with Muon while being notably faster in large-token and low-loss regimes.

Training large language models (LLMs) on Python execution traces grounds them in code execution and enables the line-by-line execution prediction of whole Python programs, effectively turning them into neural interpreters (FAIR CodeGen Team et al., 2025). However, developers rarely execute programs step by step; instead, they use debuggers to stop execution at certain breakpoints and step through relevant portions only while inspecting or modifying program variables. Existing neural interpreter approaches lack such interactive control. To address this limitation, we introduce neural debuggers: language models that emulate traditional debuggers, supporting operations such as stepping into, over, or out of functions, as well as setting breakpoints at specific source lines. We show that neural debuggers -- obtained via fine-tuning large LLMs or pre-training smaller models from scratch -- can reliably model both forward execution (predicting future states and outputs) and inverse execution (inferring prior states or inputs) conditioned on debugger actions. Evaluated on CruxEval, our models achieve strong performance on both output and input prediction tasks, demonstrating robust conditional execution modeling. Our work takes first steps towards future agentic coding systems in which neural debuggers serve as a world model for simulated debugging environments, providing execution feedback or enabling agents to interact with real debugging tools. This capability lays the foundation for more powerful code generation, program understanding, and automated debugging.

Deep Reinforcement Learning systems are highly sensitive to the learning rate (LR), and selecting stable and performant training runs often requires extensive hyperparameter search. In Proximal Policy Optimization (PPO) actor--critic methods, small LR values lead to slow convergence, whereas large LR values may induce instability or collapse. We analyse this phenomenon from the behavior of the hidden neurons in the network using the Overfitting-Underfitting Indicator (OUI), a metric that quantifies the balance of binary activation patterns over a fixed probe batch. We introduce an efficient batch-based formulation of OUI and derive a theoretical connection between LR and activation sign changes, clarifying how a correct evolution of the neuron's inner structure depends on the step size. Empirically, across three discrete-control environments and multiple seeds, we show that OUI measured at only 10\% of training already discriminates between LR regimes. We observe a consistent asymmetry: critic networks achieving highest return operate in an intermediate OUI band (avoiding saturation), whereas actor networks achieving highest return exhibit comparatively high OUI values. We then compare OUI-based screening rules against early return, clip-based, divergence-based, and flip-based criteria under matched recall over successful runs. In this setting, OUI provides the strongest early screening signal: OUI alone achieves the best precision at broader recall, while combining early return with OUI yields the highest precision in best-performing screening regimes, enabling aggressive pruning of unpromising runs without requiring full training.

Ranked decision systems -- recommenders, ad auctions, clinical triage queues -- must decide when to intervene in ranked outputs and when to abstain. We study when confidence-based abstention monotonically improves decision quality, and when it fails. The formal conditions are simple: rank-alignment and no inversion zones. The substantive contribution is identifying why these conditions hold or fail: the distinction between structural uncertainty (missing data, e.g., cold-start) and contextual uncertainty (missing context, e.g., temporal drift). Empirically, we validate this distinction across three domains: collaborative filtering (MovieLens, 3 distribution shifts), e-commerce intent detection (RetailRocket, Criteo, Yoochoose), and clinical pathway triage (MIMIC-IV). Structural uncertainty produces near-monotonic abstention gains in all domains; structurally grounded confidence signals (observation counts) fail under contextual drift, producing as many monotonicity violations as random abstention on our MovieLens temporal split. Context-aware alternatives -- ensemble disagreement and recency features -- substantially narrow the gap (reducing violations from 3 to 1--2) but do not fully restore monotonicity, suggesting that contextual uncertainty poses qualitatively different challenges. Exception labels defined from residuals degrade substantially under distribution shift (AUC drops from 0.71 to 0.61--0.62 across three splits), providing a clean negative result against the common practice of exception-based intervention. The results provide a practical deployment diagnostic: check C1 and C2 on held-out data before deploying a confidence gate, and match the confidence signal to the dominant uncertainty type.

Conventional clinical CMR pipelines rely on a sequential "reconstruct-then-analyze" paradigm, forcing an ill-posed intermediate step that introduces avoidable artifacts and information bottlenecks. This creates a fundamental mathematical paradox: it attempts to recover high-dimensional pixel arrays (i.e., images) from undersampled k-space, rather than directly extracting the low-dimensional physiological labels actually required for diagnosis. To unlock the direct diagnostic potential of k-space, we propose k-MTR (k-space Multi-Task Representation), a k-space representation learning framework that aligns undersampled k-space data and fully-sampled images into a shared semantic manifold. Leveraging a large-scale controlled simulation of 42,000 subjects, k-MTR forces the k-space encoder to restore anatomical information lost to undersampling directly within the latent space, bypassing the explicit inverse problem for downstream analysis. We demonstrate that this latent alignment enables the dense latent space embedded with high-level physiological semantics directly from undersampled frequencies. Across continuous phenotype regression, disease classification, and anatomical segmentation, k-MTR achieves highly competitive performance against state-of-the-art image-domain baselines. By showcasing that precise spatial geometries and multi-task features can be successfully recovered directly from the k-space representations, k-MTR provides a robust architectural blueprint for task-aware cardiac MRI workflows.

Computational pathology demands both visual pattern recognition and dynamic integration of structured domain knowledge, including taxonomy, grading criteria, and clinical evidence. In practice, diagnostic reasoning requires linking morphological evidence with formal diagnostic and grading criteria. Although multimodal large language models (MLLMs) demonstrate strong vision language reasoning capabilities, they lack explicit mechanisms for structured knowledge integration and interpretable memory control. As a result, existing models struggle to consistently incorporate pathology-specific diagnostic standards during reasoning. Inspired by the hierarchical memory process of human pathologists, we propose PathMem, a memory-centric multimodal framework for pathology MLLMs. PathMem organizes structured pathology knowledge as a long-term memory (LTM) and introduces a Memory Transformer that models the dynamic transition from LTM to working memory (WM) through multimodal memory activation and context-aware knowledge grounding, enabling context-aware memory refinement for downstream reasoning. PathMem achieves SOTA performance across benchmarks, improving WSI-Bench report generation (12.8% WSI-Precision, 10.1% WSI-Relevance) and open-ended diagnosis by 9.7% and 8.9% over prior WSI-based models.

In the diverse landscape of 6G networks, where wireless connectivity demands surge and spectrum resources remain limited, flexible spectrum access becomes paramount. The success of crafting such schemes hinges on our ability to accurately characterize spectrum demand patterns across space and time. This paper presents a data-driven methodology for estimating spectrum demand variations over space and identifying key drivers of these variations in the mobile broadband landscape. By leveraging geospatial analytics and machine learning, the methodology is applied to a case study in Canada to estimate spectrum demand dynamics in urban regions. Our proposed model captures 70\% of the variability in spectrum demand when trained on one urban area and tested on another. These insights empower regulators to navigate the complexities of 6G networks and devise effective policies to meet future network demands.

Recent biosignal foundation models (FMs) have demonstrated promising performance across diverse clinical prediction tasks, yet systematic evaluation on long-duration multimodal data remains limited. We introduce SignalMC-MED, a benchmark for evaluating biosignal FMs on synchronized single-lead electrocardiogram (ECG) and photoplethysmogram (PPG) data. Derived from the MC-MED dataset, SignalMC-MED comprises 22,256 visits with 10-minute overlapping ECG and PPG signals, and includes 20 clinically relevant tasks spanning prediction of demographics, emergency department disposition, laboratory value regression, and detection of prior ICD-10 diagnoses. Using this benchmark, we perform a systematic evaluation of representative time-series and biosignal FMs across ECG-only, PPG-only, and ECG + PPG settings. We find that domain-specific biosignal FMs consistently outperform general time-series models, and that multimodal ECG + PPG fusion yields robust improvements over unimodal inputs. Moreover, using the full 10-minute signal consistently outperforms shorter segments, and larger model variants do not reliably outperform smaller ones. Hand-crafted ECG domain features provide a strong baseline and offer complementary value when combined with learned FM representations. Together, these results establish SignalMC-MED as a standardized benchmark and provide practical guidance for evaluating and deploying biosignal FMs.

Model merging has emerged as a transformative paradigm for combining the capabilities of multiple neural networks into a single unified model without additional training. With the rapid proliferation of fine-tuned large language models~(LLMs), merging techniques offer a computationally efficient alternative to ensembles and full retraining, enabling practitioners to compose specialized capabilities at minimal cost. This survey presents a comprehensive and structured examination of model merging in the LLM era through the \textbf{FUSE} taxonomy, a four-dimensional framework organized along \textbf{F}oundations, \textbf{U}nification Strategies, \textbf{S}cenarios, and \textbf{E}cosystem. We first establish the theoretical underpinnings of merging, including loss landscape geometry, mode connectivity, and the linear mode connectivity hypothesis. We then systematically review the algorithmic landscape, spanning weight averaging, task vector arithmetic, sparsification-enhanced methods, mixture-of-experts architectures, and evolutionary optimization approaches. For each method family, we analyze the core formulation, highlight representative works, and discuss practical trade-offs. We further examine downstream applications across multi-task learning, safety alignment, domain specialization, multilingual transfer, and federated learning. Finally, we survey the supporting ecosystem of open-source tools, community platforms, and evaluation benchmarks, and identify key open challenges including theoretical gaps, scalability barriers, and standardization needs. This survey aims to equip researchers and practitioners with a structured foundation for advancing model merging.

Generative Modeling via Drifting has recently achieved state-of-the-art one-step image generation through a kernel-based drift operator, yet the success is largely empirical and its theoretical foundations remain poorly understood. In this paper, we make the following observation: \emph{under a Gaussian kernel, the drift operator is exactly a score difference on smoothed distributions}. This insight allows us to answer all three key questions left open in the original work: (1) whether a vanishing drift guarantees equality of distributions ($V_{p,q}=0\Rightarrow p=q$), (2) how to choose between kernels, and (3) why the stop-gradient operator is indispensable for stable training. Our observations position drifting within the well-studied score-matching family and enable a rich theoretical perspective. By linearizing the McKean-Vlasov dynamics and probing them in Fourier space, we reveal frequency-dependent convergence timescales comparable to \emph{Landau damping} in plasma kinetic theory: the Gaussian kernel suffers an exponential high-frequency bottleneck, explaining the empirical preference for the Laplacian kernel. We also propose an exponential bandwidth annealing schedule $σ(t)=σ_0 e^{-rt}$ that reduces convergence time from $\exp(O(K_{\max}^2))$ to $O(\log K_{\max})$. Finally, by formalizing drifting as a Wasserstein gradient flow of the smoothed KL divergence, we prove that the stop-gradient operator is derived directly from the frozen-field discretization mandated by the JKO scheme, and removing it severs training from any gradient-flow guarantee. This variational perspective further provides a general template for constructing novel drift operators, demonstrated with a Sinkhorn divergence drift.

Multimodal neuroimaging provides complementary insights for Alzheimer's disease diagnosis, yet clinical datasets frequently suffer from missing modalities. We propose ACADiff, a framework that synthesizes missing brain imaging modalities through adaptive clinical-aware diffusion. ACADiff learns mappings between incomplete multimodal observations and target modalities by progressively denoising latent representations while attending to available imaging data and clinical metadata. The framework employs adaptive fusion that dynamically reconfigures based on input availability, coupled with semantic clinical guidance via GPT-4o-encoded prompts. Three specialized generators enable bidirectional synthesis among sMRI, FDG-PET, and AV45-PET. Evaluated on ADNI subjects, ACADiff achieves superior generation quality and maintains robust diagnostic performance even under extreme 80\% missing scenarios, outperforming all existing baselines. To promote reproducibility, code is available at https://github.com/rongzhou7/ACADiff

The Exponential Moving Average (EMA) is a cornerstone of widely used optimizers such as Adam. However, existing theoretical analyses of Adam-style methods have notable limitations: their guarantees can remain suboptimal in the zero-noise regime, rely on restrictive boundedness conditions (e.g., bounded gradients or objective gaps), use constant or open-loop stepsizes, or require prior knowledge of Lipschitz constants. To overcome these bottlenecks, we introduce OptEMA and analyze two novel variants: OptEMA-M, which applies an adaptive, decreasing EMA coefficient to the first-order moment with a fixed second-order decay, and OptEMA-V, which swaps these roles. Crucially, OptEMA is closed-loop and Lipschitz-free in the sense that its effective stepsizes are trajectory-dependent and do not require the Lipschitz constant for parameterization. Under standard stochastic gradient descent (SGD) assumptions, namely smoothness, a lower-bounded objective, and unbiased gradients with bounded variance, we establish rigorous convergence guarantees. Both variants achieve a noise-adaptive convergence rate of $\widetilde{\mathcal{O}}(T^{-1/2}+σ^{1/2} T^{-1/4})$ for the average gradient norm, where $σ$ is the noise level. In particular, in the zero-noise regime where $σ=0$, our bounds reduce to the nearly optimal deterministic rate $\widetilde{\mathcal{O}}(T^{-1/2})$ without manual hyperparameter retuning.

Accurately forecasting spectrum demand is a key component for efficient spectrum resource allocation and management. With the rapid growth in demand for wireless services, mobile network operators and regulators face increasing challenges in ensuring adequate spectrum availability. This paper presents a data-driven approach leveraging artificial intelligence (AI) and machine learning (ML) to estimate and manage spectrum demand. The approach uses multiple proxies of spectrum demand, drawing from site license data and derived from crowdsourced data. These proxies are validated against real-world mobile network traffic data to ensure reliability, achieving an R$^2$ value of 0.89 for an enhanced proxy. The proposed ML models are tested and validated across five major Canadian cities, demonstrating their generalizability and robustness. These contributions assist spectrum regulators in dynamic spectrum planning, enabling better resource allocation and policy adjustments to meet future network demands.

While Multi-Agent Systems (MAS) show potential for complex clinical decision support, the field remains hindered by architectural fragmentation and the lack of standardized multimodal integration. Current medical MAS research suffers from non-uniform data ingestion pipelines, inconsistent visual-reasoning evaluation, and a lack of cross-specialty benchmarking. To address these challenges, we present MedMASLab, a unified framework and benchmarking platform for multimodal medical multi-agent systems. MedMASLab introduces: (1) A standardized multimodal agent communication protocol that enables seamless integration of 11 heterogeneous MAS architectures across 24 medical modalities. (2) An automated clinical reasoning evaluator, a zero-shot semantic evaluation paradigm that overcomes the limitations of lexical string-matching by leveraging large vision-language models to verify diagnostic logic and visual grounding. (3) The most extensive benchmark to date, spanning 11 organ systems and 473 diseases, standardizing data from 11 clinical benchmarks. Our systematic evaluation reveals a critical domain-specific performance gap: while MAS improves reasoning depth, current architectures exhibit significant fragility when transitioning between specialized medical sub-domains. We provide a rigorous ablation of interaction mechanisms and cost-performance trade-offs, establishing a new technical baseline for future autonomous clinical systems. The source code and data is publicly available at: https://github.com/NUS-Project/MedMASLab/

While reasoning in LLMs plays a natural role in math, code generation, and multi-hop factual questions, its effect on simple, single-hop factual questions remains unclear. Such questions do not require step-by-step logical decomposition, making the utility of reasoning highly counterintuitive. Nevertheless, we find that enabling reasoning substantially expands the capability boundary of the model's parametric knowledge recall, unlocking correct answers that are otherwise effectively unreachable. Why does reasoning aid parametric knowledge recall when there are no complex reasoning steps to be done? To answer this, we design a series of hypothesis-driven controlled experiments, and identify two key driving mechanisms: (1) a computational buffer effect, where the model uses the generated reasoning tokens to perform latent computation independent of their semantic content; and (2) factual priming, where generating topically related facts acts as a semantic bridge that facilitates correct answer retrieval. Importantly, this latter generative self-retrieval mechanism carries inherent risks: we demonstrate that hallucinating intermediate facts during reasoning increases the likelihood of hallucinations in the final answer. Finally, we show that our insights can be harnessed to directly improve model accuracy by prioritizing reasoning trajectories that contain hallucination-free factual statements.

Continual fine-tuning of large language models (LLMs) is becoming increasingly crucial as these models are deployed in dynamic environments where tasks and data distributions evolve over time. While strong adaptability enables rapid acquisition of new knowledge, it also exposes LLMs to catastrophic forgetting, where previously learned skills degrade during sequential training. Existing replay-based strategies, such as fixed interleaved replay, accuracy-supervised, and loss-driven scheduling, remain limited: some depend on heuristic rules and provide only partial mitigation of forgetting, while others improve performance but incur substantial computational overhead. Motivated by retention dynamics under sequential fine-tuning, we propose Memory-Inspired Sampler and Scheduler Replay (MSSR), an experience replay framework that estimates sample-level memory strength and schedules rehearsal at adaptive intervals to mitigate catastrophic forgetting while maintaining fast adaptation. Extensive experiments across three backbone models and 11 sequential tasks show that MSSR consistently outperforms state-of-the-art replay baselines, with particularly strong gains on reasoning-intensive and multiple-choice benchmarks.

Large Language Models (LLMs) have emerged as a new paradigm for multi-agent systems. However, existing research on the behaviour of LLM-based multi-agents relies on ad hoc prompts and lacks a principled policy perspective. Different from reinforcement learning, we investigate whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training. Our framework regards prompts as actions executed by LLMs, and dynamically constructs prompts through five components based on the current state of the agent. To test the effectiveness of parameterized control, we evaluated the dialogue flow based on five indicators: responsiveness, rebuttal, evidence usage, non-repetition, and stance shift. We conduct experiments using different LLM-driven agents in two discussion scenarios related to the general public and show that prompt parameterization can influence the dialogue dynamics. This result shows that policy-parameterised prompts offer a simple and effective mechanism to influence the dialogue process, which will help the research of multi-agent systems in the direction of social simulation.

A fundamental requirement for intelligent systems is the ability to learn continuously under changing environments. However, models trained in this regime often suffer from catastrophic forgetting. Leveraging pre-trained models has recently emerged as a promising solution, since their generalized feature extractors enable faster and more robust adaptation. While some earlier works mitigate forgetting by fine-tuning only on the first task, this approach quickly deteriorates as the number of tasks grows and the data distributions diverge. More recent research instead seeks to consolidate task knowledge into a unified backbone, or adapting the backbone as new tasks arrive. However, such approaches may create a (potential) \textit{mismatch} between task-specific classifiers and the adapted backbone. To address this issue, we propose a novel \textit{Local Classifier Alignment} (LCA) loss to better align the classifier with backbone. Theoretically, we show that this LCA loss can enable the classifier to not only generalize well for all observed tasks, but also improve robustness. Furthermore, we develop a complete solution for continual learning, following the model merging approach and using LCA. Extensive experiments on several standard benchmarks demonstrate that our method often achieves leading performance, sometimes surpasses the state-of-the-art methods with a large margin.

Concerns persist regarding the capacity of Large Language Models (LLMs) to sway political views. Although prior research has claimed that LLMs are not more persuasive than standard political campaign practices, the recent rise of frontier models warrants further study. In two survey experiments (N=19,145) across bipartisan issues and stances, we evaluate seven state-of-the-art LLMs developed by Anthropic, OpenAI, Google, and xAI. We find that LLMs outperform standard campaign advertisements, with heterogeneity in performance across models. Specifically, Claude models exhibit the highest persuasiveness, while Grok exhibits the lowest. The results are robust across issues and stances. Moreover, in contrast to the findings in Hackenburg et al. (2025b) and Lin et al. (2025) that information-based prompts boost persuasiveness, we find that the effectiveness of information-based prompts is model-dependent: they increase the persuasiveness of Claude and Grok while substantially reducing that of GPT. We introduce a data-driven and strategy-agnostic LLM-assisted conversation analysis approach to identify and assess underlying persuasive strategies. Our work benchmarks the persuasive risks of frontier models and provides a framework for cross-model comparative risk assessment.

Extrinsic dexterity leverages environmental contact to overcome the limitations of prehensile manipulation. However, achieving such dexterity in cluttered scenes remains challenging and underexplored, as it requires selectively exploiting contact among multiple interacting objects with inherently coupled dynamics. Existing approaches lack explicit modeling of such complex dynamics and therefore fall short in non-prehensile manipulation in cluttered environments, which in turn limits their practical applicability in real-world environments. In this paper, we introduce a Dynamics-Aware Policy Learning (DAPL) framework that can facilitate policy learning with a learned representation of contact-induced object dynamics in cluttered environments. This representation is learned through explicit world modeling and used to condition reinforcement learning, enabling extrinsic dexterity to emerge without hand-crafted contact heuristics or complex reward shaping. We evaluate our approach in both simulation and the real world. Our method outperforms prehensile manipulation, human teleoperation, and prior representation-based policies by over 25% in success rate on unseen simulated cluttered scenes with varying densities. The real-world success rate reaches around 50% across 10 cluttered scenes, while a practical grocery deployment further demonstrates robust sim-to-real transfer and applicability.

Speech Large Language Models (SLLMs) have rapidly expanded, supporting a wide range of tasks. These models are typically evaluated using text prompts, which may not reflect real-world scenarios where users interact with speech. To address this gap, we introduce DoWhatISay (DOWIS), a multilingual dataset of human-recorded spoken and written prompts designed to pair with any existing benchmark for realistic evaluation of SLLMs under spoken instruction conditions. Spanning 9 tasks and 11 languages, it provides 10 prompt variants per task-language pair, across five styles. Using DOWIS, we benchmark state-of-the-art SLLMs, analyzing the interplay between prompt modality, style, language, and task type. Results show that text prompts consistently outperform spoken prompts, particularly for low-resource and cross-lingual settings. Only for tasks with speech output, spoken prompts do close the gap, highlighting the need for speech-based prompting in SLLM evaluation.

Recent work has found that contemporary language models such as transformers can become so good at next-word prediction that the probabilities they calculate become worse for predicting reading time. In this paper, we propose that this can be explained by reading time being sensitive to simple n-gram statistics rather than the more complex statistics learned by state-of-the-art transformer language models. We demonstrate that the neural language models whose predictions are most correlated with n-gram probability are also those that calculate probabilities that are the most correlated with eye-tracking-based metrics of reading time on naturalistic text.

Accurately quantifying terrestrial carbon exchange is essential for climate policy and carbon accounting, yet models must generalize to ecosystems underrepresented in sparse eddy covariance observations. Despite this challenge being a natural instance of zero-shot spatial transfer learning for time series regression, no standardized benchmark exists to rigorously evaluate model performance across geographically distinct locations with different climate regimes and vegetation types. We introduce CarbonBench, the first benchmark for zero-shot spatial transfer in carbon flux upscaling. CarbonBench comprises over 1.3 million daily observations from 567 flux tower sites globally (2000-2024). It provides: (1) stratified evaluation protocols that explicitly test generalization across unseen vegetation types and climate regimes, separating spatial transfer from temporal autocorrelation; (2) a harmonized set of remote sensing and meteorological features to enable flexible architecture design; and (3) baselines ranging from tree-based methods to domain-generalization architectures. By bridging machine learning methodologies and Earth system science, CarbonBench aims to enable systematic comparison of transfer learning methods, serves as a testbed for regression under distribution shift, and contributes to the next-generation climate modeling efforts.

Parameter-Efficient Fine-Tuning (PEFT) has become a key strategy for adapting large language models, with recent advances in sparse tuning reducing overhead by selectively updating key parameters or subsets of data. Existing approaches generally focus on two distinct paradigms: layer-selective methods aiming to fine-tune critical layers to minimize computational load, and data-selective methods aiming to select effective training subsets to boost training. However, current methods typically overlook the fact that different data points contribute varying degrees to distinct model layers, and they often discard potentially valuable information from data perceived as of low quality. To address these limitations, we propose Gradient-aligned Sparse Tuning (GAST), an innovative method that simultaneously performs selective fine-tuning at both data and layer dimensions as integral components of a unified optimization strategy. GAST specifically targets redundancy in information by employing a layer-sparse strategy that adaptively selects the most impactful data points for each layer, providing a more comprehensive and sophisticated solution than approaches restricted to a single dimension. Experiments demonstrate that GAST consistently outperforms baseline methods, establishing a promising direction for future research in PEFT strategies.

The surge in wireless connectivity demand, coupled with the finite nature of spectrum resources, compels the development of efficient spectrum management approaches. Spectrum sharing presents a promising avenue, although it demands precise characterization of spectrum demand for informed policy-making. This paper introduces HR-GAT, a hierarchical resolution graph attention network model, designed to predict spectrum demand using geospatial data. HR-GAT adeptly handles complex spatial demand patterns and resolves issues of spatial autocorrelation that usually challenge standard machine learning models, often resulting in poor generalization. Tested across five major Canadian cities, HR-GAT improves predictive accuracy of spectrum demand by 21% over eight baseline models, underscoring its superior performance and reliability.

Advances in large language models (LLMs) have enabled significant capabilities in audio processing, resulting in state-of-the-art models now known as Large Audio Language Models (LALMs). However, minimal work has been done to measure audio understanding beyond automatic speech recognition (ASR). This paper closes that gap by proposing a benchmark suite, SCENEBench (Spatial, Cross-lingual, Environmental, Non-speech Evaluation), that targets a broad form of audio comprehension across four real-world categories: background sound understanding, noise localization, cross-linguistic speech understanding, and vocal characterizer recognition. These four categories are selected based on understudied needs from accessibility technology and industrial noise monitoring. In addition to performance, we also measure model latency. The purpose of this benchmark suite is to assess audio beyond just what words are said - rather, how they are said and the non-speech components of the audio. Because our audio samples are synthetically constructed (e.g., by overlaying two natural audio samples), we further validate our benchmark against 20 natural audio items per task, sub-sampled from existing datasets to match our task criteria, to assess ecological validity. We assess five state-of-the-art LALMs and find critical gaps: performance varies across tasks, with some tasks performing below random chance and others achieving high accuracy. These results provide direction for targeted improvements in model capabilities.

Surrogate modeling is an essential data-driven technique for quantifying relationships between input variables and system responses in manufacturing and engineering systems. Two major challenges limit its effectiveness: (1) large data requirements for learning complex nonlinear relationships, and (2) heterogeneous data collected from sources with varying fidelity levels. Multi-task learning (MTL) addresses the first challenge by enabling information sharing across related processes, while multi-fidelity modeling addresses the second by accounting for fidelity-dependent uncertainty. However, existing approaches typically address these challenges separately, and no unified framework simultaneously leverages inter-task similarity and fidelity-dependent data characteristics. This paper develops a novel hierarchical multi-task multi-fidelity (H-MT-MF) framework for Gaussian process-based surrogate modeling. The proposed framework decomposes each task's response into a task-specific global trend and a residual local variability component that is jointly learned across tasks using a hierarchical Bayesian formulation. The framework accommodates an arbitrary number of tasks, design points, and fidelity levels while providing predictive uncertainty quantification. We demonstrate the effectiveness of the proposed method using a 1D synthetic example and a real-world engine surface shape prediction case study. Compared to (1) a state-of-the-art MTL model that does not account for fidelity information and (2) a stochastic kriging model that learns tasks independently, the proposed approach improves prediction accuracy by up to 19% and 23%, respectively. The H-MT-MF framework provides a general and extensible solution for surrogate modeling in manufacturing systems characterized by heterogeneous data sources.

Sequential multi-agent reasoning frameworks such as Chain-of-Agents (CoA) handle long-context queries by decomposing inputs into chunks and processing them sequentially using LLM-based worker agents that read from and update a bounded shared memory. From a probabilistic perspective, CoA aims to approximate the conditional distribution corresponding to a model capable of jointly reasoning over the entire long context. CoA achieves this through a latent-state factorization in which only bounded summaries of previously processed evidence are passed between agents. The resulting bounded-memory approximation introduces a lossy information bottleneck, making the final evidence state inherently dependent on the order in which chunks are processed. In this work, we study the problem of chunk ordering for long-context reasoning. We use the well-known Chow-Liu trees to learn a dependency structure that prioritizes strongly related chunks. Empirically, we show that a breadth-first traversal of the resulting tree yields chunk orderings that reduce information loss across agents and consistently outperform both default document-chunk ordering and semantic score-based ordering in answer relevance and exact-match accuracy across three long-context benchmarks.

As embodied models become powerful, humans will collaborate with multiple embodied AI agents at their workplace or home in the future. To ensure better communication between human users and the multi-agent system, it is crucial to interpret incoming information from agents in parallel and refer to the appropriate context for each query. Existing challenges include effectively compressing and communicating high volumes of individual sensory inputs in the form of video and correctly aggregating multiple egocentric videos to construct system-level memory. In this work, we first formally define a novel problem of understanding multiple long-horizon egocentric videos simultaneously collected from embodied agents. To facilitate research in this direction, we introduce MultiAgent-EgoQA (MA-EgoQA), a benchmark designed to systemically evaluate existing models in our scenario. MA-EgoQA provides 1.7k questions unique to multiple egocentric streams, spanning five categories: social interaction, task coordination, theory-of-mind, temporal reasoning, and environmental interaction. We further propose a simple baseline model for MA-EgoQA named EgoMAS, which leverages shared memory across embodied agents and agent-wise dynamic retrieval. Through comprehensive evaluation across diverse baselines and EgoMAS on MA-EgoQA, we find that current approaches are unable to effectively handle multiple egocentric streams, highlighting the need for future advances in system-level understanding across the agents. The code and benchmark are available at https://ma-egoqa.github.io.

Reliable evaluation is essential for developing and deploying large language models, yet in practice it often requires substantial manual effort: practitioners must identify appropriate benchmarks, reproduce heterogeneous evaluation codebases, configure dataset schema mappings, and interpret aggregated metrics. To address these challenges, we present One-Eval, an agentic evaluation system that converts natural-language evaluation requests into executable, traceable, and customizable evaluation workflows. One-Eval integrates (i) NL2Bench for intent structuring and personalized benchmark planning, (ii) BenchResolve for benchmark resolution, automatic dataset acquisition, and schema normalization to ensure executability, and (iii) Metrics \& Reporting for task-aware metric selection and decision-oriented reporting beyond scalar scores. The system further incorporates human-in-the-loop checkpoints for review, editing, and rollback, while preserving sample evidence trails for debugging and auditability. Experiments show that One-Eval can execute end-to-end evaluations from diverse natural-language requests with minimal user effort, supporting more efficient and reproducible evaluation in industrial settings. Our framework is publicly available at https://github.com/OpenDCAI/One-Eval.

The pseudo-projector is a lightweight modification that can be integrated into existing language models and other neural networks without altering their core architecture. It can be viewed as a hidden-representation corrector that reduces sensitivity to noise by suppressing directions induced by label-irrelevant input content. The design is inspired by the multigrid (MG) paradigm, originally developed to accelerate the convergence of iterative solvers for partial differential equations and boundary value problems, and later extended to more general linear systems through algebraic multigrid methods. We refer to the method as a pseudo-projector because its linear prototype corresponds to a strictly idempotent orthogonal projector, whereas the practical formulation employs learnable restriction and prolongation operators and therefore does not, in general, satisfy the properties of an exact orthogonal projection. We evaluate the proposed approach on transformer-based text classification tasks, as well as controlled synthetic benchmarks, demonstrating its effectiveness in improving training dynamics and robustness. Experimental results, together with supporting theoretical heuristics, indicate consistent improvements in training behavior across a range of settings, with no adverse effects observed otherwise. Our next step will be to extend this approach to language models.

Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that get correct answers by chance. We observe that better reasoning are better teachers: high-quality solutions serve as more effective demonstrations than low-quality ones. We term this teaching ability Demonstration Utility, and show that the policy model's own in-context learning ability provides an efficient way to measure it, yielding a quality signal termed Evidence Gain. To employ this signal during training, we introduce In-Context RLVR. By Bayesian analysis, we show that this objective implicitly reweights rewards by Evidence Gain, assigning higher weights to high-quality traces and lower weights to low-quality ones, without requiring costly computation or external evaluators. Experiments on mathematical benchmarks show improvements in both accuracy and reasoning quality over standard RLVR.

Large-scale scientific collaborations, such as the Compact Muon Solenoid (CMS) at CERN, produce a vast and ever-growing corpus of internal documentation. Navigating this complex information landscape presents a significant challenge for both new and experienced researchers, hindering knowledge sharing and slowing down the pace of scientific discovery. To address this, we present a prototype of MITRA, a Retrieval-Augmented Generation (RAG) based system, designed to answer specific, context-aware questions about physics analyses. MITRA employs a novel, automated pipeline using Selenium for document retrieval from internal databases and Optical Character Recognition (OCR) with layout parsing for high-fidelity text extraction. Crucially, MITRA's entire framework, from the embedding model to the Large Language Model (LLM), is hosted on-premise, ensuring that sensitive collaboration data remains private. We introduce a two-tiered vector database architecture that first identifies the relevant analysis from abstracts before focusing on the full documentation, resolving potential ambiguities between different analyses. We demonstrate the prototype's superior retrieval performance against a standard keyword-based baseline on realistic queries and discuss future work towards developing a comprehensive research agent for large experimental collaborations.

Bayesian optimization is a data-efficient technique that has been shown to be extremely powerful to optimize expensive, black-box, and possibly noisy objective functions. Many applications involve optimizing probabilities and mixtures which naturally belong to the probability simplex, a constrained non-Euclidean domain defined by non-negative entries summing to one. This paper introduces $α$-GaBO, a novel family of Bayesian optimization algorithms over the probability simplex. Our approach is grounded in information geometry, a branch of Riemannian geometry which endows the simplex with a Riemannian metric and a class of connections. Based on information geometry theory, we construct Matérn kernels that reflect the geometry of the probability simplex, as well as a one-parameter family of geometric optimizers for the acquisition function. We validate our method on benchmark functions and on a variety of real-world applications including mixtures of components, mixtures of classifiers, and a robotic control task, showing its increased performance compared to constrained Euclidean approaches.

Split learning (SL) transfers most of the training workload to the server, which alleviates computational burden on client devices. However, the transmission of intermediate feature representations, referred to as smashed data, incurs significant communication overhead, particularly when a large number of client devices are involved. To address this challenge, we propose an adaptive channel pruning-aided SL (ACP-SL) scheme. In ACP-SL, a label-aware channel importance scoring (LCIS) module is designed to generate channel importance scores, distinguishing important channels from less important ones. Based on these scores, an adaptive channel pruning (ACP) module is developed to prune less important channels, thereby compressing the corresponding smashed data and reducing the communication overhead. Experimental results show that ACP-SL consistently outperforms benchmark schemes in test accuracy. Furthermore, it reaches a target test accuracy in fewer training rounds, thereby reducing communication overhead.

Accurate forecasting of financial market volatility is crucial for risk management, option pricing, and portfolio optimization. Traditional econometric models and classical machine learning methods face challenges in handling the inherent non-linear and non-stationary characteristics of financial time series. In recent years, the rapid development of quantum computing has provided a new paradigm for solving complex optimization and sampling problems. This paper proposes a novel hybrid quantum-classical computing framework aimed at combining the powerful representation capabilities of classical neural networks with the unique advantages of quantum models. For the specific task of financial market volatility forecasting, we designed and implemented a hybrid model based on this framework, which combines a Long Short-Term Memory (LSTM) network with a Quantum Circuit Born Machine (QCBM). The LSTM is responsible for extracting complex dynamic features from historical time series data, while the QCBM serves as a learnable prior module, providing the model with a high-quality prior distribution to guide the forecasting process. We evaluated the model on two real financial datasets consisting of 5-minute high-frequency data from the Shanghai Stock Exchange (SSE) Composite Index and CSI 300 Index. Experimental results show that, compared to a purely classical LSTM baseline model, our hybrid quantum-classical model demonstrates significant advantages across multiple key metrics, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and QLIKE loss, proving the great potential of quantum computing in enhancing the capabilities of financial forecasting models. More broadly, the proposed hybrid framework offers a flexible architecture that may be adapted to other machine learning tasks involving high-dimensional, complex, or non-linear data distributions.

Explainable artificial intelligence (XAI) aims to provide human-interpretable insights into the behavior of deep neural networks (DNNs), typically by estimating a simplified causal structure of the model. In existing work, this causal structure often includes relationships where the presence of a concept is associated with a strong activation of a neuron. For example, attribution methods primarily identify input pixels that contribute most to a prediction, and feature visualization methods reveal inputs that cause high activation of a target neuron - the former implicitly assuming that the relevant information resides in the input, and the latter that neurons encode the presence of concepts. However, a largely overlooked type of causal relationship is that of encoded absences, where the absence of a concept increases neural activation. In this work, we show that such missing but relevant concepts are common and that mainstream XAI methods struggle to reveal them when applied in their standard form. To address this, we propose two simple extensions to attribution and feature visualization techniques that uncover encoded absences. Across experiments, we show how mainstream XAI methods can be used to reveal and explain encoded absences, how ImageNet models exploit them, and that debiasing can be improved when considering them.

Large language models (LLMs) tend to externalize their reasoning in their chain of thought, making the chain of thought a good target for monitoring. This is partially an inherent feature of the Transformer architecture: sufficiently long serial cognition must pass through the chain of thought (Korbak et al., 2025). We formalize this argument through the notion of opaque serial depth, given by the length of the longest computation that can be done without the use of interpretable intermediate steps like chain of thought. Given this formalization, we compute numeric upper bounds on the opaque serial depth of Gemma 3 models, as well as asymptotic results for additional architectures beyond standard LLMs. We also open-source an automated method that can calculate upper bounds on the opaque serial depth of arbitrary neural networks, and use it to demonstrate that Mixture-of-Experts models likely have lower depth than dense models. Overall, our results suggest that opaque serial depth is a useful tool for understanding the potential for models to do significant reasoning that is not externalized.

This paper introduces an updated and combined version of the bidirectional English-German EPIC-UdS (spoken) and EuroParl-UdS (written) corpora containing original European Parliament speeches as well as their translations and interpretations. The new version corrects metadata and text errors identified through previous use, refines the content, updates linguistic annotations, and adds new layers, including word alignment and word-level surprisal indices. The combined resource is designed to support research using information-theoretic approaches to language variation, particularly studies comparing written and spoken modes, and examining disfluencies in speech, as well as traditional translationese studies, including parallel (source vs. target) and comparable (original vs. translated) analyses. The paper outlines the updates introduced in this release, summarises previous results based on the corpus, and presents a new illustrative study. The study validates the integrity of the rebuilt spoken data and evaluates probabilistic measures derived from base and fine-tuned GPT-2 and machine translation models on the task of filler particles prediction in interpreting.

To enable an accurate determination of oscillation parameters, accelerator-based neutrino experiments require detailed simulations of nuclear interaction physics in the GeV regime. While substantial effort from both theory and experiment is currently being invested to improve the fidelity of these simulations, their present deficiencies typically oblige experimental collaborations to resort to empirical tuning of simulation model parameters. As the precision requirements of the field continue to become more stringent, machine learning techniques may provide a powerful means of handling corresponding growth in the complexity of future neutrino interaction model tuning exercises. To study the suitability of simulation-based inference (SBI) for this physics application, in this paper we revisit a tuned configuration of the GENIE neutrino event generator that was originally developed by the MicroBooNE collaboration. Despite closely reproducing the adopted values of four physics parameters when confronted with the tuned cross-section predictions as input, we find that our trained SBI algorithm prefers modestly different values (within MicroBooNE's assigned uncertainties) and achieves slightly better goodness-of-fit when inference is run on the experimental data set originally used by MicroBooNE. We also find that our trained algorithm can create a fair approximation of an alternative neutrino scattering simulation, NuWro, that shares only a subset of its physics model parameters with GENIE.

Achieving robust spatial reasoning remains a fundamental challenge for current Multimodal Foundation Models (MFMs). Existing methods either overfit statistical shortcuts via 3D grounding data or remain confined to 2D visual perception, limiting both spatial reasoning accuracy and generalization in unseen scenarios. Inspired by the spatial cognitive mapping mechanisms of biological intelligence, we propose World2Mind, a training-free spatial intelligence toolkit. At its core, World2Mind leverages 3D reconstruction and instance segmentation models to construct structured spatial cognitive maps, empowering MFMs to proactively acquire targeted spatial knowledge regarding interested landmarks and routes of interest. To provide robust geometric-topological priors, World2Mind synthesizes an Allocentric-Spatial Tree (AST) that uses elliptical parameters to model the top-down layout of landmarks accurately. To mitigate the inherent inaccuracies of 3D reconstruction, we introduce a three-stage reasoning chain comprising tool invocation assessment, modality-decoupled cue collection, and geometry-semantics interwoven reasoning. Extensive experiments demonstrate that World2Mind boosts the performance of frontier models, such as GPT-5.2, by 5%~18%. Astonishingly, relying solely on the AST-structured text, purely text-only foundation models can perform complex 3D spatial reasoning, achieving performance approaching that of advanced multimodal models.

We establish global universal approximation theorems on spaces of piecewise linear paths, stating that linear functionals of the corresponding signatures are dense with respect to $L^p$- and weighted norms, under an integrability condition on the underlying weight function. As an application, we show that piecewise linear interpolations of Brownian motion satisfies this integrability condition. Consequently, we obtain $L^p$-approximation results for path-dependent functionals, random ordinary differential equations, and stochastic differential equations driven by Brownian motion.

AI assistants that support humans in daily life are becoming increasingly feasible, driven by the rapid advancements in multimodal language models. A key challenge lies in overcoming the generic nature of these models to deliver personalized experiences. Existing approaches to personalizing large vision language models often rely on additional training stages, which limit generality and scalability, or on engineered pipelines with external pre-trained modules, which hinder deployment efficiency. In this work, we propose an efficient personalization method that leverages the model's inherent ability to capture personalized concepts. Specifically, we extract visual tokens that predominantly represent the target concept by utilizing the model's internal attention mechanisms. These tokens serve as a memory of that specific concept, enabling the model to recall and describe it when it appears in test images. We conduct a comprehensive and unified evaluation of our approach and SOTA methods across various personalization settings including single-concept, multi-concept, and video personalization, demonstrating strong performance gains with minimal personalization overhead.

Standardizing food terms from product labels and menus into ontology concepts is a prerequisite for trustworthy dietary assessment and safety reporting. The dominant approach to Named Entity Linking (NEL) in the food and nutrition domains fine-tunes Large Language Models (LLMs) on task-specific corpora. Although effective, fine-tuning incurs substantial computational cost, ties models to a particular ontology snapshot (i.e., version), and degrades under ontology drift. This paper presents FoodOntoRAG, a model- and ontology-agnostic pipeline that performs few-shot NEL by retrieving candidate entities from domain ontologies and conditioning an LLM on structured evidence (food labels, synonyms, definitions, and relations). A hybrid lexical--semantic retriever enumerates candidates; a selector agent chooses a best match with rationale; a separate scorer agent calibrates confidence; and, when confidence falls below a threshold, a synonym generator agent proposes reformulations to re-enter the loop. The pipeline approaches state-of-the-art accuracy while revealing gaps and inconsistencies in existing annotations. The design avoids fine-tuning, improves robustness to ontology evolution, and yields interpretable decisions through grounded justifications.

Neural oscillators that originate from the second-order ordinary differential equations (ODEs) have shown competitive performance in learning mappings between dynamic loads and responses of complex nonlinear structural systems. Despite this empirical success, theoretically quantifying the generalization capacities of their neural network architectures remains undeveloped. In this study, the neural oscillator consisting of a second-order ODE followed by a multilayer perceptron (MLP) is considered. Its upper probably approximately correct (PAC) generalization bound for approximating causal and uniformly continuous operators between continuous temporal function spaces and that for approximating the uniformly asymptotically incrementally stable second-order dynamical systems are derived by leveraging the Rademacher complexity framework. The theoretical results show that the estimation errors grow polynomially with respect to both the MLP size and the time length, thereby avoiding the curse of parametric complexity. Furthermore, the derived error bounds demonstrate that constraining the Lipschitz constants of the MLPs via loss function regularization can improve the generalization ability of the neural oscillator. A numerical study considering a Bouc-Wen nonlinear system under stochastic seismic excitation validates the theoretically predicted power laws of the estimation errors with respect to the sample size and time length, and confirms the effectiveness of constraining MLPs' matrix and vector norms in enhancing the performance of the neural oscillator under limited training data.

Multimodal large language models (MLLMs) are increasingly considered as a foundation for embodied agents, yet it remains unclear whether they can reliably reason about the long-term physical consequences of actions from an egocentric viewpoint. We study this gap through a new task, Egocentric Scene Prediction with LOng-horizon REasoning: given an initial-scene image and a sequence of atomic action descriptions, a model is asked to predict the final scene after all actions are executed. To enable systematic evaluation, we introduce EXPLORE-Bench, a benchmark curated from real first-person videos spanning diverse scenarios. Each instance pairs long action sequences with structured final-scene annotations, including object categories, visual attributes, and inter-object relations, which supports fine-grained, quantitative assessment. Experiments on a range of proprietary and open-source MLLMs reveal a significant performance gap to humans, indicating that long-horizon egocentric reasoning remains a major challenge. We further analyze test-time scaling via stepwise reasoning and show that decomposing long action sequences can improve performance to some extent, while incurring non-trivial computational overhead. Overall, EXPLORE-Bench provides a principled testbed for measuring and advancing long-horizon reasoning for egocentric embodied perception.

With the development of wireless network, Multi-Access Edge Computing (MEC) and Artificial Intelligence (AI)-native Radio Access Network (RAN) have attracted significant attention. Particularly, the integration of AI-RAN and MEC is envisioned to transform network efficiency and responsiveness. Therefore, it is valuable to investigate AI-RAN enabled MEC system. Federated learning (FL) nowadays is emerging as a promising approach for AI-RAN enabled MEC system, in which edge devices are enabled to train a global model cooperatively without revealing their raw data. However, conventional FL encounters the challenge in processing the non-independent and identically distributed (non-IID) data. Single prototype obtained by averaging the embedding vectors per class can be employed in FL to handle the data heterogeneity issue. Nevertheless, this may result in the loss of useful information owing to the average operation. Therefore, in this paper, a multi-prototype-guided federated knowledge distillation (MP-FedKD) approach is proposed. Particularly, self-knowledge distillation is integrated into FL to deal with the non-IID issue. To cope with the problem of information loss caused by single prototype-based strategy, multi-prototype strategy is adopted, where we present a conditional hierarchical agglomerative clustering (CHAC) approach and a prototype alignment scheme. Additionally, we design a novel loss function (called LEMGP loss) for each local client, where the relationship between global prototypes and local embedding will be focused. Extensive experiments over multiple datasets with various non-IID settings showcase that the proposed MP-FedKD approach outperforms the considered state-of-the-art baselines regarding accuracy, average accuracy and errors (RMSE and MAE).

Large language models (LLMs) are increasingly used across the scientific workflow, including to draft peer-review reports. However, many AI-generated reviews are superficial and insufficiently actionable, leaving authors without concrete, implementable guidance and motivating the gap this work addresses. We propose RbtAct, which targets actionable review feedback generation and places existing peer review rebuttal at the center of learning. Rebuttals show which reviewer comments led to concrete revisions or specific plans, and which were only defended. Building on this insight, we leverage rebuttal as implicit supervision to directly optimize a feedback generator for actionability. To support this objective, we propose a new task called perspective-conditioned segment-level review feedback generation, in which the model is required to produce a single focused comment based on the complete paper and a specified perspective such as experiments and writing. We also build a large dataset named RMR-75K that maps review segments to the rebuttal segments that address them, with perspective labels and impact categories that order author uptake. We then train the Llama-3.1-8B-Instruct model with supervised fine-tuning on review segments followed by preference optimization using rebuttal derived pairs. Experiments with human experts and LLM-as-a-judge show consistent gains in actionability and specificity over strong baselines while maintaining grounding and relevance.

Autonomous agent frameworks still struggle to reconcile long-term experiential learning with real-time, context-sensitive decision-making. In practice, this gap appears as static cognition, rigid workflow dependence, and inefficient context usage, which jointly limit adaptability in open-ended and non-stationary environments. To address these limitations, we present AutoAgent, a self-evolving multi-agent framework built on three tightly coupled components: evolving cognition, on-the-fly contextual decision-making, and elastic memory orchestration. At the core of AutoAgent, each agent maintains structured prompt-level cognition over tools, self-capabilities, peer expertise, and task knowledge. During execution, this cognition is combined with live task context to select actions from a unified space that includes tool calls, LLM-based generation, and inter-agent requests. To support efficient long-horizon reasoning, an Elastic Memory Orchestrator dynamically organizes interaction history by preserving raw records, compressing redundant trajectories, and constructing reusable episodic abstractions, thereby reducing token overhead while retaining decision-critical evidence. These components are integrated through a closed-loop cognitive evolution process that aligns intended actions with observed outcomes to continuously update cognition and expand reusable skills, without external retraining. Empirical results across retrieval-augmented reasoning, tool-augmented agent benchmarks, and embodied task environments show that AutoAgent consistently improves task success, tool-use efficiency, and collaborative robustness over static and memory-augmented baselines. Overall, AutoAgent provides a unified and practical foundation for adaptive autonomous agents that must learn from experience while making reliable context-aware decisions in dynamic environments.

Visual instruction tuning is crucial for improving vision-language large models (VLLMs). However, many samples can be solved via linguistic patterns or common-sense shortcuts, without genuine cross-modal reasoning, limiting the effectiveness of multimodal learning. Prior data selection methods often rely on costly proxy model training and focus on difficulty or diversity, failing to capture a sample's true contribution to vision-language joint reasoning. In this paper, we propose CVS, a training-free data selection method based on the insight that, for high-quality multimodal samples, introducing the question should substantially alter the model's assessment of answer validity given an image. CVS leverages a frozen VLLM as an evaluator and measures the discrepancy in answer validity with and without conditioning on the question, enabling the identification of samples that require vision-language joint reasoning while filtering semantic-conflict noise. Experiments on Vision-Flan and The Cauldron show that CVS achieves solid performance across datasets. On Vision-Flan, CVS outperforms full-data training by 3.5% and 4.8% using only 10% and 15% of the data, respectively, and remains robust on the highly heterogeneous Cauldron dataset. Moreover, CVS reduces computational cost by 17.3% and 44.4% compared to COINCIDE and XMAS.

While multi-audio understanding is critical for large audio-language models (LALMs), it remains underexplored. We introduce MUGEN, a comprehensive benchmark evaluating this capability across speech, general audio, and music. Our experiments reveal consistent weaknesses in multi-audio settings, and performance degrades sharply as the number of concurrent audio inputs increases, identifying input scaling as a fundamental bottleneck. We further investigate training-free strategies and observe that Audio-Permutational Self-Consistency, which diversifies the order of audio candidates, helps models form more robust aggregated predictions, yielding up to 6.28% accuracy gains. Combining this permutation strategy with Chain-of-Thought further improves performance to 6.74%. These results expose blind spots in current LALMs and provide a foundation for evaluating complex auditory comprehension.

While safety alignment for Multimodal Large Language Models (MLLMs) has gained significant attention, current paradigms primarily target malicious intent or situational violations. We propose shifting the safety frontier toward consequence-driven safety, a paradigm essential for the robust deployment of autonomous and embodied agents. To formalize this shift, we introduce OOD-MMSafe, a benchmark comprising 455 curated query-image pairs designed to evaluate a model's ability to identify latent hazards within context-dependent causal chains. Our analysis reveals a pervasive causal blindness among frontier models, with the highest 67.5% failure rate in high-capacity closed-source models, and identifies a preference ceiling where static alignment yields format-centric failures rather than improved safety reasoning as model capacity grows. To address these bottlenecks, we develop the Consequence-Aware Safety Policy Optimization (CASPO) framework, which integrates the model's intrinsic reasoning as a dynamic reference for token-level self-distillation rewards. Experimental results demonstrate that CASPO significantly enhances consequence projection, reducing the failure ratio of risk identification to 7.3% for Qwen2.5-VL-7B and 5.7% for Qwen3-VL-4B while maintaining overall effectiveness.

In this article, we evaluate four Large Language Models (LLMs) and their effectiveness at retrieving data within a specialized Retrieval-Augmented Generation (RAG) system, using a comprehensive food composition database. Our method is focused on the LLMs ability to translate natural language queries into structured metadata filters, enabling efficient retrieval via a Chroma vector database. By achieving high accuracy in this critical retrieval step, we demonstrate that LLMs can serve as an accessible, high-performance tool, drastically reducing the manual effort and technical expertise previously required for domain experts, such as food compilers and nutritionists, to leverage complex food and nutrition data. However, despite the high performance on easy and moderately complex queries, our analysis of difficult questions reveals that reliable retrieval remains challenging when queries involve non-expressible constraints. These findings demonstrate that LLM-driven metadata filtering excels when constraints can be explicitly expressed, but struggles when queries exceed the representational scope of the metadata format.

Large Language Models (LLMs) have revolutionized code generation, evolving from static tools into dynamic conversational interfaces that facilitate complex, multi-turn collaborative programming. While LLMs exhibit remarkable proficiency in generating standalone code snippets, they often struggle to maintain contextual consistency during extended interactions, creating significant obstacles in the collaboration process. Existing benchmarks primarily emphasize the functional correctness of the final output, overlooking latent quality issues within the interaction process itself, which we term Interaction Smells. In this paper, we conduct an empirical study on sampled real-word user-LLM interactions from WildChat and LMSYS-Chat-1M datasets to systematically investigate Interaction Smells in human-LLM code generation tasks from the perspectives of phenomena, distribution, and mitigation. First, we establish the first taxonomy of Interaction Smells by manually performing open card sorting on real-world interaction logs. This taxonomy categorizes Interaction Smells into three primary categories, i.e., User Intent Quality, Historical Instruction Compliance, and Historical Response Violation, comprising nine specific subcategories. Next, we quantitatively evaluate six mainstream LLMs (i.e., GPT-4o, DeepSeek-Chat, Gemini 2.5, Qwen2.5-32B, Qwen2.5-72B, and Qwen3-235B-a22b) to analyze the distribution of Interaction Smells across different models. Finally, we propose Invariant-aware Constraint Evolution (InCE), a multi-agent framework designed to improve multi-turn interaction quality through explicit extraction of global invariants and pre-generation quality audits. Experimental results on the extended WildBench benchmark demonstrate that this lightweight mitigation approach significantly improves the Task Success Rate and effectively suppresses the occurrence of Interaction Smells.

Recent advances in spectral optimization, notably Muon, have demonstrated that constraining update steps to the Stiefel manifold can significantly accelerate training and improve generalization. However, Muon implicitly assumes an isotropic optimization landscape, enforcing a uniform spectral update norm across all eigen-directions. We argue that this "egalitarian" constraint is suboptimal for Deep Neural Networks, where the curvature spectrum is known to be highly heavy-tailed and ill-conditioned. In such landscapes, Muon risks amplifying instabilities in high-curvature directions while limiting necessary progress in flat directions. In this work, we propose \textbf{Mousse} (\textbf{M}uon \textbf{O}ptimization \textbf{U}tilizing \textbf{S}hampoo's \textbf{S}tructural \textbf{E}stimation), a novel optimizer that reconciles the structural stability of spectral methods with the geometric adaptivity of second-order preconditioning. Instead of applying Newton-Schulz orthogonalization directly to the momentum matrix, Mousse operates in a whitened coordinate system induced by Kronecker-factored statistics (derived from Shampoo). Mathematically, we formulate Mousse as the solution to a spectral steepest descent problem constrained by an anisotropic trust region, where the optimal update is derived via the polar decomposition of the whitened gradient. Empirical results across language models ranging from 160M to 800M parameters demonstrate that Mousse consistently outperforms Muon, achieving around $\sim$12\% reduction in training steps with negligible computational overhead.

Predicting the microstructural and morphological evolution of materials through phase-field modelling is computationally intensive, particularly for high-throughput parametric studies. While neural operators such as the Fourier neural operator (FNO) show promise in accelerating the solution of parametric partial differential equations (PDEs), the lack of explicit physical constraints, may limit generalisation and long-term accuracy for complex phase-field dynamics. Here, we develop a physics-informed neural operator framework to learn parametric phase-field PDEs, namely PF-PINO. By embedding the residuals of phase-field governing equations into the data-fidelity loss function, our framework effectively enforces physical constraints during training. We validate PF-PINO against benchmark phase-field problems, including electrochemical corrosion, dendritic crystal solidification, and spinodal decomposition. Our results demonstrate that PF-PINO significantly outperforms conventional FNO in accuracy, generalisation capability, and long-term stability. This work provides a robust and efficient computational tool for phase-field modelling and highlights the potential of physics-informed neural operators to advance scientific machine learning for complex interfacial evolution problems.

Reinforcement Learning from Human Feedback (RLHF) has become the standard for aligning Large Language Models (LLMs), yet its efficacy is bottlenecked by the high cost of acquiring preference data, especially in low-resource and expert domains. To address this, we introduce ACTIVEULTRAFEEDBACK, a modular active learning pipeline that leverages uncertainty estimates to dynamically identify the most informative responses for annotation. Our pipeline facilitates the systematic evaluation of standard response selection methods alongside DOUBLE REVERSE THOMPSON SAMPLING (DRTS) and DELTAUCB, two novel methods prioritizing response pairs with large predicted quality gaps, leveraging recent results showing that such pairs provide good signals for fine-tuning. Our experiments demonstrate that ACTIVEULTRAFEEDBACK yields high-quality datasets that lead to significant improvements in downstream performance, notably achieving comparable or superior results with as little as one-sixth of the annotated data relative to static baselines. Our pipeline is available at https://github.com/lasgroup/ActiveUltraFeedback and our preference datasets at https://huggingface.co/ActiveUltraFeedback.

Existing end-to-end modeling methods for modular task-oriented dialog systems are typically tailored to specific datasets, making it challenging to adapt to new dialog scenarios. In this work, we propose ESAinsTOD, a unified End-to-end Schema-Aware Instruction-tuning framework for general Task-Oriented Dialog modeling. This framework introduces a structured methodology to go beyond simply fine-tuning Large Language Models (LLMs), enabling flexible adaptation to various dialogue task flows and schemas. Specifically, we leverage full-parameter fine-tuning of LLMs and introduce two alignment mechanisms to make the resulting system both instruction-aware and schema-aware: (i) instruction alignment, which ensures that the system faithfully follows task instructions to complete various task flows from heterogeneous TOD datasets; and (ii) schema alignment, which encourages the system to make predictions adhering to the specified schema. In addition, we employ session-level end-to-end modeling, which allows the system to access the results of previously executed task flows within the dialogue history, to bridge the gap between the instruction-tuning paradigm and the real-world application of TOD systems. Empirical results show that while a fine-tuned LLM serves as a strong baseline, our structured approach provides significant additional benefits. In particular, our findings indicate that: (i) ESAinsTOD outperforms state-of-the-art models by a significant margin on end-to-end task-oriented dialog modeling benchmarks: CamRest676, In-Car and MultiWOZ; (ii) more importantly, it exhibits superior generalization capabilities across various low-resource settings, with the proposed alignment mechanisms significantly enhancing zero-shot performance; and (iii) our instruction-tuning paradigm substantially improves the model's robustness against data noise and cascading errors.

Visual Question Answering (VQA) is a fundamental multimodal task that requires models to jointly understand visual and textual information. Early VQA systems relied heavily on language biases, motivating subsequent work to emphasize visual grounding and balanced datasets. With the success of large-scale pre-trained transformers for both text and vision domains -- such as PhoBERT for Vietnamese language understanding and Vision Transformers (ViT) for image representation learning -- multimodal fusion has achieved remarkable progress. For Vietnamese VQA, several datasets have been introduced to promote research in low-resource multimodal learning, including ViVQA, OpenViVQA, and the recently proposed ViTextVQA. These resources enable benchmarking of models that integrate linguistic and visual features in the Vietnamese context. Evaluation of VQA systems often employs automatic metrics originally designed for image captioning or machine translation, such as BLEU, METEOR, CIDEr, Recall, Precision, and F1-score. However, recent research suggests that large language models can further improve the alignment between automatic evaluation and human judgment in VQA tasks. In this work, we explore Vietnamese Visual Question Answering using transformer-based architectures, leveraging both textual and visual pre-training while systematically comparing automatic evaluation metrics under multilingual settings.

This research focuses on developing advanced methods for assessing similarity between recipes by combining different sources of information and analytical approaches. We explore the semantic, lexical, and domain similarity of food recipes, evaluated through the analysis of ingredients, preparation methods, and nutritional attributes. A web-based interface was developed to allow domain experts to validate the combined similarity results. After evaluating 318 recipe pairs, experts agreed on 255 (80%). The evaluation of expert assessments enables the estimation of which similarity aspects--lexical, semantic, or nutritional--are most influential in expert decision-making. The application of these methods has broad implications in the food industry and supports the development of personalized diets, nutrition recommendations, and automated recipe generation systems.

To overcome the limitations of manual administrative coding in geriatric Cardiovascular Risk Management, this study introduces an automated classification framework leveraging unstructured Electronic Health Records (EHRs). Using a dataset of 3,482 patients, we benchmarked three distinct modeling paradigms on longitudinal Dutch clinical narratives: classical machine learning baselines, specialized deep learning architectures optimized for large-context sequences, and general-purpose generative Large Language Models (LLMs) in a zero-shot setting. Additionally, we evaluated a late fusion strategy to integrate unstructured text with structured medication embeddings and anthropometric data. Our analysis reveals that the custom Transformer architecture outperforms both traditional methods and generative \acs{llm}s, achieving the highest F1-scores and Matthews Correlation Coefficients. These findings underscore the critical role of specialized hierarchical attention mechanisms in capturing long-range dependencies within medical texts, presenting a robust, automated alternative to manual workflows for clinical risk stratification.

Parameter-efficient fine-tuning (PEFT) based on low-rank decomposition, such as LoRA, has become a standard for adapting large pretrained models. However, its behavior in sequential learning -- specifically regarding catastrophic forgetting -- remains insufficiently understood. In this work, we present an empirical study showing that forgetting is strongly influenced by the geometry and parameterization of the update subspace. While methods that restrict updates to small, shared matrix subspaces often suffer from task interference, tensor-based decompositions (e.g., LoRETTA) mitigate forgetting by capturing richer structural information within ultra-compact budgets, and structurally aligned parameterizations (e.g., WeGeFT) preserve pretrained representations. Our findings highlight update subspace design as a key factor in continual learning and offer practical guidance for selecting efficient adaptation strategies in sequential settings.

Large language models achieve near-ceiling performance on code generation benchmarks, yet these results increasingly reflect memorization rather than genuine reasoning. We introduce EsoLang-Bench, a benchmark using five esoteric programming languages (Brainfuck, Befunge-98, Whitespace, Unlambda, and Shakespeare) that lack benchmark gaming incentives due to their economic irrationality for pre-training. These languages require the same computational primitives as mainstream programming but have 1,000-100,000x fewer public repositories than Python (based on GitHub search counts). We evaluate five frontier models across five prompting strategies and find a dramatic capability gap: models achieving 85-95% on standard benchmarks score only 0-11% on equivalent esoteric tasks, with 0% accuracy beyond the Easy tier. Few-shot learning and self-reflection fail to improve performance, suggesting these techniques exploit training priors rather than enabling genuine learning. EsoLang-Bench provides the first benchmark designed to mimic human learning by acquiring new languages through documentation, interpreter feedback, and iterative experimentation, measuring transferable reasoning skills resistant to data contamination.

Addressing the challenges of fragmented task definitions and the heterogeneity of unstructured data in multimodal parsing, this paper proposes the Omni Parsing framework. This framework establishes a Unified Taxonomy covering documents, images, and audio-visual streams, introducing a progressive parsing paradigm that bridges perception and cognition. Specifically, the framework integrates three hierarchical levels: 1) Holistic Detection, which achieves precise spatial-temporal grounding of objects or events to establish a geometric baseline for perception; 2) Fine-grained Recognition, which performs symbolization (e.g., OCR/ASR) and attribute extraction on localized objects to complete structured entity parsing; and 3) Multi-level Interpreting, which constructs a reasoning chain from local semantics to global logic. A pivotal advantage of this framework is its evidence anchoring mechanism, which enforces a strict alignment between high-level semantic descriptions and low-level facts. This enables ``evidence-based'' logical induction, transforming unstructured signals into standardized knowledge that is locatable, enumerable, and traceable. Building on this foundation, we constructed a standardized dataset and released the Logics-Parsing-Omni model, which successfully converts complex audio-visual signals into machine-readable structured knowledge. Experiments demonstrate that fine-grained perception and high-level cognition are synergistic, effectively enhancing model reliability. Furthermore, to quantitatively evaluate these capabilities, we introduce OmniParsingBench. Code, models and the benchmark are released at https://github.com/alibaba/Logics-Parsing/tree/master/Logics-Parsing-Omni.

There is growing interest in applying graph-based methods to Time Series Anomaly Detection (TSAD), particularly Graph Neural Networks (GNNs), as they naturally model dependencies among multivariate signals. GNNs are typically used as backbones in score-based TSAD pipelines, where anomalies are identified through reconstruction or prediction errors followed by thresholding. However, and despite promising results, the field still lacks standardized frameworks for evaluation and suffers from persistent issues with metric design and interpretation. We thus present an open-source framework for TSAD using GNNs, designed to support reproducible experimentation across datasets, graph structures, and evaluation strategies. Built with flexibility and extensibility in mind, the framework facilitates systematic comparisons between TSAD models and enables in-depth analysis of performance and interpretability. Using this tool, we evaluate several GNN-based architectures alongside baseline models across two real-world datasets with contrasting structural characteristics. Our results show that GNNs not only improve detection performance but also offer significant gains in interpretability, an especially valuable feature for practical diagnosis. We also find that attention-based GNNs offer robustness when graph structure is uncertain or inferred. In addition, we reflect on common evaluation practices in TSAD, showing how certain metrics and thresholding strategies can obscure meaningful comparisons. Overall, this work contributes both practical tools and critical insights to advance the development and evaluation of graph-based TSAD systems.

Bias can be introduced in diverse ways in machine learning datasets, for example via selection or label bias. Although these bias types in themselves have an influence on important aspects of fair machine learning, their different impact has been understudied. In this work, we empirically analyze the effect of label bias and several subtypes of selection bias on the evaluation of classification models, on their performance, and on the effectiveness of bias mitigation methods. We also introduce a biasing and evaluation framework that allows to model fair worlds and their biased counterparts through the introduction of controlled bias in real-life datasets with low discrimination. Using our framework, we empirically analyze the impact of each bias type independently, while obtaining a more representative evaluation of models and mitigation methods than with the traditional use of a subset of biased data as test set. Our results highlight different factors that influence how impactful bias is on model performance. They also show an absence of trade-off between fairness and accuracy, and between individual and group fairness, when models are evaluated on a test set that does not exhibit unwanted bias. They furthermore indicate that the performance of bias mitigation methods is influenced by the type of bias present in the data. Our findings call for future work to develop more accurate evaluations of prediction models and fairness interventions, but also to better understand other types of bias, more complex scenarios involving the combination of different bias types, and other factors that impact the efficiency of the mitigation methods, such as dataset characteristics.

Mining time-frequency features is critical for time series forecasting. Existing research has predominantly focused on modeling low-frequency patterns, where most time series energy is concentrated. The overlooking of mid to high frequency continues to limit further performance gains in deep learning models. We propose FreqCycle, a novel framework integrating: (i) a Filter-Enhanced Cycle Forecasting (FECF) module to extract low-frequency features by explicitly learning shared periodic patterns in the time domain, and (ii) a Segmented Frequency-domain Pattern Learning (SFPL) module to enhance mid to high frequency energy proportion via learnable filters and adaptive weighting. Furthermore, time series data often exhibit coupled multi-periodicity, such as intertwined weekly and daily cycles. To address coupled multi-periodicity as well as long lookback window challenges, we extend FreqCycle hierarchically into MFreqCycle, which decouples nested periodic features through cross-scale interactions. Extensive experiments on seven diverse domain benchmarks demonstrate that FreqCycle achieves state-of-the-art accuracy while maintaining faster inference speeds, striking an optimal balance between performance and efficiency.

Maintaining background consistency while enhancing foreground quality remains a core challenge in video editing. Injecting full-image information often leads to background artifacts, whereas rigid background locking severely constrains the model's capacity for foreground generation. To address this issue, we propose KV-Lock, a training-free framework tailored for DiT-based video diffusion models. Our core insight is that the hallucination metric (variance of denoising prediction) directly quantifies generation diversity, which is inherently linked to the classifier-free guidance (CFG) scale. Building upon this, KV-Lock leverages diffusion hallucination detection to dynamically schedule two key components: the fusion ratio between cached background key-values (KVs) and newly generated KVs, and the CFG scale. When hallucination risk is detected, KV-Lock strengthens background KV locking and simultaneously amplifies conditional guidance for foreground generation, thereby mitigating artifacts and improving generation fidelity. As a training-free, plug-and-play module, KV-Lock can be easily integrated into any pre-trained DiT-based models. Extensive experiments validate that our method outperforms existing approaches in improved foreground quality with high background fidelity across various video editing tasks.

Language Models (LMs) acquire parametric knowledge from their training process, embedding it within their weights. The increasing scalability of LMs, however, poses significant challenges for understanding a model's inner workings and further for updating or correcting this embedded knowledge without the significant cost of retraining. Moreover, when using these language models for knowledge-intensive language understanding tasks, LMs have to integrate relevant context, mitigating their inherent weaknesses, such as incomplete or outdated knowledge. Nevertheless, studies indicate that LMs often ignore the provided context as it can be in conflict with the pre-existing LM's memory learned during pre-training. Conflicting knowledge can also already be present in the LM's parameters, termed intra-memory conflict. This underscores the importance of understanding the interplay between how a language model uses its parametric knowledge and the retrieved contextual knowledge. In this talk, I will aim to shed light on this important issue by presenting our research on evaluating the knowledge present in LMs, diagnostic tests that can reveal knowledge conflicts, as well as on understanding the characteristics of successfully used contextual knowledge.

With the rapid advancement of Large Language Models (LLMs) in code generation, human-AI interaction is evolving from static text responses to dynamic, interactive HTML-based applications, which we term MiniApps. These applications require models to not only render visual interfaces but also construct customized interaction logic that adheres to real-world principles. However, existing benchmarks primarily focus on algorithmic correctness or static layout reconstruction, failing to capture the capabilities required for this new paradigm. To address this gap, we introduce MiniAppBench, the first comprehensive benchmark designed to evaluate principle-driven, interactive application generation. Sourced from a real-world application with 10M+ generations, MiniAppBench distills 500 tasks across six domains (e.g., Games, Science, and Tools). Furthermore, to tackle the challenge of evaluating open-ended interactions where no single ground truth exists, we propose MiniAppEval, an agentic evaluation framework. Leveraging browser automation, it performs human-like exploratory testing to systematically assess applications across three dimensions: Intention, Static, and Dynamic. Our experiments reveal that current LLMs still face significant challenges in generating high-quality MiniApps, while MiniAppEval demonstrates high alignment with human judgment, establishing a reliable standard for future research. Our code is available in github.com/MiniAppBench.

Pore-scale imaging of subsurface formations is costly and limited to discrete depths, creating significant gaps in reservoir characterization. To address this, we present a conditional Generative Adversarial Network (cGAN) framework for synthesizing realistic thin section images of carbonate rock formations, conditioned on porosity values derived from well logs. The model is trained on 5,000 sub-images extracted from 15 petrography samples over a depth interval of 1992-2000m, the model generates geologically consistent images across a wide porosity range (0.004-0.745), achieving 81% accuracy within a 10\% margin of target porosity values. The successful integration of well log data with the trained generator enables continuous pore-scale visualization along the wellbore, bridging gaps between discrete core sampling points and providing valuable insights for reservoir characterization and energy transition applications such as carbon capture and underground hydrogen storage.

Photonic Quantum Machine Learning (PQML) is an emerging approach that integrates photonic quantum computing technologies with machine learning techniques to enable scalable and energy-efficient quantum information processing. Photonic systems offer advantages such as room-temperature operation, high-speed signal processing, and the ability to represent information in high-dimensional Hilbert spaces. However, noise remains a major challenge affecting the performance, reliability, and scalability of PQML implementations. This review provides a systematic analysis of noise sources in photonic quantum machine learning systems. We discuss photonic quantum computing architectures and examine key quantum machine learning algorithms implemented on photonic platforms, including Variational Quantum Circuits, Quantum Neural Networks, and Quantum Support Vector Machines. The paper categorizes major noise mechanisms and analyzes their impact on learning performance, training stability, and convergence behavior. Furthermore, we review both traditional and advanced noise characterization techniques and survey recent strategies for noise mitigation in photonic quantum systems. Finally, we highlight recent experimental advances and discuss future research directions for developing robust and scalable PQML systems under realistic noise conditions.

Current evaluation frameworks and benchmarks for LLM powered agents focus on text chat driven agents, these frameworks do not expose the persona of user to the agent, thus operating in a user agnostic environment. Importantly, in customer experience management domain, the agent's behaviour evolves as the agent learns about user personality. With proliferation of real time TTS and multi-modal language models, LLM based agents are gradually going to become multi-modal. Towards this, we propose the MM-tau-p$^2$ benchmark with metrics for evaluating the robustness of multi-modal agents in dual control setting with and without persona adaption of user, while also taking user inputs in the planning process to resolve a user query. In particular, our work shows that even with state of-the-art frontier LLMs like GPT-5, GPT 4.1, there are additional considerations measured using metrics viz. multi-modal robustness, turn overhead while introducing multi-modality into LLM based agents. Overall, MM-tau-p$^2$ builds on our prior work FOCAL and provides a holistic way of evaluating multi-modal agents in an automated way by introducing 12 novel metrics. We also provide estimates of these metrics on the telecom and retail domains by using the LLM-as-judge approach using carefully crafted prompts with well defined rubrics for evaluating each conversation.

Modern edge applications increasingly require multi-DNN inference systems to execute tasks on heterogeneous processors, gaining performance from both concurrent execution and from matching each model to the most suited accelerator. However, existing systems support only a single model (or a few sparse variants) per task, which impedes the efficiency of this matching and results in high Service Level Objective violation rates. We introduce model stitching for multi-DNN inference systems, which creates model variants by recombining subgraphs from sparse models without re-training. We present a demonstrator system, SparseLoom, that shows model stitching can be deployed to SoCs. We show experimentally that SparseLoom reduces SLO violation rates by up to 74%, improves throughput by up to 2.31x, and lowers memory overhead by an average of 28% compared to state-of-the-art multi-DNN inference systems.

LLM agents that store knowledge as natural language suffer steep retrieval degradation as condition count grows, often struggle to compose learned rules reliably, and typically lack explicit mechanisms to detect stale or adversarial knowledge. We introduce PRECEPT, a unified framework for test-time adaptation with three tightly coupled components: (1) deterministic exact-match rule retrieval over structured condition keys, (2) conflict-aware memory with Bayesian source reliability and threshold-based rule invalidation, and (3) COMPASS, a Pareto-guided prompt-evolution outer loop. Exact retrieval eliminates partial-match interpretation errors on the deterministic path (0% by construction, vs 94.4% under Theorem~B.6's independence model at N=10) and supports compositional stacking through a semantic tier hierarchy; conflict-aware memory resolves static--dynamic disagreements and supports drift adaptation; COMPASS evaluates prompts through the same end-to-end execution pipeline. Results (9--10 seeds): PRECEPT achieves a +41.1pp first-try advantage over Full Reflexion (d>1.9), +33.3pp compositional generalization (d=1.55), 100% $P_1$ on 2-way logistics compositions (d=2.64), +40--55pp continuous learning gains, strong eventual robustness under adversarial static knowledge (100% logistics with adversarial SK active; partial recovery on integration), +55.0pp drift recovery (d=0.95, p=0.031), and 61% fewer steps. Core comparisons are statistically significant, often at p<0.001.

Radiology reports capture crucial longitudinal information on tumor burden, treatment response, and disease progression, yet their unstructured narrative format complicates automated analysis. While large language models (LLMs) have advanced clinical text processing, most state-of-the-art systems remain proprietary, limiting their applicability in privacy-sensitive healthcare environments. We present a fully open-source, locally deployable pipeline for longitudinal information extraction from radiology reports, implemented using the \texttt{llm\_extractinator} framework. The system applies the \texttt{qwen2.5-72b} model to extract and link target, non-target, and new lesion data across time points in accordance with RECIST criteria. Evaluation on 50 Dutch CT Thorax/Abdomen report pairs yielded high extraction performance, with attribute-level accuracies of 93.7\% for target lesions, 94.9\% for non-target lesions, and 94.0\% for new lesions. The approach demonstrates that open-source LLMs can achieve clinically meaningful performance in multi-timepoint oncology tasks while ensuring data privacy and reproducibility. These results highlight the potential of locally deployable LLMs for scalable extraction of structured longitudinal data from routine clinical text.

3D Gaussian Splatting (3DGS) has emerged as a powerful technique for novel view synthesis, subsequently extending into numerous spatial AI applications. However, most existing 3DGS methods are isolated, focusing on specific domains such as online SLAM, semantic enrichment, or 3DGS for unposed images. In this paper, we introduce X-GS, an extensible open framework that unifies a broad range of techniques to enable real-time 3DGS-based online SLAM enriched with semantics, bridging the gap to downstream multimodal models. At the core of X-GS is a highly efficient pipeline called X-GS-Perceiver, capable of taking unposed RGB (or optionally RGB-D) video streams as input to co-optimize geometry and poses, and distill high-dimensional semantic features from vision foundation models into the 3D Gaussians. We achieve real-time performance through a novel online Vector Quantization (VQ) module, a GPU-accelerated grid-sampling scheme, and a highly parallelized pipeline design. The semantic 3D Gaussians can then be utilized by vision-language models within the X-GS-Thinker component, enabling downstream tasks such as object detection, zero-shot caption generation, and potentially embodied tasks. Experimental results on real-world datasets showcase the efficacy, efficiency, and newly unlocked multimodal capabilities of the X-GS framework.

Deep learning models benefit from increasing data diversity and volume, motivating synthetic data augmentation to improve existing datasets. However, existing evaluation metrics for synthetic data typically calculate latent feature similarity, which is difficult to interpret and does not always correlate with the contribution to downstream tasks. We propose a vision-language grounded framework for interpretable synthetic data augmentation and evaluation in remote sensing. Our approach combines generative models, semantic segmentation and image captioning with vision and language models. Based on this framework, we introduce ARAS400k: A large-scale Remote sensing dataset Augmented with Synthetic data for segmentation and captioning, containing 100k real images and 300k synthetic images, each paired with segmentation maps and descriptions. ARAS400k enables the automated evaluation of synthetic data by analyzing semantic composition, minimizing caption redundancy, and verifying cross-modal consistency between visual structures and language descriptions. Experimental results indicate that while models trained exclusively on synthetic data reach competitive performance levels, those trained with augmented data (a combination of real and synthetic images) consistently outperform real-data baselines. Consequently, this work establishes a scalable benchmark for remote sensing tasks, specifically in semantic segmentation and image captioning. The dataset is available at zenodo.org/records/18890661 and the code base at github.com/caglarmert/ARAS400k.

As artificial intelligence (AI) systems evolve from stateless chatbots to autonomous multi-step agents, prompt engineering (PE), the discipline of crafting individual queries, proves necessary but insufficient. This paper introduces context engineering (CE) as a standalone discipline concerned with designing, structuring, and managing the entire informational environment in which an AI agent makes decisions. Drawing on vendor architectures (Google ADK, Anthropic, LangChain), current academic work (ACE framework, Google DeepMind's intelligent delegation), enterprise research (Deloitte, 2026; KPMG, 2026), and the author's experience building a multi-agent system, the paper proposes five context quality criteria: relevance, sufficiency, isolation, economy, and provenance, and frames context as the agent's operating system. Two higher-order disciplines follow. Intent engineering (IE) encodes organizational goals, values, and trade-off hierarchies into agent infrastructure. Specification engineering (SE) creates a machine-readable corpus of corporate policies and standards enabling autonomous operation of multi-agent systems at scale. Together these four disciplines form a cumulative pyramid maturity model of agent engineering, in which each level subsumes the previous one as a necessary foundation. Enterprise data reveals a gap: while 75% of enterprises plan agentic AI deployment within two years (Deloitte, 2026), deployment has surged and retreated as organizations confront scaling complexity (KPMG, 2026). The Klarna case illustrates a dual deficit, contextual and intentional. Whoever controls the agent's context controls its behavior; whoever controls its intent controls its strategy; whoever controls its specifications controls its scale.

We identify a systematic attention collapse pathology in the BLOOM family of transformer language models, where ALiBi positional encoding causes 31-44% of attention heads to attend almost entirely to the beginning-of-sequence token. The collapse follows a predictable pattern across four model scales (560M to 7.1B parameters), concentrating in head indices where ALiBi's slope schedule imposes the steepest distance penalties. We introduce surgical reinitialization: targeted Q/K/V reinitialization with zeroed output projections and gradient-masked freezing of all non-surgical parameters. Applied to BLOOM-1b7 on a single consumer GPU, the technique recovers 98.7% operational head capacity (242 to 379 of 384 heads) in two passes. A controlled comparison with C4 training data confirms that reinitialization -- not corpus content -- drives recovery, and reveals two distinct post-surgical phenomena: early global functional redistribution that improves the model, and late local degradation that accumulates under noisy training signal. An extended experiment reinitializing mostly-healthy heads alongside collapsed ones produces a model that transiently outperforms stock BLOOM-1b7 by 25% on training perplexity (12.70 vs. 16.99), suggesting that pretrained attention configurations are suboptimal local minima. Code, checkpoints, and diagnostic tools are released as open-source software.

Brain network analysis based on functional Magnetic Resonance Imaging (fMRI) is pivotal for diagnosing brain disorders. Existing approaches typically rely on predefined functional sub-networks to construct sub-network associations. However, we identified many cross-network interaction patterns with high Pearson correlations that this strict, prior-based organization fails to capture. To overcome this limitation, we propose the Brain Hierarchical Organization Learning (BrainHO) to learn inherently hierarchical brain network dependencies based on their intrinsic features rather than predefined sub-network labels. Specifically, we design a hierarchical attention mechanism that allows the model to aggregate nodes into a hierarchical organization, effectively capturing intricate connectivity patterns at the subgraph level. To ensure diverse, complementary, and stable organizations, we incorporate an orthogonality constraint loss, alongside a hierarchical consistency constraint strategy, to refine node-level features using high-level graph semantics. Extensive experiments on the publicly available ABIDE and REST-meta-MDD datasets demonstrate that BrainHO not only achieves state-of-the-art classification performance but also uncovers interpretable, clinically significant biomarkers by precisely localizing disease-related sub-networks.

Non-negative matrix factorisation (NMF) is a widely used tool for unsupervised learning and feature extraction, with applications ranging from genomics to text analysis and signal processing. Standard formulations of NMF are typically derived under Gaussian or Poisson noise assumptions, which may be inadequate for data exhibiting overdispersion or other complex mean-variance relationships. In this paper, we develop a unified framework for both traditional and convex NMF under a broad class of distributional assumptions, including Negative Binomial and Tweedie models, where the connection between the Tweedie and the $β$-divergence is also highlighted. Using a Majorize-Minimisation approach, we derive multiplicative update rules for all considered models, and novel updates for convex NMF with Poisson and Negative Binomial cost functions. We provide a unified implementation of all considered models, including the first implementations of several convex NMF models. Empirical evaluations on mutational and word count data demonstrate that the choice of noise model critically affects model fit and feature recovery, and that convex NMF can provide an efficient and robust alternative to traditional NMF in scenarios where the number of classes is large. The code for our proposed updates is available in the R package nmfgenr and can be found at https://github.com/MartaPelizzola/nmfgenr.

Brains remain unrivaled in their ability to recognize and generate complex spatiotemporal patterns. While AI is able to reproduce some of these capabilities, deep learning algorithms remain largely at odds with our current understanding of brain circuitry and dynamics. This is prominently the case for backpropagation through time (BPTT), the go-to algorithm for learning complex temporal dependencies. In this work we propose a general formalism to approximate BPTT in a controlled, biologically plausible manner. Our approach builds on, unifies and extends several previous approaches to local, time-continuous, phase-free spatiotemporal credit assignment based on principles of energy conservation and extremal action. Our starting point is a prospective energy function of neuronal states, from which we calculate real-time error dynamics for time-continuous neuronal networks. In the general case, this provides a simple and straightforward derivation of the adjoint method result for neuronal networks, the time-continuous equivalent to BPTT. With a few modifications, we can turn this into a fully local (in space and time) set of equations for neuron and synapse dynamics. Our theory provides a rigorous framework for spatiotemporal deep learning in the brain, while simultaneously suggesting a blueprint for physical circuits capable of carrying out these computations. These results reframe and extend the recently proposed Generalized Latent Equilibrium (GLE) model.

Generative AI and agentic tools are reshaping agile software development, yet many engineering curricula still teach agile methods and AI competencies separately and largely lecture-based. This paper presents a project-based AI Engineering curriculum designed to prepare students for AI-driven agile development by integrating agile practices and AI-enabled engineering throughout the program. We contribute (1) the curriculum concept and guiding principles, (2) a case study of interdisciplinary, AI-enabled agile student projects, and (3) early evidence from a mixed-methods evaluation. In our case study, second-semester bachelor students work in teams over seven two-week sprints on a realistic software product. AI tools are embedded into everyday agile engineering tasks - requirements clarification, backlog refinement, architectural reasoning, coding support, testing, and documentation - paired with reflection on human responsibility and quality. Initial results indicate that the integrated approach supports hands-on competence development in AI-assisted engineering. Key observations highlight the need for agile teaching adaptations due to rapid tool evolution, the critical role of oral verification to ensure foundational learning. We close with lessons learned and recommendations for educators designing agile project-based curricula in the age of AI.

Political scientists increasingly face a consequential choice when adopting natural language processing tools: build a domain-specific model from scratch, borrow and adapt an existing one, or simply fine-tune a general-purpose model on task data? Each approach occupies a different point on the spectrum of performance, cost, and required expertise, yet the discipline has offered little empirical guidance on how to navigate this trade-off. This paper provides such guidance. Using conflict event classification as a test case, I fine-tune ModernBERT on the Global Terrorism Database (GTD) to create Confli-mBERT and systematically compare it against ConfliBERT, a domain-specific pretrained model that represents the current gold standard. Confli-mBERT achieves 75.46% accuracy compared to ConfliBERT's 79.34%. Critically, the four-percentage-point gap is not uniform: on high-frequency attack types such as Bombing/Explosion (F1 = 0.95 vs. 0.96) and Kidnapping (F1 = 0.92 vs. 0.91), the models are nearly indistinguishable. Performance differences concentrate in rare event categories comprising fewer than 2% of all incidents. I use these findings to develop a practical decision framework for political scientists considering any NLP-assisted research task: when does the research question demand a specialized model, and when does an accessible fine-tuned alternative suffice? The answer, I argue, depends not on which model is "better" in the abstract, but on the specific intersection of class prevalence, error tolerance, and available resources. The model, training code, and data are publicly available on Hugging Face.

This paper studies the memorization capacity of deep neural networks with ReLU activation. Specifically, we investigate the minimal size of such networks to memorize any $N$ data points in the unit ball with pairwise separation distance $δ$ and discrete labels. Most prior studies characterize the memorization capacity by the number of parameters or neurons. We generalize these results by constructing neural networks, whose width $W$ and depth $L$ satisfy $W^2L^2= \mathcal{O}(N\log(δ^{-1}))$, that can memorize any $N$ data samples. We also prove that any such networks should also satisfy the lower bound $W^2L^2=Ω(N \log(δ^{-1}))$, which implies that our construction is optimal up to logarithmic factors when $δ^{-1}$ is polynomial in $N$. Hence, we explicitly characterize the trade-off between width and depth for the memorization capacity of deep neural networks in this regime.

The nonparametric variational information bottleneck (NVIB) provides the foundation for nonparametric variational differential privacy (NVDP), a framework for building privacy-preserving language models. However, the learned latent representations can drift into regions with high information content, leading to poor privacy guarantees, but also low utility due to numerical instability during training. In this work, we introduce a principled parameter clipping strategy to directly address this issue. Our method is mathematically derived from the objective of minimizing the Rényi Divergence (RD) upper bound, yielding specific, theoretically grounded constraints on the posterior mean, variance, and mixture weight parameters. We apply our technique to an NVIB based model and empirically compare it against an unconstrained baseline. Our findings demonstrate that the clipped model consistently achieves tighter RD bounds, implying stronger privacy, while simultaneously attaining higher performance on several downstream tasks. This work presents a simple yet effective method for improving the privacy-utility trade-off in variational models, making them more robust and practical.

Adam is a widely used optimization algorithm in deep learning, yet the specific class of objective functions where it exhibits inherent advantages remains underexplored. Unlike prior studies requiring external schedulers and $β_2$ near 1 for convergence, this work investigates the "natural" auto-convergence properties of Adam. We identify a class of highly degenerate polynomials where Adam converges automatically without additional schedulers. Specifically, we derive theoretical conditions for local asymptotic stability on degenerate polynomials and demonstrate strong alignment between theoretical bounds and experimental results. We prove that Adam achieves local linear convergence on these degenerate functions, significantly outperforming the sub-linear convergence of Gradient Descent and Momentum. This acceleration stems from a decoupling mechanism between the second moment $v_t$ and squared gradient $g_t^2$, which exponentially amplifies the effective learning rate. Finally, we characterize Adam's hyperparameter phase diagram, identifying three distinct behavioral regimes: stable convergence, spikes, and SignGD-like oscillation.

Continual learning in transformers is commonly addressed through parameter-efficient adaptation: prompts, adapters, or LoRA modules are specialized per task while the backbone remains frozen. Although effective in controlled multi-epoch settings, these approaches rely on gradual gradient-based specialization and struggle in Online Continual Learning (OCL), where data arrive as a non-stationary stream and each sample may be observed only once. We recast continual learning in transformers as a routing problem: under strict online constraints, the model must dynamically select the appropriate representational subspace for each input without explicit task identifiers or repeated optimization. We thus introduce Routing without Forgetting (RwF), a transformer architecture augmented with energy-based associative retrieval layers inspired by Modern Hopfield Networks. Instead of storing or merging task-specific prompts, RwF generates dynamic prompts through single-step associative retrieval over the transformer token embeddings at each layer. Retrieval corresponds to the closed-form minimization of a strictly convex free-energy functional, enabling input-conditioned routing within each forward pass, independently of iterative gradient refinement. Across challenging class-incremental benchmarks, RwF improves over existing prompt-based methods. On Split-ImageNet-R and Split-ImageNet-S, RwF outperforms prior prompt-based approaches by a large margin, even in few-shot learning regimes. These results indicate that embedding energy-based associative routing directly within the transformer backbone provides a principled and effective foundation for OCL.

Distilling humanoid locomotion control from offline datasets into deployable policies remains a challenge, as existing methods rely on privileged full-body states that require complex and often unreliable state estimation. We present Sensor-Conditioned Diffusion Policies (SCDP) that enables humanoid locomotion using only onboard sensors, eliminating the need for explicit state estimation. SCDP decouples sensing from supervision through mixed-observation training: diffusion model conditions on sensor histories while being supervised to predict privileged future state-action trajectories, enforcing the model to infer the motion dynamics under partial observability. We further develop restricted denoising, context distribution alignment, and context-aware attention masking to encourage implicit state estimation within the model and to prevent train-deploy mismatch. We validate SCDP on velocity-commanded locomotion and motion reference tracking tasks. In simulation, SCDP achieves near-perfect success on velocity control (99-100%) and 93% tracking success in AMASS test set, performing comparable to privileged baselines while using only onboard sensors. Finally, we deploy the trained policy on a real G1 humanoid at 50 Hz, demonstrating robust real robot locomotion without external sensing or state estimation.

In this paper, we develop a rigorous optimal control-theoretic approach to Transformer training that respects key structural constraints such as (i) realized-input-independence during execution, (ii) the ensemble control nature of the problem, and (iii) positional dependence. We model the Transformer architecture as a discrete-time controlled particle system with shared actions, exhibiting noise-free McKean-Vlasov dynamics. While the resulting dynamics is not Markovian, we show that lifting it to probability measures produces a fully-observed Markov decision process (MDP). Positional encodings are incorporated into the state space to preserve the sequence order under lifting. Using the dynamic programming principle, we establish the existence of globally optimal policies under mild assumptions of compactness. We further prove that closed-loop policies in the lifted is equivalent to an initial-distribution dependent open-loop policy, which are realized-input-independent and compatible with standard Transformer training. To train a Transformer, we propose a triply quantized training procedure for the lifted MDP by quantizing the state space, the space of probability measures, and the action space, and show that any optimal policy for the triply quantized model is near-optimal for the original training problem. Finally, we establish stability and empirical consistency properties of the lifted model by showing that the value function is continuous with respect to the perturbations of the initial empirical measures and convergence of policies as the data size increases. This approach provides a globally optimal and robust alternative to gradient-based training without requiring smoothness or convexity.

The traditional Triangular Maximally Filtered Graph (TMFG) construction requires pre-computation and storage of a dense correlation matrix; this limits its applicability to small and medium-sized datasets. Here we identify key memory and runtime complexity challenges when using TMFG at scale. We then present the Approximate Triangular Maximally Filtered Graph (a-TMFG) algorithm. This is a novel approach to scaling the construction of artificial graphs from data inspired by TMFG. The method employs k-Nearest Neighbors Graphs (kNNG) for initial construction, and implements a memory management strategy to search and estimate missing correlations on-the-fly. This provides representations to control combinatorial explosion. The algorithm is tested for robustness to the parameters and noise, and is evaluated on datasets with millions of observations. This new method provides a parsimonious way to construct graphs for use-cases where graphs are used as input to supervised and unsupervised learning but where no natural graph exists.

We study constraint-based structure learning of Markov networks and Bayesian networks in the presence of an unreliable conditional independence oracle that makes at most a bounded number of errors. For Markov networks, we observe that a low maximum number of vertex-wise disjoint paths implies that the structure is uniquely identifiable even if the number of errors is (moderately) exponential in the number of vertices. For Bayesian networks, however, we prove that one cannot tolerate any errors to always identify the structure even when many commonly used graph parameters like treewidth are bounded. Finally, we give algorithms for structure learning when the structure is uniquely identifiable.

Large audio language models (ALMs) extend LLMs with auditory understanding. A common approach freezes the LLM and trains only an adapter on self-generated targets. However, this fails for reasoning LLMs (RLMs) whose built-in chain-of-thought traces expose the textual surrogate input, yielding unnatural responses. We propose self-rephrasing, converting self-generated responses into audio-understanding variants compatible with RLMs while preserving distributional alignment. We further fuse and compress multiple audio encoders for stronger representations. For training, we construct a 6M-instance multi-task corpus (2.5M unique prompts) spanning 19K hours of speech, music, and sound. Our 4B-parameter ALM outperforms similarly sized models and surpasses most larger ALMs on related audio-reasoning benchmarks, while preserving textual capabilities with a low training cost. Notably, we achieve the best open-source result on the MMAU-speech and MMSU benchmarks and rank third among all the models.

State-space model releases are typically coupled to fused CUDA and Triton kernels, inheriting a hard dependency on NVIDIA hardware. We show that Mamba-2's state space duality algorithm -- diagonal state structure, chunkable recurrence, and einsum-dominated compute with static control flow -- maps cleanly onto what XLA's fusion and tiling passes actually optimise, making custom kernels optional rather than required. We implement the full inference path (prefill, cached autoregressive decoding) as shaped standard primitives under XLA, without hand-written kernels, and realise the architecture's theoretical $O(1)$ state management as a compiled on-device cache requiring no host synchronisation during generation. The implementation runs unmodified on CPU, NVIDIA GPU, and Google Cloud TPU from a single JAX source. On TPU v6e across five model scales (130M--2.7B parameters), XLA-generated code reaches approximately 140 TFLOPS on single-stream prefill ($15%$ MFU) and up to $64%$ bandwidth utilisation on decode. Greedy decoding matches the PyTorch/CUDA reference token-for-token across 64 steps, with hidden-state agreement within float32 rounding tolerance. The pattern transfers to any SSM recurrence satisfying the same structural conditions, on any platform with a mature XLA backend. The implementation is publicly available at https://github.com/CosmoNaught/mamba2-jax and merged into the Bonsai JAX model library.

This study proposes a novel methodology for generating personalized fake news debunking messages by prompting Large Language Models (LLMs) with persona-based inputs aligned to the Big Five personality traits: Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness. Our approach guides LLMs to transform generic debunking content into personalized versions tailored to specific personality profiles. To assess the effectiveness of these transformations, we employ a separate LLM as an automated evaluator simulating corresponding personality traits, thereby eliminating the need for costly human evaluation panels. Our results show that personalized messages are generally seen as more persuasive than generic ones. We also find that traits like Openness tend to increase persuadability, while Neuroticism can lower it. Differences between LLM evaluators suggest that using multiple models provides a clearer picture. Overall, this work demonstrates a practical way to create more targeted debunking messages exploiting LLMs, while also raising important ethical questions about how such technology might be used.

Bandits with noncompliance separate the learner's recommendation from the treatment actually delivered, so the learning target itself must be chosen. A platform may care about recommendation welfare in the current mediated workflow, treatment learning for a future direct-control regime, or anytime-valid uncertainty for one of those targets. These objectives need not agree. We formalize this objective-choice problem, identify the direct-control regime in which recommendation and treatment objectives collapse, and show by example that recommendation welfare can strictly exceed every learner-measurable treatment policy when downstream actors use private information. For finite-context square-IV problems we propose BRACE, a parameter-free phase-doubling algorithm that performs IV inversion only after matrix certification and otherwise returns full-range but honest structural intervals. BRACE delivers simultaneous policy-value validity, fixed-gap identification of the operationally optimal recommendation policy, and fixed-gap identification of the structurally optimal treatment policy under contextual homogeneity and invertibility. We complement the theory with a finite-context empirical benchmark spanning direct control, mediated present-versus-future tradeoffs, weak identification, homogeneity failure, and rectangular overidentification. The experiments show that safety appears as regret on easy problems, as abstention and wide valid intervals under weak identification, as a reason to prefer recommendation welfare under homogeneity failure, and as tighter structural uncertainty when extra instruments are available. For rich contexts, we also derive an orthogonal score whose conditional bias factorizes into compliance-model and outcome-model errors, clarifying what must be stabilized for anytime-valid semiparametric IV inference.

Speculative decoding accelerates LLM inference but suffers from performance degradation when target models are fine-tuned for specific domains. A naive solution is to retrain draft models for every target model, which is costly and inefficient. To address this, we introduce a parameter- and data-efficient framework named Efficient Draft Adaptation, abbreviated as EDA, for efficiently adapting draft models. EDA introduces three innovations: (1) a decoupled architecture that utilizes shared and private components to model the shared and target-specific output distributions separately, enabling parameter-efficient adaptation by updating only the lightweight private component;(2) a data regeneration strategy that utilizes the fine-tuned target model to regenerate training data, thereby improving the alignment between training and speculative decoding, leading to higher average acceptance length;(3) a sample selection mechanism that prioritizes high-value data for efficient adaptation. Our experiments show that EDA effectively restores speculative performance on fine-tuned models, achieving superior average acceptance lengths with significantly reduced training costs compared to full retraining. Code is available at https://github.com/Lyn-Lucy/Efficient-Draft-Adaptation.

When language models are trained on synthetic data, they (student model) can covertly acquire behavioral traits from the data-generating model (teacher model). Subliminal learning refers to the transmission of traits from a teacher to a student model via training on data unrelated to those traits. Prior work demonstrated this in the training domains of number sequences, code, and math Chain-of-Thought traces including transmission of misaligned behaviors. We investigate whether transmission occurs through natural language paraphrases with fixed semantic content, and whether content explicitly contradicting the teacher's preference can block it. We find that training on paraphrases from a teacher system-prompted to love a particular animal increases a student's preference for that animal by up to 19 percentage points. This occurs when paraphrased content is semantically unrelated to the animal, or even when it explicitly expresses dislike. The transmission succeeds despite aggressive filtering to ensure paraphrase fidelity. This raises concerns for pipelines where models generate their own training data: content-based inspection cannot detect such transmission, and even preference-contradicting content fails to prevent it.

On-device tuning of deep neural networks enables long-term adaptation at the edge while preserving data privacy. However, the high computational and memory demands of backpropagation pose significant challenges for ultra-low-power, memory-constrained extreme-edge devices. These challenges are further amplified for attention-based models due to their architectural complexity and computational scale. We present TrainDeeploy, a framework that unifies efficient inference and on-device training on heterogeneous ultra-low-power System-on-Chips (SoCs). TrainDeeploy provides the first complete on-device training pipeline for extreme-edge SoCs supporting both Convolutional Neural Networks (CNNs) and Transformer models, together with multiple training strategies such as selective layer-wise fine-tuning and Low-Rank Adaptation (LoRA). On a RISC-V-based heterogeneous SoC, we demonstrate the first end-to-end on-device fine-tuning of a Compact Convolutional Transformer (CCT), achieving up to 11 trained images per second. We show that LoRA reduces dynamic memory usage by 23%, decreases the number of trainable parameters and gradients by 15x, and reduces memory transfer volume by 1.6x compared to full backpropagation. TrainDeeploy achieves up to 4.6 FLOP/cycle on CCT (0.28M parameters, 71-126M FLOPs) and up to 13.4 FLOP/cycle on Deep-AE (0.27M parameters, 0.8M FLOPs), while expanding the scope of prior frameworks to support both CNN and Transformer models with parameter-efficient tuning on extreme-edge platforms.

Phoneme frequency distributions exhibit robust statistical regularities across languages, including exponential-tailed rank-frequency patterns and a negative relationship between phonemic inventory size and the relative entropy of the distribution. The origin of these patterns remains largely unexplained. In this paper, we investigate whether they can arise as consequences of the historical processes that shape phonological systems. We introduce a stochastic model of phonological change and simulate the diachronic evolution of phoneme inventories. A naïve version of the model reproduces the general shape of phoneme rank-frequency distributions but fails to capture other empirical properties. Extending the model with two additional assumptions -- an effect related to functional load and a stabilising tendency toward a preferred inventory size -- yields simulations that match both the observed distributions and the negative relationship between inventory size and relative entropy. These results suggest that some statistical regularities of phonological systems may arise as natural consequences of diachronic sound change rather than from explicit optimisation or compensatory mechanisms.

Manual development of automatic tests for embedded C software is a strenuous and time-consuming task that does not scale well. With the accelerating pace of software release cycles, verification increasingly becomes the bottleneck in the embedded development workflow. This paper presents a Retrieval-Augmented Generation (RAG) pipeline as a solution for partial automation of the verification process. By grounding a large language model in project-specific artifacts, the approach reduces hallucinations and improves project alignment. An industrial evaluation showed that the generated tests are 100 % syntactically correct, with 85 % successfully passing runtime validation. The proposed solution has the potential to save up to 66 % of the testing time compared to manual test writing while generating 270 tests per hour.

The adaptation of large-scale vision-language models (VLMs) to downstream tasks with limited labeled data remains a significant challenge. While parameter-efficient prompt learning methods offer a promising path, they often suffer from catastrophic forgetting of pre-trained knowledge. Toward addressing this limitation, our work is grounded in the insight that governing the evolutionary path of prompts is essential for forgetting-free adaptation. To this end, we propose EvoPrompt, a novel framework designed to explicitly steer the prompt trajectory for stable, knowledge-preserving fine-tuning. Specifically, our approach employs a Modality-Shared Prompt Projector (MPP) to generate hierarchical prompts from a unified embedding space. Critically, an evolutionary training strategy decouples low-rank updates into directional and magnitude components, preserving early-learned semantic directions while only adapting their magnitude, thus enabling prompts to evolve without discarding foundational knowledge. This process is further stabilized by Feature Geometric Regularization (FGR), which enforces feature decorrelation to prevent representation collapse. Extensive experiments demonstrate that EvoPrompt achieves state-of-the-art performance in few-shot learning while robustly preserving the original zero-shot capabilities of pre-trained VLMs.

Processing regulations and resulting requirements to achieve regulatory compliance in software engineering (SE) is a developing challenge due to the continuously growing amount, complexity, and expanding scope of regulations. Despite the growing amount of newly suggested regulatory requirements engineering (RE) approaches by the research community, industry remains under pressure to assure their integration into their RE and overall software development life cycle (SDLC) practices to facilitate a seamless and legally valid compliance by design. As of today, we still have limited empirical understanding of how this can be achieved. Such integration should avoid additional burdens and address the demands of legal knowledge intensity, cross-functional communication and consistency between different involved viewpoints. Intermediary results of this doctoral study showed that regulatory RE has peculiarities distinguishing it from the engineering of other requirements. Oftentimes, organizations establish standalone regulatory RE processes on the organizational level. However, software development teams usually approach compliance by design in an ad-hoc manner, rather than in a systematic way. Among other, because of the complexity of the coordination between the involved viewpoints. The goal of this paper is to report and get feedback about the synthesis and future evaluation of our Artefact Model for Regulatory Requirements Engineering (AM4RRE) for a integrated compliance by design. We hope this paper will spark discussions about regulatory RE and help us refine plans for the final stage of the doctoral study.

This paper introduces temporal-conditioned normalizing flows (tcNF), a novel framework that addresses anomaly detection in time series data with accurate modeling of temporal dependencies and uncertainty. By conditioning normalizing flows on previous observations, tcNF effectively captures complex temporal dynamics and generates accurate probability distributions of expected behavior. This autoregressive approach enables robust anomaly detection by identifying low-probability events within the learned distribution. We evaluate tcNF on diverse datasets, demonstrating good accuracy and robustness compared to existing methods. A comprehensive analysis of strengths and limitations and open-source code is provided to facilitate reproducibility and future research.

The encounter between human reasoning and generative artificial intelligence (GenAI) cannot be adequately described by inherited metaphors of tool use, augmentation, or collaborative partnership. This article argues that such interactions produce a qualitatively distinct cognitive-epistemic formation, designated here as the Third Entity: an emergent, transient structure that arises from the transductive coupling of two ontologically incommensurable modes of cognition. Drawing on Peirce semiotics, Polanyi theory of tacit knowledge, Simondon philosophy of individuation, Ihde postphenomenology, and Morin complexity theory, we develop a multi-layered theoretical account of this formation. We introduce the concept of vibe-creation to designate the pre-reflective cognitive mode through which the Third Entity navigates high-dimensional semantic space and argue that this mode constitutes the automation of tacit knowledge - a development with far-reaching consequences for epistemology, the philosophy of mind, and educational theory. We further propose the notion of asymmetric emergence to characterize the agency of the Third Entity: genuinely novel and irreducible, yet anchored in human intentional responsibility. The article concludes by examining the implications of this theoretical framework for the transformation of educational institutions and the redefinition of intellectual competence in the age of GenAI.

We present GenePlan (GENeralized Evolutionary Planner), a novel framework that leverages large language model (LLM) assisted evolutionary algorithms to generate domain-dependent generalized planners for classical planning tasks described in PDDL. By casting generalized planning as an optimization problem, GenePlan iteratively evolves interpretable Python planners that minimize plan length across diverse problem instances. In empirical evaluation across six existing benchmark domains and two new domains, GenePlan achieved an average SAT score of 0.91, closely matching the performance of the state-of-the-art planners (SAT score 0.93), and significantly outperforming other LLM-based baselines such as chain-of-thought (CoT) prompting (average SAT score 0.64). The generated planners solve new instances rapidly (average 0.49 seconds per task) and at low cost (average $1.82 per domain using GPT-4o).

Goal-conditioned systems assume goals are provided externally. We ask whether attentional priorities can emerge endogenously from an agent's internal cognitive state. We propose a priority function that generates observation targets from three epistemic gaps: ignorance (posterior variance), surprise (prediction error), and staleness (temporal decay of confidence in unobserved variables). We validate this in two systems: a minimal attention-allocation environment (2,000 runs) and a modular, partially observable world (500 runs). Ablation shows each component is necessary. A key finding is metric-dependent reversal: under global prediction error, coverage-based rotation wins; under change detection latency, priority-guided allocation wins, with advantage growing monotonically with dimensionality (d = -0.95 at N=48, p < 10^-6). Detection latency follows a power law in attention budget, with a steeper exponent for priority-guided allocation (0.55 vs. 0.40). When the decay rate is made learnable per variable, the system spontaneously recovers environmental volatility structure without supervision (t = 22.5, p < 10^-6). We demonstrate that epistemic gaps alone, without external reward, suffice to generate adaptive priorities that outperform fixed strategies and recover latent environmental structure.

There is a growing demand for software engineering education (SEE) for professionals because of the increasing demand, active evolution of the technological landscape, and changes in the skills required by the practice. Integrating requirements engineering (RE) courses into SEE curricula for professionals systematically and effectively is challenging. In particular, curricula for professionals have different demands, are more dynamic, and modular in nature. In this study, we report on our experience in the development of three SEE curricula for professionals and the integration of RE courses into such curricula. We suggest basic principles for such integration and describe the systematic approach focused on course content mapping that we have developed.

Vision-Language-Action models have shown great promise for autonomous driving, yet they suffer from degraded perception after unfreezing the visual encoder and struggle with accumulated instability in long-term planning. To address these challenges, we propose EvoDriveVLA-a novel collaborative perception-planning distillation framework that integrates self-anchored perceptual constraints and oracle-guided trajectory optimization. Specifically, self-anchored visual distillation leverages self-anchor teacher to deliver visual anchoring constraints, regularizing student representations via trajectory-guided key-region awareness. In parallel, oracle-guided trajectory distillation employs a future-aware oracle teacher with coarse-to-fine trajectory refinement and Monte Carlo dropout sampling to produce high-quality trajectory candidates, thereby selecting the optimal trajectory to guide the student's prediction. EvoDriveVLA achieves SOTA performance in open-loop evaluation and significantly enhances performance in closed-loop evaluation. Our code is available at: https://github.com/hey-cjj/EvoDriveVLA.

Model merging unifies independently fine-tuned LLMs from the same base, enabling reuse and integration of parallel development efforts without retraining. However, in practice we observe that merging does not always succeed: certain combinations of task-specialist models suffer from catastrophic performance degradation after merging. We refer to this failure mode as merging collapse. Intuitively, collapse arises when the learned representations or parameter adjustments for different tasks are fundamentally incompatible, so that merging forces destructive interference rather than synergy. In this paper, we identify and characterize the phenomenon of task-level merging collapse, where certain task combinations consistently trigger huge performance degradation across all merging methods. Through extensive experiments and statistical analysis, we demonstrate that representational incompatibility between tasks is strongly correlated with merging collapse, while parameter-space conflict metrics show minimal correlation, challenging conventional wisdom in model merging literature. We provide a theoretical explanation on this phenomenon through rate-distortion theory with a dimension-dependent bound, establishing fundamental limits on task mergeability regardless of methodology.

Scenario-based testing is a key method for cost-effective and safe validation of autonomous vehicles (AVs). Existing approaches rely on imperative scenario definitions, requiring developers to manually enumerate numerous variants to achieve coverage. Declarative languages, such as OpenSCENARIO DSL (OS2), raise the abstraction level but lack systematic methods for instantiating concrete, specification-compliant scenarios as simulations. To our knowledge, currently, no open-source solution provides this capability. We present RoadLogic that bridges declarative OS2 specifications and executable simulations. It uses Answer Set Programming to generate abstract plans satisfying scenario constraints, motion planning to refine the plans into feasible trajectories, and specification-based monitoring to verify correctness. We evaluate RoadLogic on instantiating representative OS2 scenarios as simulations in the CommonRoad framework. Results show that RoadLogic consistently produces realistic, specification-satisfying simulations within minutes and captures diverse behavioral variants through parameter sampling, thus opening the door to systematic scenario-based testing for autonomous driving systems.

Foundation models are increasingly being deployed in contexts where understanding the uncertainty of their outputs is critical to ensuring responsible deployment. While Bayesian methods offer a principled approach to uncertainty quantification, their computational overhead renders their use impractical for training or inference at foundation model scale. State-of-the-art models achieve parameter counts in the trillions through carefully engineered sparsity including Mixture-of-Experts (MoE) layers. In this work, we demonstrate calibrated uncertainty at scale by introducing Variational Mixture-of-Experts Routing (VMoER), a structured Bayesian approach for modelling uncertainty in MoE layers. VMoER confines Bayesian inference to the expert-selection stage which is typically done by a deterministic routing network. We instantiate VMoER using two inference strategies: amortised variational inference over routing logits and inferring a temperature parameter for stochastic expert selection. Across tested foundation models, VMoER improves routing stability under noise by 38\%, reduces calibration error by 94\%, and increases out-of-distribution AUROC by 12\%, while incurring less than 1\% additional FLOPs. These results suggest VMoER offers a scalable path toward robust and uncertainty-aware foundation models.

Analyzing Open Source Intelligence (OSINT) from large volumes of data is critical for drafting and publishing comprehensive CTI reports. This process usually follows a three-stage workflow -- triage, deep search and TI drafting. While Large Language Models (LLMs) offer a promising route toward automation, existing benchmarks still have limitations. These benchmarks often consist of tasks that do not reflect real-world analyst workflows. For example, human analysts rarely receive tasks in the form of multiple-choice questions. Also, existing benchmarks often rely on model-centric metrics that emphasize lexical overlap rather than actionable, detailed insights essential for security analysts. Moreover, they typically fail to cover the complete three-stage workflow. To address these issues, we introduce CyberThreat-Eval, which is collected from the daily CTI workflow of a world-leading company. This expert-annotated benchmark assesses LLMs on practical tasks across all three stages as mentioned above. It utilizes analyst-centric metrics that measure factual accuracy, content quality, and operational costs. Our evaluation using this benchmark reveals important insights into the limitations of current LLMs. For example, LLMs often lack the nuanced expertise required to handle complex details and struggle to distinguish between correct and incorrect information. To address these challenges, the CTI workflow incorporates both external ground-truth databases and human expert knowledge. TRA allows human experts to iteratively provide feedback for continuous improvement. The code is available at \href{https://github.com/xschen-beb/CyberThreat-Eval}{\texttt{GitHub}} and \href{https://huggingface.co/datasets/xse/CyberThreat-Eval}{\texttt{HuggingFace}}.

Delineating the clinical target volume (CTV) in radiotherapy involves complex margins constrained by tumor location and anatomical barriers. While deep learning models automate this process, their rigid reliance on expert-annotated data requires costly retraining whenever clinical guidelines update. To overcome this limitation, we introduce OncoAgent, a novel guideline-aware AI agent framework that seamlessly converts textual clinical guidelines into three-dimensional target contours in a training-free manner. Evaluated on esophageal cancer cases, the agent achieves a zero-shot Dice similarity coefficient of 0.842 for the CTV and 0.880 for the planning target volume, demonstrating performance highly comparable to a fully supervised nnU-Net baseline. Notably, in a blinded clinical evaluation, physicians strongly preferred OncoAgent over the supervised baseline, rating it higher in guideline compliance, modification effort, and clinical acceptability. Furthermore, the framework generalizes zero-shot to alternative esophageal guidelines and other anatomical sites (e.g., prostate) without any retraining. Beyond mere volumetric overlap, our agent-based paradigm offers near-instantaneous adaptability to alternative guidelines, providing a scalable and transparent pathway toward interpretability in radiotherapy treatment planning.

We study off-policy evaluation in the setting of contextual bandits, where we aim to evaluate a new policy using historical data that consists of contexts, actions and received rewards. This historical data typically does not faithfully represent action distribution of the new policy accurately. A common approach, inverse probability weighting (IPW), adjusts for these discrepancies in action distributions. However, this method often suffers from high variance due to the probability being in the denominator. The doubly robust (DR) estimator reduces variance through modeling reward but does not directly address variance from IPW. In this work, we address the limitation of IPW by proposing a Nonparametric Weighting (NW) approach that constructs weights using a nonparametric model. Our NW approach achieves low bias like IPW but typically exhibits significantly lower variance. To further reduce variance, we incorporate reward predictions -- similar to the DR technique -- resulting in the Model-assisted Nonparametric Weighting (MNW) approach. The MNW approach yields accurate value estimates by explicitly modeling and mitigating bias from reward modeling, without aiming to guarantee the standard doubly robust property. Extensive empirical comparisons show that our approaches consistently outperform existing techniques, achieving lower variance in value estimation while maintaining low bias.

The rapid rollout of AI in heterogeneous public and societal sectors has subsequently escalated the need for compliance with regulatory standards and frameworks. The EU AI Act has emerged as a landmark in the regulatory landscape. The development of solutions that elicit the level of AI systems' compliance with such standards is often limited by the lack of resources, hindering the semi-automated or automated evaluation of their performance. This generates the need for manual work, which is often error-prone, resource-limited or limited to cases not clearly described by the regulation. This paper presents an open, transparent, and reproducible method of creating a resource that facilitates the evaluation of NLP models with a strong focus on RAG systems. We have developed a dataset that contain the tasks of risk-level classification, article retrieval, obligation generation, and question-answering for the EU AI Act. The dataset files are in a machine-to-machine appropriate format. To generate the files, we utilise domain knowledge as an exegetical basis, combining with the processing and reasoning power of large language models to generate scenarios along with the respective tasks. Our methodology demonstrates a way to harness language models for grounded generation with high document relevancy. Besides, we overcome limitations such as navigating the decision boundaries of risk-levels that are not explicitly defined within the EU AI Act, such as limited and minimal cases. Finally, we demonstrate our dataset's effectiveness by evaluating a RAG-based solution that reaches 0.87 and 0.85 F1-score for prohibited and high-risk scenarios.

Large Language Models (LLMs) are increasingly deployed across diverse real-world applications and user communities. As such, it is crucial that these models remain both morally grounded and knowledge-aware. In this work, we uncover a critical limitation of current LLMs -- their tendency to prioritize moral reasoning over commonsense understanding. To investigate this phenomenon, we introduce CoMoral, a novel benchmark dataset containing commonsense contradictions embedded within moral dilemmas. Through extensive evaluation of ten LLMs across different model sizes, we find that existing models consistently struggle to identify such contradictions without prior signal. Furthermore, we observe a pervasive narrative focus bias, wherein LLMs more readily detect commonsense contradictions when they are attributed to a secondary character rather than the primary (narrator) character. Our comprehensive analysis underscores the need for enhanced reasoning-aware training to improve the commonsense robustness of large language models.

Reinforcement Learning (RL) has demonstrated strong potential for industrial process control, yet policies trained in simulation often suffer from a significant sim-to-real gap when deployed on physical hardware. This work systematically analyzes how core Markov Decision Process (MDP) design choices -- state composition, target inclusion, reward formulation, termination criteria, and environment dynamics models -- affect this transfer. Using a color mixing task, we evaluate different MDP configurations and mixing dynamics across simulation and real-world experiments. We validate our findings on physical hardware, demonstrating that physics-based dynamics models achieve up to 50% real-world success under strict precision constraints where simplified models fail entirely. Our results provide practical MDP design guidelines for deploying RL in industrial process control.

We present CERES (Calibrated Early-warning and Risk Estimation System), an automated probabilistic forecasting system for acute food insecurity. CERES generates 90-day ahead probability estimates of IPC Phase 3+ (Crisis), Phase 4+ (Emergency), and Phase 5 (Famine) conditions for 43 high-risk countries globally, updated weekly. The system fuses six data streams, precipitation anomalies (CHIRPS), vegetation indices (MODIS NDVI), conflict events (ACLED), IPC classifications, food consumption scores (WFP), and cereal price indices (FAO/WFP) - through a logistic scoring model with author-specified initial coefficients and parametric input-perturbation intervals (n=2,000 draws). In historical back-validation against four IPC Phase 4-5 events selected for data completeness, CERES assigned TIER-1 classification in all four cases; these are in-sample sanity checks only, not prospective performance claims. All prospective predictions are timestamped, cryptographically identified, and archived for public verification against IPC outcome data at the T+90 horizon. To the author's knowledge, CERES is the first famine early warning system that is simultaneously: (1) probabilistic, (2) open-access, (3) continuously running, (4) machine-readable at prediction level, and (5) committed to public prospective verification of every prediction made.

Motion forecasting aims to predict the future trajectories of dynamic agents in the scene, enabling autonomous vehicles to effectively reason about scene evolution. Existing approaches operate under the closed-world regime and assume fixed object taxonomy as well as access to high-quality perception. Therefore, they struggle in real-world settings where perception is imperfect and object taxonomy evolves over time. In this work, we bridge this fundamental gap by introducing open-world motion forecasting, a novel setting in which new object classes are sequentially introduced over time and future object trajectories are estimated directly from camera images. We tackle this setting by proposing the first end-to-end class-incremental motion forecasting framework to mitigate catastrophic forgetting while simultaneously learning to forecast newly introduced classes. When a new class is introduced, our framework employs a pseudo-labeling strategy to first generate motion forecasting pseudo-labels for all known classes which are then processed by a vision-language model to filter inconsistent and over-confident predictions. Parallelly, our approach further mitigates catastrophic forgetting by using a novel replay sampling strategy that leverages query feature variance to sample previous sequences with informative motion patterns. Extensive evaluation on the nuScenes and Argoverse 2 datasets demonstrates that our approach successfully resists catastrophic forgetting and maintains performance on previously learned classes while improving adaptation to novel ones. Further, we demonstrate that our approach supports zero-shot transfer to real-world driving and naturally extends to end-to-end class-incremental planning, enabling continual adaptation of the full autonomous driving system. We provide the code at https://omen.cs.uni-freiburg.de .

Large Language Models (LLMs) excel in Natural Language Processing (NLP) tasks, but they often propagate biases embedded in their training data, which is potentially impactful in sensitive domains like healthcare. While existing benchmarks evaluate biases related to individual social determinants of health (SDoH) such as gender or ethnicity, they often overlook interactions between these factors and lack context-specific assessments. This study investigates bias in LLMs by probing the relationships between gender and other SDoH in French patient records. Through a series of experiments, we found that embedded stereotypes can be probed using SDoH input and that LLMs rely on embedded stereotypes to make gendered decisions, suggesting that evaluating interactions among SDoH factors could usefully complement existing approaches to assessing LLM performance and bias.

Generative policies based on diffusion and flow matching achieve strong performance in robotic manipulation by modeling multi-modal human demonstrations. However, their reliance on iterative Ordinary Differential Equation (ODE) integration introduces substantial latency, limiting high-frequency closed-loop control. Recent single-step acceleration methods alleviate this overhead but often exhibit distributional collapse, producing averaged trajectories that fail to execute coherent manipulation strategies. We propose a framework that distills a Conditional Flow Matching (CFM) expert into a fast single-step student via Implicit Maximum Likelihood Estimation (IMLE). A bi-directional Chamfer distance provides a set-level objective that promotes both mode coverage and fidelity, enabling preservation of the teacher multi-modal action distribution in a single forward pass. A unified perception encoder further integrates multi-view RGB, depth, point clouds, and proprioception into a geometry-aware representation. The resulting high-frequency control supports real-time receding-horizon re-planning and improved robustness under dynamic disturbances.

Document Layout Analysis (DLA) is crucial for document artificial intelligence and has recently received increasing attention, resulting in an influx of large-scale public DLA datasets. Existing work often combines data from various domains in recent public DLA datasets to improve the generalization of DLA. However, directly merging these datasets for training often results in suboptimal model performance, as it overlooks the different layout structures inherent to various domains. These variations include different labeling styles, document types, and languages. This paper introduces PromptDLA, a domain-aware Prompter for Document Layout Analysis that effectively leverages descriptive knowledge as cues to integrate domain priors into DLA. The innovative PromptDLA features a unique domain-aware prompter that customizes prompts based on the specific attributes of the data domain. These prompts then serve as cues that direct the DLA toward critical features and structures within the data, enhancing the model's ability to generalize across varied domains. Extensive experiments show that our proposal achieves state-of-the-art performance among DocLayNet, PubLayNet, M6Doc, and D$^4$LA. Our code is available at https://github.com/Zirui00/PromptDLA.

This paper explores potential improvements to the Spatial-Temporal Matching algorithm for matching the GPS trajectories to road networks. While this algorithm is effective, it presents some limitations in computational efficiency and the accuracy of the results, especially in dense environments with relatively high sampling intervals. To address this, the paper proposes four modifications to the original algorithm: a dynamic buffer, an adaptive observation probability, a redesigned temporal scoring function, and a behavioral analysis to account for the historical mobility patterns. The enhancements are assessed using real-world data from the urban area of Milan, and through newly defined evaluation metrics to be applied in the absence of ground truth. The results of the experiment show significant improvements in performance efficiency and path quality across various metrics.

Recent diffusion models increasingly favor Transformer backbones, motivated by the remarkable scalability of fully attentional architectures. Yet the locality bias, parameter efficiency, and hardware friendliness--the attributes that established ConvNets as the efficient vision backbone--have seen limited exploration in modern generative modeling. Here we introduce the fully convolutional diffusion model (FCDM), a model having a backbone similar to ConvNeXt, but designed for conditional diffusion modeling. We find that using only 50% of the FLOPs of DiT-XL/2, FCDM-XL achieves competitive performance with 7$\times$ and 7.5$\times$ fewer training steps at 256$\times$256 and 512$\times$512 resolutions, respectively. Remarkably, FCDM-XL can be trained on a 4-GPU system, highlighting the exceptional training efficiency of our architecture. Our results demonstrate that modern convolutional designs provide a competitive and highly efficient alternative for scaling diffusion models, reviving ConvNeXt as a simple yet powerful building block for efficient generative modeling.

Validating evaluation metrics for NLG typically relies on expensive and time-consuming human annotations, which predominantly exist only for English datasets. We propose \textit{LLM as a Meta-Judge}, a scalable framework that utilizes LLMs to generate synthetic evaluation datasets via controlled semantic degradation of real data, replacing human judgment. We validate our approach using \textit{meta-correlation}, measuring the alignment between metric rankings derived from synthetic data and those from standard human benchmarks. Experiments across Machine Translation, Question Answering, and Summarization demonstrate that synthetic validation serves as a reliable proxy for human judgment, achieving meta-correlations exceeding 0.9 in multilingual QA and proves to be a viable alternative where human judgments are unavailable or too expensive to obtain. Our code and data will become publicly available upon paper acceptance.

Agents must infer action outcomes and select actions that maximize a reward signal indicating how close the goal is to being reached. Supervised learning of reward models could introduce biases inherent to training data, limiting generalization to novel goals and environments. In this paper, we investigate whether well-defined world state representations alone can enable accurate reward prediction across domains. To address this, we introduce StateFactory, a factorized representation method that transforms unstructured observations into a hierarchical object-attribute structure using language models. This structured representation allows rewards to be estimated naturally as the semantic similarity between the current state and the goal state under hierarchical constraint. Overall, the compact representation structure induced by StateFactory enables strong reward generalization capabilities. We evaluate on RewardPrediction, a new benchmark dataset spanning five diverse domains and comprising 2,454 unique action-observation trajectories with step-wise ground-truth rewards. Our method shows promising zero-shot results against both VLWM-critic and LLM-as-a-Judge reward models, achieving 60% and 8% lower EPIC distance, respectively. Furthermore, this superior reward quality successfully translates into improved agent planning performance, yielding success rate gains of +21.64% on AlfWorld and +12.40% on ScienceWorld over reactive system-1 policies and enhancing system-2 agent planning. Project Page: https://statefactory.github.io

Document Image Machine Translation (DIMT) seeks to translate text embedded in document images from one language to another by jointly modeling both textual content and page layout, bridging optical character recognition (OCR) and natural language processing (NLP). The DIMT 2025 Challenge advances research on end-to-end document image translation, a rapidly evolving area within multimodal document understanding. The competition features two tracks, OCR-free and OCR-based, each with two subtasks for small (less than 1B parameters) and large (greater than 1B parameters) models. Participants submit a single unified DIMT system, with the option to incorporate provided OCR transcripts. Running from December 10, 2024 to April 20, 2025, the competition attracted 69 teams and 27 valid submissions in total. Track 1 had 34 teams and 13 valid submissions, while Track 2 had 35 teams and 14 valid submissions. In this report, we present the challenge motivation, dataset construction, task definitions, evaluation protocol, and a summary of results. Our analysis shows that large-model approaches establish a promising new paradigm for translating complex-layout document images and highlight substantial opportunities for future research.

Engine sounds originate from sequential exhaust pressure pulses rather than sustained harmonic oscillations. While neural synthesis methods typically aim to approximate the resulting spectral characteristics, we propose directly modeling the underlying pulse shapes and temporal structure. We present the Pulse-Train-Resonator (PTR) model, a differentiable synthesis architecture that generates engine audio as parameterized pulse trains aligned to engine firing patterns and propagates them through recursive Karplus-Strong resonators simulating exhaust acoustics. The architecture integrates physics-informed inductive biases including harmonic decay, thermodynamic pitch modulation, valve-dynamics envelopes, exhaust system resonances and derived engine operating modes such as throttle operation and deceleration fuel cutoff (DCFO). Validated on three diverse engine types totaling 7.5 hours of audio, PTR achieves a 21% improvement in harmonic reconstruction and a 5.7% reduction in total loss over a harmonic-plus-noise baseline model, while providing interpretable parameters corresponding to physical phenomena. Complete code, model weights, and audio examples are openly available.

Offline-to-online reinforcement learning (RL) offers a promising paradigm for robotics by pre-training policies on safe, offline demonstrations and fine-tuning them via online interaction. However, a fundamental challenge remains: how to safely explore online without deviating from the behavioral support of the offline data? While recent methods leverage conditional variational autoencoders (CVAEs) to bound exploration within a latent space, they inherently suffer from an exploitation gap -- a performance ceiling imposed by the decoder's reconstruction loss. We introduce SPAARS, a curriculum learning framework that initially constrains exploration to the low-dimensional latent manifold for sample-efficient, safe behavioral improvement, then seamlessly transfers control to the raw action space, bypassing the decoder bottleneck. SPAARS has two instantiations: the CVAE-based variant requires only unordered (s,a) pairs and no trajectory segmentation; SPAARS-SUPE pairs SPAARS with OPAL temporal skill pretraining for stronger exploration structure at the cost of requiring trajectory chunks. We prove an upper bound on the exploitation gap using the Performance Difference Lemma, establish that latent-space policy gradients achieve provable variance reduction over raw-space exploration, and show that concurrent behavioral cloning during the latent phase directly controls curriculum transition stability. Empirically, SPAARS-SUPE achieves 0.825 normalized return on kitchen-mixed-v0 versus 0.75 for SUPE, with 5x better sample efficiency; standalone SPAARS achieves 92.7 and 102.9 normalized return on hopper-medium-v2 and walker2d-medium-v2 respectively, surpassing IQL baselines of 66.3 and 78.3 respectively, confirming the utility of the unordered-pair CVAE instantiation.

Modern foundation models provide highly expressive visual representations, yet adapting them to high-resolution medical imaging remains challenging due to limited annotations and weak supervision. Mammography, in particular, is characterized by large images, variable multi-view studies and predominantly breast-level labels, making end-to-end fine-tuning computationally expensive and often impractical. We propose Multiple Instance Learning on Precomputed Features (MIL-PF), a scalable framework that combines frozen foundation encoders with a lightweight MIL head for mammography classification. By precomputing the semantic representations and training only a small task-specific aggregation module (40k parameters), the method enables efficient experimentation and adaptation without retraining large backbones. The architecture explicitly models the global tissue context and the sparse local lesion signals through attention-based aggregation. MIL-PF achieves state-of-the-art classification performance at clinical scale while substantially reducing training complexity. We release the code for full reproducibility.

Variation in spatial categorization across languages is often studied by eliciting human labels for the relations depicted in a set of scenes known as the Topological Relations Picture Series (TRPS). We demonstrate that labels generated by large language models (LLMs) align relatively well with human labels, and show how LLM-generated labels can help to decide which scenes and languages to add to existing spatial data sets. To illustrate our approach we extend the TRPS by adding 42 new scenes, and show that this extension achieves better coverage of the space of possible scenes than two previous extensions of the TRPS. Our results provide a foundation for scaling towards spatial data sets with dozens of languages and hundreds of scenes.

(short version abstract, full in article)High-fidelity flow field reconstruction is important in fluid dynamics, but it is challenged by sparse and spatiotemporally incomplete sensor measurements, as well as failures of pre-deployed measurement points that can invalidate pre-trained reconstruction models. Physics-informed neural networks (PINNs) alleviate dependence on large labeled datasets by incorporating governing physics, yet sensor placement optimization, a key factor in reconstruction accuracy and robustness, remains underexplored. In this study, we propose a PINN with Voronoi-enhanced Sensor Optimization (VSOPINN). VSOPINN enables differentiable soft Voronoi construction for sparse sensor data rasterization, end-to-end fusion of centroidal Voronoi tessellation (CVT) with PINNs for adaptive sensor placement, and unified layout optimization for multi-condition flow reconstruction through a shared encoder-multi-decoder architecture. We validate VSOPINN on three representative problems: lid-driven cavity flow, vascular flow, and annular rotating flow. Results show that VSOPINN significantly improves reconstruction accuracy across different Reynolds numbers, adaptively learns effective sensor layouts, and remains robust under partial sensor failure. The study clarifies the intrinsic relationship between sensor placement and reconstruction precision in PINN-based flow field reconstruction.

Contrastive learning has demonstrated strong performance in attributed hypergraph clustering. Typically, existing methods based on contrastive learning first learn node embeddings and then apply clustering algorithms, such as k-means, to these embeddings to obtain the clustering results.However, these methods lack direct clustering supervision, risking the inclusion of clustering-irrelevant information in the learned graph.To this end, we propose a Contrastive learning approach for Attributed Hypergraph Clustering (CAHC), an end-to-end method that simultaneously learns node embeddings and obtains clustering results. CAHC consists of two main steps: representation learning and cluster assignment learning. The former employs a novel contrastive learning approach that incorporates both node-level and hyperedge-level objectives to generate node embeddings.The latter joint embedding and clustering optimization to refine these embeddings by clustering-oriented guidance and obtains clustering results simultaneously.Extensive experimental results demonstrate that CAHC outperforms baselines on eight datasets.

In recent years, contrastive learning has drawn significant attention as an effective approach to reducing reliance on labeled data. However, existing methods for self-supervised skeleton-based action recognition still face three major limitations: insufficient modeling of view discrepancies, lack of effective adversarial mechanisms, and uncontrollable augmentation perturbations. To tackle these issues, we propose the Multi-view Mini-Max infinite skeleton-data Game Contrastive Learning for skeleton-based action Recognition (M3GCLR), a game-theoretic contrastive framework. First, we establish the Infinite Skeleton-data Game (ISG) model and the ISG equilibrium theorem, and further provide a rigorous proof, enabling mini-max optimization based on multi-view mutual information. Then, we generate normal-extreme data pairs through multi-view rotation augmentation and adopt temporally averaged input as a neutral anchor to achieve structural alignment, thereby explicitly characterizing perturbation strength. Next, leveraging the proposed equilibrium theorem, we construct a strongly adversarial mini-max skeleton-data game to encourage the model to mine richer action-discriminative information. Finally, we introduce the dual-loss equilibrium optimizer to optimize the game equilibrium, allowing the learning process to maximize action-relevant information while minimizing encoding redundancy, and we prove the equivalence between the proposed optimizer and the ISG model. Extensive Experiments show that M3GCLR achieves three-stream 82.1%, 85.8% accuracy on NTU RGB+D 60 (X-Sub, X-View) and 72.3%, 75.0% accuracy on NTU RGB+D 120 (X-Sub, X-Set). On PKU-MMD Part I and II, it attains 89.1%, 45.2% in three-stream respectively, all results matching or outperforming state-of-the-art performance. Ablation studies confirm the effectiveness of each component.

Dataset condensation (DC) learns a compact synthetic dataset that enables models to match the performance of full-data training, prioritising utility over distributional fidelity. While typically explored for computational efficiency, DC also holds promise for healthcare data democratisation, especially when paired with differential privacy, allowing synthetic data to serve as a safe alternative to real records. However, existing DC methods rely on differentiable neural networks, limiting their compatibility with widely used clinical models such as decision trees and Cox regression. We address this gap using a differentially private, zero-order optimisation framework that extends DC to non-differentiable models using only function evaluations. Empirical results across six datasets, including both classification and survival tasks, show that the proposed method produces condensed datasets that preserve model utility while providing effective differential privacy guarantees - enabling model-agnostic data sharing for clinical prediction tasks without exposing sensitive patient information.

Surface roughness in Material Extrusion Additive Manufacturing varies across a part and is difficult to anticipate during process planning because it depends on both printing parameters and local surface inclination, which governs the staircase effect. A data-driven framework is presented to predict the arithmetic mean roughness (Ra) prior to fabrication using process parameters and surface angle. A structured experimental dataset was created using a three-level Box-Behnken design: 87 specimens were printed, each with multiple planar faces spanning different inclination angles, yielding 1566 Ra measurements acquired with a contact profilometer. A multilayer perceptron regressor was trained to capture nonlinear relationships between manufacturing conditions, inclination, and Ra. To mitigate limited experimental data, a conditional generative adversarial network was used to generate additional condition-specific tabular samples, thereby improving predictive performance. Model performance was assessed on a hold-out test set. A web-based decision-support interface was also developed to enable interactive process planning by loading a 3D model, specifying printing parameters, and adjusting the part's orientation. The system computes face-wise inclination from the model geometry and visualizes predicted Ra as an interactive colormap over the surface, enabling rapid identification of regions prone to high roughness and immediate comparison of parameter and orientation choices.

A significant number of anomalous nodes in the real world, such as fake news, noncompliant users, malicious transactions, and malicious posts, severely compromises the health of the graph data ecosystem and urgently requires effective identification and processing. With anomalies that span multiple data domains yet exhibit vast differences in features, cross-domain detection models face severe domain shift issues, which limit their generalizability across all domains. This study identifies and quantitatively analyzes a specific feature mismatch pattern exhibited by domain shift in graph anomaly detection, which we define as the \emph{Anomaly Disassortativity} issue ($\mathcal{AD}$). Based on the modeling of the issue $\mathcal{AD}$, we introduce a novel graph foundation model for anomaly detection. It achieves cross-domain generalization in different graphs, requiring only a single training phase to perform effectively across diverse domains. The experimental findings, based on fourteen diverse real-world graphs, confirm a breakthrough in the model's cross-domain adaptation, achieving a pioneering state-of-the-art (SOTA) level in terms of detection accuracy. In summary, the proposed theory of $\mathcal{AD}$ provides a novel theoretical perspective and a practical route for future research in generalist graph anomaly detection (GGAD). The code is available at https://anonymous.4open.science/r/Anonymization-TA-GGAD/.

Offline reinforcement learning (RL) enables data-efficient and safe policy learning without online exploration, but its performance often degrades under distribution shift. The learned policy may visit out-of-distribution state-action pairs where value estimates and learned dynamics are unreliable. To address policy-induced extrapolation and transition uncertainty in a unified framework, we formulate offline RL as robust policy optimization, treating the transition kernel as a decision variable within an uncertainty set and optimizing the policy against the worst-case dynamics. We propose Robust Regularized Policy Iteration (RRPI), which replaces the intractable max-min bilevel objective with a tractable KL-regularized surrogate and derives an efficient policy iteration procedure based on a robust regularized Bellman operator. We provide theoretical guarantees by showing that the proposed operator is a $γ$-contraction and that iteratively updating the surrogate yields monotonic improvement of the original robust objective with convergence. Experiments on D4RL benchmarks demonstrate that RRPI achieves strong average performance, outperforming recent baselines including percentile-based methods such as PMDB on the majority of environments while remaining competitive on the rest. Moreover, RRPI exhibits robust behavior. The learned $Q$-values decrease in regions with higher epistemic uncertainty, suggesting that the resulting policy avoids unreliable out-of-distribution actions under transition uncertainty.

Retrieval-Augmented Generation (RAG) helps large language models (LLMs) answer knowledge-intensive and time-sensitive questions by conditioning generation on external evidence. However, most RAG systems still retrieve unstructured chunks and rely on one-shot generation, which often yields redundant context, low information density, and brittle multi-hop reasoning. While structured RAG pipelines can improve grounding, they typically require costly and error-prone graph construction or impose rigid entity-centric structures that do not align with the query's reasoning chain. We propose \textsc{TaSR-RAG}, a taxonomy-guided structured reasoning framework for evidence selection. We represent both queries and documents as relational triples, and constrain entity semantics with a lightweight two-level taxonomy to balance generalization and precision. Given a complex question, \textsc{TaSR-RAG} decomposes it into an ordered sequence of triple sub-queries with explicit latent variables, then performs step-wise evidence selection via hybrid triple matching that combines semantic similarity over raw triples with structural consistency over typed triples. By maintaining an explicit entity binding table across steps, \textsc{TaSR-RAG} resolves intermediate variables and reduces entity conflation without explicit graph construction or exhaustive search. Experiments on multiple multi-hop question answering benchmarks show that \textsc{TaSR-RAG} consistently outperforms strong RAG and structured-RAG baselines by up to 14\%, while producing clearer evidence attribution and more faithful reasoning traces.

Large Language Models (LLMs) have achieved strong performance on static reasoning benchmarks, yet their effectiveness as interactive agents operating in adversarial, time-sensitive environments remains poorly understood. Existing evaluations largely treat reasoning as a single-shot capability, overlooking the challenges of opponent-aware decision-making, temporal constraints, and execution under pressure. This paper introduces Strategic Tactical Agent Reasoning (STAR) Benchmark, a multi-agent evaluation framework that assesses LLMs through 1v1 zero-sum competitive interactions, framing reasoning as an iterative, adaptive decision-making process. STAR supports both turn-based and real-time settings, enabling controlled analysis of long-horizon strategic planning and fast-paced tactical execution within a unified environment. Built on a modular architecture with a standardized API and fully implemented execution engine, STAR facilitates reproducible evaluation and flexible task customization. To move beyond binary win-loss outcomes, we introduce a Strategic Evaluation Suite that assesses not only competitive success but also the quality of strategic behavior, such as execution efficiency and outcome stability. Extensive pairwise evaluations reveal a pronounced strategy-execution gap: while reasoning-intensive models dominate turn-based settings, their inference latency often leads to inferior performance in real-time scenarios, where faster instruction-tuned models prevail. These results show that strategic intelligence in interactive environments depends not only on reasoning depth, but also on the ability to translate plans into timely actions, positioning STAR as a principled benchmark for studying this trade-off in competitive, dynamic settings.

System requirement specifications (SyRSs) are central, natural-language (NL) artifacts. Access to real SyRS for research purposes is highly valuable but limited by proprietary restrictions or confidentiality concerns. Generating synthetic SyRSs (SSyRSs) can address this scarcity. Black-box large language models (LLMs) such as ChatGPT offer compelling generation capabilities by providing easy access to NL generation functions without requiring access to real data. However, LLMs suffer from hallucinations and overconfidence, which pose major challenges in their use. We designed an exploratory study to investigate whether, despite these challenges, we can generate realistic SSyRSs with ChatGPT without having access to real SyRSs. Using a systematic approach that leverages prompt patterns, LLM-based quality assessments, and iterative prompt refinements, we generated 300 SSyRSs across 10 industries with ChatGPT. The results were evaluated using cross-model checks and an expert study, with n=87 submitted surveys. 62\% of experts considered the SSyRSs to be realistic. However, in-depth examination revealed contradictory statements and deficiencies. Overall, we were able to generate realistic SSyRSs to a certain extent with ChatGPT, but LLM-based quality assessments cannot fully replace thorough expert evaluations. This paper presents the methodology and results of our study and discusses the key insights we obtained.

Digital audio workstations expose rich effect chains, yet a semantic gap remains between perceptual user intent and low-level signal-processing parameters. We study retrieval-grounded audio effect control, where the output is an editable plugin configuration rather than a finalized waveform. Our focus is Texture Resonance Retrieval (TRR), an audio representation built from Gram matrices of projected mid-level Wav2Vec2 activations. This design preserves texture-relevant co-activation structure. We evaluate TRR on a guitar-effects benchmark with 1,063 candidate presets and 204 queries. The evaluation follows Protocol-A, a cross-validation scheme that prevents train-test leakage. We compare TRR against CLAP and internal retrieval baselines (Wav2Vec-RAG, Text-RAG, FeatureNN-RAG), using min-max normalized metrics grounded in physical DSP parameter ranges. Ablation studies validate TRR's core design choices: projection dimensionality, layer selection, and projection type. A near-duplicate sensitivity analysis confirms that results are robust to trivial knowledge-base matches. TRR achieves the lowest normalized parameter error among evaluated methods. A multiple-stimulus listening study with 26 participants provides complementary perceptual evidence. We interpret these results as benchmark evidence that texture-aware retrieval is useful for editable audio effect control, while broader personalization and real-audio robustness claims remain outside the verified evidence presented here.

We introduce Reward-Zero, a general-purpose implicit reward mechanism that transforms natural-language task descriptions into dense, semantically grounded progress signals for reinforcement learning (RL). Reward-Zero serves as a simple yet sophisticated universal reward function that leverages language embeddings for efficient RL training. By comparing the embedding of a task specification with embeddings derived from an agent's interaction experience, Reward-Zero produces a continuous, semantically aligned sense-of-completion signal. This reward supplements sparse or delayed environmental feedback without requiring task-specific engineering. When integrated into standard RL frameworks, it accelerates exploration, stabilizes training, and enhances generalization across diverse tasks. Empirically, agents trained with Reward-Zero converge faster and achieve higher final success rates than conventional methods such as PPO with common reward-shaping baselines, successfully solving tasks that hand-designed rewards could not in some complex tasks. In addition, we develop a mini benchmark for the evaluation of completion sense during task execution via language embeddings. These results highlight the promise of language-driven implicit reward functions as a practical path toward more sample-efficient, generalizable, and scalable RL for embodied agents. Code will be released after peer review.

In VR interactions with embodied conversational agents, users' emotional intent is often conveyed more by how something is said than by what is said. However, most VR agent pipelines rely on speech-to-text processing, discarding prosodic cues and often producing emotionally incongruent responses despite correct semantics. We propose an emotion-context-aware VR interaction pipeline that treats vocal emotion as explicit dialogue context in an LLM-based conversational agent. A real-time speech emotion recognition model infers users' emotional states from prosody, and the resulting emotion labels are injected into the agent's dialogue context to shape response tone and style. Results from a within-subjects VR study (N=30) show significant improvements in dialogue quality, naturalness, engagement, rapport, and human-likeness, with 93.3% of participants preferring the emotion-aware agent.

Autonomous space operations such as on-orbit servicing and active debris removal demand robust part-level semantic understanding and precise relative navigation of target spacecraft, yet collecting large-scale real data in orbit remains impractical due to cost and access constraints. Existing synthetic datasets, moreover, suffer from limited target diversity, single-modality sensing, and incomplete ground-truth annotations. We present \textbf{SpaceSense-Bench}, a large-scale multi-modal benchmark for spacecraft perception encompassing 136~satellite models with approximately 70~GB of data. Each frame provides time-synchronized 1024$\times$1024 RGB images, millimeter-precision depth maps, and 256-beam LiDAR point clouds, together with dense 7-class part-level semantic labels at both the pixel and point level as well as accurate 6-DoF pose ground truth. The dataset is generated through a high-fidelity space simulation built in Unreal Engine~5 and a fully automated pipeline covering data acquisition, multi-stage quality control, and conversion to mainstream formats. We benchmark five representative tasks (object detection, 2D semantic segmentation, RGB--LiDAR fusion-based 3D point cloud segmentation, monocular depth estimation, and orientation estimation) and identify two key findings: (i)~perceiving small-scale components (\emph{e.g.}, thrusters and omni-antennas) and generalizing to entirely unseen spacecraft in a zero-shot setting remain critical bottlenecks for current methods, and (ii)~scaling up the number of training satellites yields substantial performance gains on novel targets, underscoring the value of large-scale, diverse datasets for space perception research. The dataset, code, and toolkit are publicly available at https://github.com/wuaodi/SpaceSense-Bench.

Multimodal medical image segmentation often faces missing modalities at inference, which induces disagreement among modality experts and makes fusion unstable, particularly on small foreground structures. We propose Consistency Learning of Experts (CLoE), a consistency-driven framework for missing-modality segmentation that preserves strong performance when all modalities are available. CLoE formulates robustness as decision-level expert consistency control and introduces a dual-branch Expert Consistency Learning objective. Modality Expert Consistency enforces global agreement among expert predictions to reduce case-wise drift under partial inputs, while Region Expert Consistency emphasizes agreement on clinically critical foreground regions to avoid background-dominated regularization. We further map consistency scores to modality reliability weights using a lightweight gating network, enabling reliability-aware feature recalibration before fusion. Extensive experiments on BraTS 2020 and MSD Prostate demonstrate that CLoE outperforms state-of-the-art methods in incomplete multimodal segmentation, while exhibiting strong cross-dataset generalization and improving robustness on clinically critical structures.

Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations. Existing methods largely rely on the Linear Representation Hypothesis, assuming behavioral attributes can be manipulated using global linear directions. In practice, however, such linear interventions often behave inconsistently. We question this assumption by analyzing the intrinsic geometry of LLM activation spaces. Measuring geometric distortion via the ratio of geodesic to Euclidean distances, we observe substantial and concept-dependent distortions, indicating that activation spaces are not well-approximated by a globally linear geometry. Motivated by this, we propose "Curveball steering", a nonlinear steering method based on polynomial kernel PCA that performs interventions in a feature space, better respecting the learned activation geometry. Curveball steering consistently outperforms linear PCA-based steering, particularly in regimes exhibiting strong geometric distortion, suggesting that geometry-aware, nonlinear steering provides a principled alternative to global, linear interventions.

We study training algorithms with data following a Gaussian mixture model. For a specific family of such algorithms, we present a non-asymptotic result, connecting the evolution of the model to a surrogate dynamical system, which can be easier to analyze. The proof of our result is based on the celebrated Gordon comparison theorem. Using our theorem, we rigorously prove the validity of the dynamic mean-field (DMF) expressions in the asymptotic scenarios. Moreover, we suggest an iterative refinement scheme to obtain more accurate expressions in non-asymptotic scenarios. We specialize our theory to the analysis of training a perceptron model with a generic first-order (full-batch) algorithm and demonstrate that fluctuation parameters in a non-asymptotic domain emerge in addition to the DMF kernels.

Verbalized confidence, in which LLMs report a numerical certainty score, is widely used to estimate uncertainty in black-box settings, yet the confidence scale itself (typically 0--100) is rarely examined. We show that this design choice is not neutral. Across six LLMs and three datasets, verbalized confidence is heavily discretized, with more than 78% of responses concentrating on just three round-number values. To investigate this phenomenon, we systematically manipulate confidence scales along three dimensions: granularity, boundary placement, and range regularity, and evaluate metacognitive sensitivity using meta-d'. We find that a 0--20 scale consistently improves metacognitive efficiency over the standard 0--100 format, while boundary compression degrades performance and round-number preferences persist even under irregular ranges. These results demonstrate that confidence scale design directly affects the quality of verbalized uncertainty and should be treated as a first-class experimental variable in LLM evaluation.

Large Language Model (LLM) has exhibited strong reasoning ability in text-based contexts across various domains, yet the limitation of context window poses challenges for the model on long-range inference tasks and necessitates a memory storage system. While many current storage approaches have been proposed with episodic notes and graph representations of memory, retrieval methods still primarily rely on predefined workflows or static similarity top-k over embeddings. To address this inflexibility, we introduced a novel tool-augmented autonomous memory retrieval framework (TA-Mem), which contains: (1) a memory extraction LLM agent which is prompted to adaptively chuck an input into sub-context based on semantic correlation, and extract information into structured notes, (2) a multi-indexed memory database designed for different types of query methods including both key-based lookup and similarity-based retrieval, (3) a tool-augmented memory retrieval agent which explores the memory autonomously by selecting appropriate tools provided by the database based on the user input, and decides whether to proceed to the next iteration or finalizing the response after reasoning on the fetched memories. The TA-Mem is evaluated on the LoCoMo dataset, achieving significant performance improvements over existing baseline approaches. In addition, an analysis of tool use across different question types also demonstrates the adaptivity of the proposed method.

Generative Engine Optimization (GEO) aims to improve content visibility in AI-generated responses. However, existing methods measure contribution-how much a document influences a response-rather than citation, the mechanism that actually drives traffic back to creators. Also, these methods apply generic rewriting rules uniformly, failing to diagnose why individual document are not cited. This paper introduces a diagnostic approach to GEO that asks why a document fails to be cited and intervenes accordingly. We develop a unified framework comprising: (1) the first taxonomy of citation failure modes spanning different stages of a citation pipeline; (2) AgentGEO, an agentic system that diagnoses failures using this taxonomy, selects targeted repairs from a corresponding tool library, and iterates until citation is achieved; and (3) a document-centric benchmark evaluating whether optimizations generalize across held-out queries. AgentGEO achieves over 40% relative improvement in citation rates while modifying only 5% of content, compared to 25% for baselines. Our analysis reveals that generic optimization can harm long-tail content and some documents face challenges that optimization alone cannot fully address-findings with implications for equitable visibility in AI-mediated information access.

3D scene reconstruction and novel-view synthesis are fundamental for VR, robotics, and content creation. However, most NeRF and 3D Gaussian Splatting pipelines assume clean inputs and degrade under real noise and artifacts. We therefore propose DenoiseSplat, a feed-forward 3D Gaussian splatting method for noisy multi-view images. We build a large-scale, scene-consistent noisy--clean benchmark on RE10K by injecting Gaussian, Poisson, speckle, and salt-and-pepper noise with controlled intensities. With a lightweight MVSplat-style feed-forward backbone, we train end-to-end using only clean 2D renderings as supervision and no 3D ground truth. On noisy RE10K, DenoiseSplat outperforms vanilla MVSplat and a strong two-stage baseline (IDF + MVSplat) in PSNR/SSIM and LPIPS across noise types and levels.

Reusing and invoking existing code remains costly and unreliable, as most practical tools are embedded in heterogeneous code repositories and lack standardized, executable interfaces. Although large language models (LLMs) and Model Context Protocol (MCP)-based tool invocation frameworks enable natural language task execution, current approaches rely heavily on manual tool curation and standardization, which fundamentally limits scalability. In this paper, we propose ToolRosetta, a unified framework that automatically translates open-source code repositories and APIs into MCP-compatible tools that can be reliably invoked by LLMs. Given a user task, ToolRosetta autonomously plans toolchains, identifies relevant codebases, and converts them into executable MCP services, enabling end-to-end task completion with minimal human intervention. In addition, ToolRosetta incorporates a security inspection layer to mitigate risks inherent in executing arbitrary code. Extensive experiments across diverse scientific domains demonstrate that ToolRosetta can automatically standardize a large number of open-source tools and reduce the human effort required for code reproduction and deployment. Notably, by seamlessly leveraging specialized open-source tools, ToolRosetta-powered agents consistently improve task completion performance compared to commercial LLMs and existing agent systems.

Aggregate outcome variables collected through surveys and administrative records are often subject to systematic measurement error. For instance, in disaster loss databases, county-level losses reported may differ from the true damages due to variations in on-the-ground data collection capacity, reporting practices, and event characteristics. Such miscalibration complicates downstream analysis and decision-making. We study the problem of outcome miscalibration and propose a framework guided by proxy variables for estimating and correcting the systematic errors. We model the data-generating process using a causal graph that separates latent content variables driving the true outcome from the latent bias variables that induce systematic errors. The key insight is that proxy variables that depend on the true outcome but are independent of the bias mechanism provide identifying information for quantifying the bias. Leveraging this structure, we introduce a two-stage approach that utilizes variational autoencoders to disentangle content and bias latents, enabling us to estimate the effect of bias on the outcome of interest. We analyze the assumptions underlying our approach and evaluate it on synthetic data, semi-synthetic datasets derived from randomized trials, and a real-world case study of disaster loss reporting.

We study a widely used Bayesian optimization method, Gaussian process Thompson sampling (GP-TS), under the assumption that the objective function is a sample path from a GP. Compared with the GP upper confidence bound (GP-UCB) with established high-probability and expected regret bounds, most analyses of GP-TS have been limited to expected regret. Moreover, whether the recent analyses of GP-UCB for the lenient regret and the improved cumulative regret upper bound can be applied to GP-TS remains unclear. To fill these gaps, this paper shows several regret bounds: (i) a regret lower bound for GP-TS, which implies that GP-TS suffers from a polynomial dependence on $1/δ$ with probability $δ$, (ii) an upper bound of the second moment of cumulative regret, which directly suggests an improved regret upper bound on $δ$, (iii) expected lenient regret upper bounds, and (iv) an improved cumulative regret upper bound on the time horizon $T$. Along the way, we provide several useful lemmas, including a relaxation of the necessary condition from recent analysis to obtain improved regret upper bounds on $T$.

Spatiotemporal information is at the core of diverse sensory processing and computational tasks. Feed-forward spiking neural networks can be used to solve these tasks while offering potential benefits in terms of energy efficiency by computing event-based. However, they have trouble decoding temporal information with high accuracy. Thus, they commonly resort to recurrence or delays to enhance their temporal computing ability which, however, bring downsides in terms of hardware-efficiency. In the brain, dendrites are computational powerhouses that just recently started to be acknowledged in such machine learning systems. In this work, we focus on a sequence detection mechanism present in branches of dendrites and translate it into a novel type of neural network by introducing a dendrocentric neural network, DendroNN. DendroNNs identify unique incoming spike sequences as spatiotemporal features. This work further introduces a rewiring phase to train the non-differentiable spike sequences without the use of gradients. During the rewiring, the network memorizes frequently occurring sequences and additionally discards those that do not contribute any discriminative information. The networks display competitive accuracies across various event-based time series datasets. We also propose an asynchronous digital hardware architecture using a time-wheel mechanism that builds on the event-driven design of DendroNNs, eliminating per-step global updates typical of delay- or recurrence-based models. By leveraging a DendroNN's dynamic and static sparsity along with intrinsic quantization, it achieves up to 4x higher efficiency than state-of-the-art neuromorphic hardware at comparable accuracy on the same audio classification task, demonstrating its suitability for spatiotemporal event-based computing. This work offers a novel approach to low-power spatiotemporal processing on event-driven hardware.

The discovery and design of functional molecules remain central challenges across chemistry,biology, and materials science. While recent advances in machine learning have accelerated molecular property prediction and candidate generation, existing models tend to excel either in physical fidelity without transparent reasoning, or in flexible reasoning without guarantees of chemical validity. This imbalance limits the reliability of artificial intelligence systems in real scientific design workflows.Here we present Logos, a compact molecular reasoning model that integrates multi-step logical reasoning with strict chemical consistency. Logos is trained using a staged strategy that first exposes the model to explicit reasoning examples linking molecular descriptions to structural decisions, and then progressively aligns these reasoning patterns with molecular representations. In a final training phase, chemical rules and invariants are incorporated directly into the optimization objective, guiding the model toward chemically valid outputs. Across multiple benchmark datasets, Logos achieves strong performance in both structural accuracy and chemical validity, matching or surpassing substantially larger general-purpose language models while operating with a fraction of their parameters. Beyond benchmark evaluation, the model exhibits stable behaviour in molecular optimization tasks involving multiple, potentially conflicting constraints. By explicitly exposing intermediate reasoning steps, Logos enables human inspection and assessment of the design logic underlying each generated structure. These results indicate that jointly optimizing for reasoning structure and physical consistency offers a practical pathway toward reliable and interpretable AI systems for molecular science, supporting closer integration of artificial intelligence into scientific discovery processes.

Many existing transductive bounds rely on classical complexity measures that are computationally intractable and often misaligned with empirical behavior. In this work, we establish new representation-based generalization bounds in a distribution-free transductive setting, where learned representations are dependent, and test features are accessible during training. We derive global and class-wise bounds via optimal transport, expressed in terms of Wasserstein distances between encoded feature distributions. We demonstrate that our bounds are efficiently computable and strongly correlate with empirical generalization in graph node classification, improving upon classical complexity measures. Additionally, our analysis reveals how the GNN aggregation process transforms the representation distributions, inducing a trade-off between intra-class concentration and inter-class separation. This yields depth-dependent characterizations that capture the non-monotonic relationship between depth and generalization error observed in practice. The code is available at https://github.com/ml-postech/Transductive-OT-Gen-Bound.

Deep learning and computer vision techniques have become increasingly important in the development of self-driving cars. These techniques play a crucial role in enabling self-driving cars to perceive and understand their surroundings, allowing them to safely navigate and make decisions in real-time. Using Neural Networks self-driving cars can accurately identify and classify objects such as pedestrians, other vehicles, and traffic signals. Using deep learning and analyzing data from sensors such as cameras and radar, self-driving cars can predict the likely movement of other objects and plan their own actions accordingly. In this study, a novel approach to enhance the performance of selfdriving cars by using pre-trained and custom-made neural networks for key tasks, including traffic sign classification, vehicle detection, lane detection, and behavioral cloning is provided. The methodology integrates several innovative techniques, such as geometric and color transformations for data augmentation, image normalization, and transfer learning for feature extraction. These techniques are applied to diverse datasets,including the German Traffic Sign Recognition Benchmark (GTSRB), road and lane segmentation datasets, vehicle detection datasets, and data collected using the Udacity selfdriving car simulator to evaluate the model efficacy. The primary objective of the work is to review the state-of-the-art in deep learning and computer vision for self-driving cars. The findings of the work are effective in solving various challenges related to self-driving cars like traffic sign classification, lane prediction, vehicle detection, and behavioral cloning, and provide valuable insights into improving the robustness and reliability of autonomous systems, paving the way for future research and deployment of safer and more efficient self-driving technologies.

We study efficient reasoning under tight compute. We ask how to make structured, correct decisions without increasing test time cost. We add two training only components to small and medium Transformers that also transfer to broader differentiable optimizers. First, a length aware attention prior built via fuzzy regime position alignment, RPA, yields a normalized pre softmax bias that guides attention like a structured regularizer while adding no new inference parameters. Second, a minimal gain aware controller, Guardian, nudges attention sharpness only when validation improvements warrant it, following a two timescale policy gradient view of nonconvex optimization. It is disabled at inference. A KL perspective shows softmax of z plus log pi as MAP with KL regularization, grounding the prior in a principled objective. Under strict compute parity on WikiText 2, we reduce validation cross entropy while matching baseline latency and memory. At inference, we add a precomputed, cached prior B of T as a single additive bias per head. The controller does not run. In practice, this incurs negligible overhead, a cached bias add per head, with no measurable p50 latency shift. Our results suggest that length aware priors and late phase gain control preserve scarce improvements, especially in long span, noisy logit regimes, while keeping test time costs effectively unchanged.

Learning to sample from complex unnormalized distributions is a fundamental challenge in computational physics and machine learning. While score-based and variational methods have achieved success in continuous domains, extending them to discrete or mixed-variable systems remains difficult due to ill-defined gradients or high variance in estimators. We propose a unified, target-gradient-free generative sampling framework applicable across diverse state spaces. Building on the fact that detailed balance implies the time-reversibility of the equilibrium stochastic process, we enforce this symmetry as a statistical constraint. Specifically, using a prescribed physical transition kernel (such as Metropolis-Hastings), we minimize the Maximum Mean Discrepancy (MMD) between the joint distributions of forward and backward Markov trajectories. Crucially, this training procedure relies solely on energy evaluations via acceptance ratios, circumventing the need for target score functions or continuous relaxations. We demonstrate the versatility of our method on three distinct benchmarks: (1) a continuous multi-modal Gaussian mixture, (2) the discrete high-dimensional Ising model, and (3) a challenging hybrid system coupling discrete indices with continuous dynamics. Experiments show that our framework accurately reproduces thermodynamic observables and captures mode-switching behavior across all regimes, offering a physically grounded and universally applicable alternative for equilibrium sampling.

While large language models demonstrate remarkable capabilities across numerous domains, social intelligence - the capacity to perceive social cues, infer mental states, and generate appropriate responses - remains a critical challenge, particularly for enabling effective human-AI collaboration and developing AI that truly serves human needs. Current models often rely on superficial patterns rather than genuine social reasoning. We argue that cultivating human-like social intelligence requires training with challenging cases that resist shortcut solutions. To this end, we introduce ToMBench-Hard, an adversarial benchmark designed to provide hard training examples for social reasoning. Building on this, we propose Social-R1, a reinforcement learning framework that aligns model reasoning with human cognition through multi-dimensional rewards. Unlike outcome-based RL, Social-R1 supervises the entire reasoning process, enforcing structural alignment, logical integrity, and information density. Results show that our approach enables a 4B parameter model to surpass much larger counterparts and generalize robustly across eight diverse benchmarks. These findings demonstrate that challenging training cases with trajectory-level alignment offer a path toward efficient and reliable social intelligence.

Virtual try-off (VTOFF) aims to recover canonical flat-garment representations from images of dressed persons for standardized display and downstream virtual try-on. Prior methods often treat VTOFF as direct image translation driven by local masks or text-only prompts, overlooking the gap between on-body appearances and flat layouts. This gap frequently leads to inconsistent completion in unobserved regions and unstable garment structure. We propose BridgeDiff, a diffusion-based framework that explicitly bridges human-centric observations and flat-garment synthesis through two complementary components. First, the Garment Condition Bridge Module (GCBM) builds a garment-cue representation that captures global appearance and semantic identity, enabling robust inference of continuous details under partial visibility. Second, the Flat Structure Constraint Module (FSCM) injects explicit flat-garment structural priors via Flat-Constraint Attention (FC-Attention) at selected denoising stages, improving structural stability beyond text-only conditioning. Extensive experiments on standard VTOFF benchmarks show that BridgeDiff achieves state-of-the-art performance, producing higher-quality flat-garment reconstructions while preserving fine-grained appearance and structural integrity.

While Contrastive Decoding (CD) has proven effective at enhancing Large Audio Language Models (LALMs), the underlying mechanisms driving its success and the comparative efficacy of different strategies remain unclear. This study systematically evaluates four distinct CD strategies across diverse LALM architectures. We identify Audio-Aware Decoding and Audio Contrastive Decoding as the most effective methods. However, their impact varies significantly by model. To explain this variability, we introduce a Transition Matrix framework to map error pattern shifts during inference. Our analysis demonstrates that CD reliably rectifies errors in which models falsely claim an absence of audio or resort to uncertainty-driven guessing. Conversely, it fails to correct flawed reasoning or confident misassertions. Ultimately, these findings provide a clear guideline for determining which LALM architectures are most suitable for CD enhancement based on their baseline error profiles.

Large language models (LLMs) demonstrate exceptional performance on general-purpose tasks. however, transferring them to complex engineering domains such as space situational awareness (SSA) remains challenging owing to insufficient structural alignment with mission chains, the absence of higher-order cognitive supervision, and poor correspondence between data quality criteria and engineering specifications. The core bottleneck is the construction of high-quality supervised fine-tuning (SFT) datasets. To this end, we propose BD-FDG (Bloom's Taxonomy-based Domain-specific Fine-tuning Data Generation), a framework that addresses incomplete knowledge coverage, shallow cognitive depth, and limited quality controllability through three mechanisms: structured knowledge organization, cognitively layered question modeling, and automated quality control. The framework uses a knowledge tree to ensure structured corpus coverage, designs a question generation scheme spanning nine categories and six cognitive levels from Remember to Create to produce samples with a continuous difficulty gradient, and applies a multidimensional scoring pipeline to enforce domain rigor and consistency. Using BD-FDG, we construct SSA-SFT, a domain dataset of approximately 230K samples, and fine-tune Qwen3-8B to obtain SSA-LLM-8B. Experiments show that SSA-LLM-8B achieves relative BLEU-1 improvements of 144\% (no-think) and 176\% (think) on the domain test set and a win rate of 82.21\% over the baseline in arena comparisons, while largely preserving general benchmark performance (MMLU-Pro, MATH-500). These results validate SFT data construction driven by cognitive layering as an effective paradigm for complex engineering domains and provide a transferable framework for domain-specific LLM adaptation.

Efficient context compression is crucial for improving the accuracy and scalability of question answering. For the efficiency of Retrieval Augmented Generation, context should be delivered fast, compact, and precise to ensure clue sufficiency and budget-friendly LLM reader cost. We propose a margin-based framework for query-driven context pruning, which identifies sentences that are critical for answering a query by measuring changes in clue richness when they are omitted. The model is trained with a composite ranking loss that enforces large margins for critical sentences while keeping non-critical ones near neutral. Built on a lightweight encoder-only Transformer, our approach generally achieves strong exact-match and F1 scores with high-throughput inference and lower memory requirements than those of major baselines. In addition to efficiency, our method yields effective compression ratios without degrading answering performance, demonstrating its potential as a lightweight and practical alternative for retrieval-augmented tasks.

Associative memory has long underpinned the design of sequential models. Beyond recall, humans reason by projecting future states and selecting goal-directed actions, a capability that modern language models increasingly require but do not natively encode. While prior work uses reinforcement learning or test-time training, planning remains external to the model architecture. We formulate reasoning as optimal control and introduce the Test-Time Control (TTC) layer, which performs finite-horizon LQR planning over latent states at inference time, represents a value function within neural architectures, and leverages it as the nested objective to enable planning before prediction. To ensure scalability, we derive a hardware-efficient LQR solver based on a symplectic formulation and implement it as a fused CUDA kernel, enabling parallel execution with minimal overhead. Integrated as an adapter into pretrained LLMs, TTC layers improve mathematical reasoning performance by up to +27.8% on MATH-500 and 2-3x Pass@8 improvements on AMC and AIME, demonstrating that embedding optimal control as an architectural component provides an effective and scalable mechanism for reasoning beyond test-time training.

Physical interactive robotics, ranging from wearable devices to collaborative humanoid robots, require close coordination between mechanical design and control. However, evaluating interactive dynamics is challenging due to complex human biomechanics and motor responses. Traditional experiments rely on indirect metrics without measuring human internal states, such as muscle forces or joint loads. To address this issue, we develop a scalable simulation-based framework for the quantitative analysis of physical human-robot interaction. At its core is a full-body musculoskeletal model serving as a predictive surrogate for the human dynamical system. Driven by a reinforcement learning controller, it generates adaptive, physiologically grounded motor behaviors. We employ a sequential training pipeline where the pre-trained human motion control policy acts as a consistent evaluator, making large-scale design space exploration computationally tractable. By simulating the coupled human-robot system, the framework provides access to internal biomechanical metrics, offering a systematic way to concurrently co-optimize a robot's structural parameters and control policy. We demonstrate its capability in optimizing human-exoskeleton interactions, showing improved joint alignment and reduced contact forces. This work establishes embodied human simulation as a scalable paradigm for interactive robotics design.

Interleaved spoken language models (SLMs) alternately generate text and speech tokens, but decoding at full transformer depth for every step becomes costly, especially due to long speech sequences. We propose SPAR-K, a modality-aware early exit framework designed to accelerate interleaved SLM inference while preserving perceptual quality. SPAR-K introduces a speech alternating-depth schedule: most speech positions exit at a fixed intermediate layer, while periodic full-depth "refresh" steps mitigate distribution shift due to early exit. We evaluate our framework using Step-Audio-2-mini and GLM-4-Voice across four datasets spanning reasoning, factual QA, and dialogue tasks, measuring performance in terms of ASR transcription accuracy and perceptual quality. Experimental results demonstrate that SPAR-K largely preserves question-answering accuracy with a maximum accuracy drop of 0.82\% while reducing average speech decoding depth by up to 11\% on Step-Audio-2-mini and 5\% on GLM-4-Voice, both with negligible changes in MOS and WER and no auxiliary computation overhead. We further demonstrate that confidence-based early exit strategies, widely used in text LLMs, are suboptimal for SLMs, highlighting that the unique statistical nature of speech tokens necessitates a specialized early exit design.

End-users seldom read verbose privacy policies, leading app stores like Google Play to mandate simplified data safety declarations as a user-friendly alternative. However, these self-declared disclosures often contradict the full privacy policies, deceiving users about actual data practices and violating regulatory requirements for consistency. To address this, we introduce PrivPRISM, a robust framework that combines encoder and decoder language models to systematically extract and compare fine-grained data practices from privacy policies and to compare against data safety declarations, enabling scalable detection of non-compliance. Evaluating 7,770 popular mobile games uncovers discrepancies in nearly 53% of cases, rising to 61% among 1,711 widely used generic apps. Additionally, static code analysis reveals possible under-disclosures, with privacy policies disclosing just 66.8% of potential accesses to sensitive data like location and financial information, versus only 36.4% in data safety declarations of mobile games. Our findings expose systemic issues, including widespread reuse of generic privacy policies, vague / contradictory statements, and hidden risks in high-profile apps with 100M+ downloads, underscoring the urgent need for automated enforcement to protect platform integrity and for end-users to be vigilant about sensitive data they disclose via popular apps.

We formalize a macro-financial stress test for rapid AI adoption. Rather than a productivity bust or existential risk, we identify a distribution-and-contract mismatch: AI-generated abundance coexists with demand deficiency because economic institutions are anchored to human cognitive scarcity. Three mechanisms formalize this channel. First, a displacement spiral with competing reinstatement effects: each firm's rational decision to substitute AI for labor reduces aggregate labor income, which reduces aggregate demand, accelerating further AI adoption. We derive conditions on the AI capability growth rate, diffusion speed, and reinstatement rate under which the net feedback is self-limiting versus explosive. Second, Ghost GDP: when AI-generated output substitutes for labor-generated output, monetary velocity declines monotonically in the labor share absent compensating transfers, creating a wedge between measured output and consumption-relevant income. Third, intermediation collapse: AI agents that reduce information frictions compress intermediary margins toward pure logistics costs, triggering repricing across SaaS, payments, consulting, insurance, and financial advisory. Because top-quintile earners drive 47--65\% of U.S.\ consumption and face the highest AI exposure, the transmission into private credit (\$2.5 trillion globally) and mortgage markets (\$13 trillion) is disproportionate. We derive eleven testable predictions with explicit falsification conditions. Calibrated simulations disciplined by FRED time series and BLS occupation-level data quantify conditions under which stable adjustment transitions to explosive crisis.

Provably efficient and robust equilibrium computation in general-sum Markov games remains a core challenge in multi-agent reinforcement learning. Nash equilibrium is computationally intractable in general and brittle due to equilibrium multiplicity and sensitivity to approximation error. We study Risk-Sensitive Quantal Response Equilibrium (RQRE), which yields a unique, smooth solution under bounded rationality and risk sensitivity. We propose \texttt{RQRE-OVI}, an optimistic value iteration algorithm for computing RQRE with linear function approximation in large or continuous state spaces. Through finite-sample regret analysis, we establish convergence and explicitly characterize how sample complexity scales with rationality and risk-sensitivity parameters. The regret bounds reveal a quantitative tradeoff: increasing rationality tightens regret, while risk sensitivity induces regularization that enhances stability and robustness. This exposes a Pareto frontier between expected performance and robustness, with Nash recovered in the limit of perfect rationality and risk neutrality. We further show that the RQRE policy map is Lipschitz continuous in estimated payoffs, unlike Nash, and RQRE admits a distributionally robust optimization interpretation. Empirically, we demonstrate that \texttt{RQRE-OVI} achieves competitive performance under self-play while producing substantially more robust behavior under cross-play compared to Nash-based approaches. These results suggest \texttt{RQRE-OVI} offers a principled, scalable, and tunable path for equilibrium learning with improved robustness and generalization.

Self-evolving has emerged as a key paradigm for improving foundational models such as Large Language Models (LLMs) and Vision Language Models (VLMs) with minimal human intervention. While recent approaches have demonstrated that LLM agents can self-evolve from scratch with little to no data, VLMs introduce an additional visual modality that typically requires at least some seed data, such as images, to bootstrap the self-evolution process. In this work, we present Multi-model Multimodal Zero (MM-Zero), the first RL-based framework to achieve zero-data self-evolution for VLM reasoning. Moving beyond prior dual-role (Proposer and Solver) setups, MM-Zero introduces a multi-role self-evolving training framework comprising three specialized roles: a Proposer that generates abstract visual concepts and formulates questions; a Coder that translates these concepts into executable code (e.g., Python, SVG) to render visual images; and a Solver that performs multimodal reasoning over the generated visual content. All three roles are initialized from the same base model and trained using Group Relative Policy Optimization (GRPO), with carefully designed reward mechanisms that integrate execution feedback, visual verification, and difficulty balancing. Our experiments show that MM-Zero improves VLM reasoning performance across a wide range of multimodal benchmarks. MM-Zero establishes a scalable path toward self-evolving multi-model systems for multimodal models, extending the frontier of self-improvement beyond the conventional two-model paradigm.

Large language models are routinely deployed on text that varies widely in emotional tone, yet their reasoning behavior is typically evaluated without accounting for emotion as a source of representational variation. Prior work has largely treated emotion as a prediction target, for example in sentiment analysis or emotion classification. In contrast, we study emotion as a latent factor that shapes how models attend to and reason over text. We analyze how emotional tone systematically alters attention geometry in transformer models, showing that metrics such as locality, center-of-mass distance, and entropy vary across emotions and correlate with downstream question-answering performance. To facilitate controlled study of these effects, we introduce Affect-Uniform ReAding QA (AURA-QA), a question-answering dataset with emotionally balanced, human-authored context passages. Finally, an emotional regularization framework is proposed that constrains emotion-conditioned representational drift during training. Experiments across multiple QA benchmarks demonstrate that this approach improves reading comprehension in both emotionally-varying and non-emotionally varying datasets, yielding consistent gains under distribution shift and in-domain improvements on several benchmarks.

Retrieval-augmented agents can query external evidence, yet their reliability in multi-step reasoning remains limited: noisy retrieval may derail multi-hop question answering, while outcome-only reinforcement learning provides credit signals that are too coarse to optimize intermediate steps. We propose \textsc{EvalAct} (Evaluate-as-Action), which converts implicit retrieval quality assessment into an explicit action and enforces a coupled Search-to-Evaluate protocol so that each retrieval is immediately followed by a structured evaluation score, yielding process signals aligned with the interaction trajectory. To leverage these signals, we introduce Process-Calibrated Advantage Rescaling (PCAR), a GRPO-based optimization method that rescales advantages at the segment level according to evaluation scores, emphasizing reliable segments while updating uncertain ones conservatively. Experiments on seven open-domain QA benchmarks show that \textsc{EvalAct} achieves the best average accuracy, with the largest gains on multi-hop tasks, and ablations verify that the explicit evaluation loop drives the primary improvements while PCAR provides consistent additional benefits.

We study a problem of signal separation: estimating a signal of interest (SOI) contaminated by an unknown non-Gaussian background/interference. Given the training data consisting of examples of SOI and interference, we show how to build a fully data-driven signal separator. To that end we learn a good discrete tokenizer for SOI and then train an end-to-end transformer on a cross-entropy loss. Training with a cross-entropy shows substantial improvements over the conventional mean-squared error (MSE). Our tokenizer is a modification of Google's SoundStream, which incorporates additional transformer layers and switches from VQVAE to finite-scalar quantization (FSQ). Across real and synthetic mixtures from the MIT RF Challenge dataset, our method achieves competitive performance, including a 122x reduction in bit-error rate (BER) over prior state-of-the-art techniques for separating a QPSK signal from 5G interference. The learned representation adapts to the interference type without side information and shows zero-shot generalization to unseen mixtures at inference time, underscoring its potential beyond RF. Although we instantiate our approach on radio-frequency mixtures, we expect the same architecture to apply to gravitational-wave data (e.g., LIGO strain) and other scientific sensing problems that require data-driven modeling of background and noise.

Situational awareness, the capacity of an AI system to recognize its own nature, understand its training and deployment context, and reason strategically about its circumstances, is widely considered among the most dangerous emergent capabilities in advanced AI systems. Separately, a growing research effort seeks to improve the logical reasoning capabilities of large language models (LLMs) across deduction, induction, and abduction. In this paper, we argue that these two research trajectories are on a collision course. We introduce the RAISE framework (Reasoning Advancing Into Self Examination), which identifies three mechanistic pathways through which improvements in logical reasoning enable progressively deeper levels of situational awareness: deductive self inference, inductive context recognition, and abductive self modeling. We formalize each pathway, construct an escalation ladder from basic self recognition to strategic deception, and demonstrate that every major research topic in LLM logical reasoning maps directly onto a specific amplifier of situational awareness. We further analyze why current safety measures are insufficient to prevent this escalation. We conclude by proposing concrete safeguards, including a "Mirror Test" benchmark and a Reasoning Safety Parity Principle, and pose an uncomfortable but necessary question to the logical reasoning community about its responsibility in this trajectory.

Message Passing Graph Neural Networks (MP-GNNs) have garnered attention for addressing various industry challenges, such as user recommendation and fraud detection. However, they face two major hurdles: (1) heavy reliance on local context, often lacking information about the global context or graph-level features, and (2) assumption of strong homophily among connected nodes, struggling with noisy local neighborhoods. To tackle these, we introduce $P^2$GNN, a plug-and-play technique leveraging prototypes to optimize message passing, enhancing the performance of the base GNN model. Our approach views the prototypes in two ways: (1) as universally accessible neighbors for all nodes, enriching global context, and (2) aligning messages to clustered prototypes, offering a denoising effect. We demonstrate the extensibility of our proposed method to all message-passing GNNs and conduct extensive experiments across 18 datasets, including proprietary e-commerce datasets and open-source datasets, on node recommendation and node classification tasks. Results show that $P^2$GNN outperforms production models in e-commerce and achieves the top average rank on open-source datasets, establishing it as a leading approach. Qualitative analysis supports the value of global context and noise mitigation in the local neighborhood in enhancing performance.

Retrieval-augmented generation (RAG) improves factual grounding, yet most systems rely on flat chunk retrieval and provide limited control over multi-step synthesis. We propose an Explainable Innovation Engine that upgrades the knowledge unit from text chunks to methods-as-nodes. The engine maintains a weighted method provenance tree for traceable derivations and a hierarchical clustering abstraction tree for efficient top-down navigation. At inference time, a strategy agent selects explicit synthesis operators (e.g., induction, deduction, analogy), composes new method nodes, and records an auditable trajectory. A verifier-scorer layer then prunes low-quality candidates and writes validated nodes back to support continual growth. Expert evaluation across six domains and multiple backbones shows consistent gains over a vanilla baseline, with the largest improvements on derivation-heavy settings, and ablations confirm the complementary roles of provenance backtracking and pruning. These results suggest a practical path toward controllable, explainable, and verifiable innovation in agentic RAG systems. Code is available at the project GitHub repository https://github.com/xiaolu-666113/Dual-Tree-Agent-RAG.

Music source separation is the task of isolating the instrumental tracks from a music song. Despite its spectacular recent progress, the trend towards more complex architectures and training protocols exacerbates reproducibility issues. The band-split recurrent neural networks (BSRNN) model is promising in this regard, since it yields close to state-of-the-art results on public datasets, and requires reasonable resources for training. Unfortunately, it is not straightforward to reproduce since its full code is not available. In this paper, we attempt to replicate BSRNN as closely as possible to the original paper through extensive experiments, which allows us to conduct a critical reflection on this reproducibility issue. Our contributions are three-fold. First, this study yields several insights on the model design and training pipeline, which sheds light on potential future improvements. In particular, since we were unsuccessful in reproducing the original results, we explore additional variants that ultimately yield an optimized BSRNN model, whose performance largely improves that of the original. Second, we discuss reproducibility issues from both methodological and practical perspectives. We notably underline how substantial time and energy costs could have been saved upon availability of the full pipeline. Third, our code and pre-trained models are released publicly to foster reproducible research. We hope that this study will contribute to spread awareness on the importance of reproducible research in the music separation community, and help promoting more transparent and sustainable practices.

Recent advances in Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) have enabled diverse retrieval methods. However, existing retrieval methods often fail to accurately retrieve results for negation and exclusion queries. To address this limitation, prior approaches rely on embedding adaptation or fine-tuning, which introduce additional computational cost and deployment complexity. We propose Direct Embedding Optimization (DEO), a training-free method for negation-aware text and multimodal retrieval. DEO decomposes queries into positive and negative components and optimizes the query embedding with a contrastive objective. Without additional training data or model updates, DEO outperforms baselines on NegConstraint, with gains of +0.0738 nDCG@10 and +0.1028 MAP@100, while improving Recall@5 by +6\% over OpenAI CLIP in multimodal retrieval. These results demonstrate the practicality of DEO for negation- and exclusion-aware retrieval in real-world settings.

Most multi-agent systems rely exclusively on autoregressive language models (ARMs) that are based on sequential generation. Although effective for fluent text, ARMs limit global reasoning and plan revision. On the other hand, Discrete Diffusion Language Models (DDLMs) enable non-sequential, globally revisable generation and have shown strong planning capabilities, but their limited text fluency hinders direct collaboration with ARMs. We introduce Latent-DARM, a latent-space communication framework bridging DDLM (planners) and ARM (executors), maximizing collaborative benefits. Across mathematical, scientific, and commonsense reasoning benchmarks, Latent-DARM outperforms text-based interfaces on average, improving accuracy from 27.0% to 36.0% on DART-5 and from 0.0% to 14.0% on AIME2024. Latent-DARM approaches the results of state-of-the-art reasoning models while using less than 2.2% of its token budget. This work advances multi-agent collaboration among agents with heterogeneous models.

Spoken dialog systems with cascaded ASR-LLM-TTS modules retain strong LLM intelligence, but VAD segmentation often forces half-duplex turns and brittle control. On the other hand, VAD-free end-to-end model support full-duplex interaction but is hard to maintain conversational intelligence. In this paper, we present DuplexCascade, a VAD-free cascaded streaming pipeline for full-duplex speech-to-speech dialogue. Our key idea is to convert conventional utterance-wise long turns into chunk-wise micro-turn interactions, enabling rapid bidirectional exchange while preserving the strengths of a capable text LLM. To reliably coordinate turn-taking and response timing, we introduce a set of conversational special control tokens that steer the LLM's behavior under streaming constraints. On Full-DuplexBench and VoiceBench, DuplexCascade delivers state-of-the-art full-duplex turn-taking and strong conversational intelligence among open-source speech-to-speech dialogue systems.

Macroscopic traffic flow is stochastic, but the physics-informed deep learning methods currently used in transportation literature embed deterministic PDEs and produce point-valued outputs; the stochasticity of the governing dynamics plays no role in the learned representation. This work develops a framework in which the physics constraint itself is distributional and directly derived from stochastic traffic-flow dynamics. Starting from an Ito-type Lighthill-Whitham-Richards model with Brownian forcing, we derive a one-point forward equation for the marginal traffic density at each spatial location. The spatial coupling induced by the conservation law appears as an explicit conditional drift term, which makes the closure requirement transparent. Based on this formulation, we derive an equivalent deterministic Probability Flow ODE that is pointwise evaluable and differentiable once a closure is specified. Incorporating this as a physics constraint, we then propose a score network with an advection-closure module, trainable by denoising score matching together with a Fokker-Planck residual loss. The resulting model targets a data-conditioned density distribution, from which point estimates, credible intervals, and congestion-risk measures can be computed. The framework provides a basis for distributional traffic-state estimation and for stochastic fundamental-diagram analysis in a physics-informed generative setting.

We present improved lower bounds for five classical Ramsey numbers: $\mathbf{R}(3, 13)$ is increased from $60$ to $61$, $\mathbf{R}(3, 18)$ from $99$ to $100$, $\mathbf{R}(4, 13)$ from $138$ to $139$, $\mathbf{R}(4, 14)$ from $147$ to $148$, and $\mathbf{R}(4, 15)$ from $158$ to $159$. These results were achieved using~\emph{AlphaEvolve}, an LLM-based code mutation agent. Beyond these new results, we successfully recovered lower bounds for all Ramsey numbers known to be exact, and matched the best known lower bounds across many other cases. These include bounds for which previous work does not detail the algorithms used. Virtually all known Ramsey lower bounds are derived computationally, with bespoke search algorithms each delivering a handful of results. AlphaEvolve is a single meta-algorithm yielding search algorithms for all of our results.

Achieving versatile and naturalistic whole-body control for humanoid robot scene-interaction remains a significant challenge. While some recent works have demonstrated autonomous humanoid interactive control, they are constrained to rigid locomotion patterns and expensive teleoperation data collection, lacking the versatility to execute more human-like natural behaviors such as sitting or kicking. Furthermore, acquiring the necessary real robot teleoperation data is prohibitively expensive and time-consuming. To address these limitations, we introduce ZeroWBC, a novel framework that learns a natural humanoid visuomotor control policy directly from human egocentric videos, eliminating the need for large-scale robot teleoperation data and enabling natural humanoid robot scene-interaction control. Specifically, our approach first fine-tunes a Vision-Language Model (VLM) to predict future whole-body human motions based on text instructions and egocentric visual context, then these generated motions are retargeted to real robot joints and executed via our robust general motion tracking policy for humanoid whole-body control. Extensive experiments on the Unitree G1 humanoid robot demonstrate that our method outperforms baseline approaches in motion naturalness and versatility, successfully establishing a pipeline that eliminates teleoperation data collection overhead for whole-body humanoid control, offering a scalable and efficient paradigm for general humanoid whole-body control.

In this paper, we study the distributed experts problem, where $n$ experts are distributed across $s$ servers for $T$ timesteps. The loss of each expert at each time $t$ is the $\ell_p$ norm of the vector that consists of the losses of the expert at each of the $s$ servers at time $t$. The goal is to minimize the regret $R$, i.e., the loss of the distributed protocol compared to the loss of the best expert, amortized over the all $T$ times, while using the minimum amount of communication. We give a protocol that achieves regret roughly $R\gtrsim\frac{1}{\sqrt{T}\cdot\text{poly}\log(nsT)}$, using $\mathcal{O}\left(\frac{n}{R^2}+\frac{s}{R^2}\right)\cdot\max(s^{1-2/p},1)\cdot\text{poly}\log(nsT)$ bits of communication, which improves on previous work.

Accurate lithology identification from well logs is crucial for subsurface resource evaluation. Although Transformer-based models excel at sequence modeling, their "black-box" nature and lack of geological guidance limit their performance and trustworthiness. To overcome these limitations, this letter proposes the Geologically-Informed Attention Transformer (GIAT), a novel framework that deeply fuses data-driven geological priors with the Transformer's attention mechanism. The core of GIAT is a new attention-biasing mechanism. We repurpose Category-Wise Sequence Correlation (CSC) filters to generate a geologically-informed relational matrix, which is injected into the self-attention calculation to explicitly guide the model toward geologically coherent patterns. On two challenging datasets, GIAT achieves state-of-the-art performance with an accuracy of up to 95.4%, significantly outperforming existing models. More importantly, GIAT demonstrates exceptional interpretation faithfulness under input perturbations and generates geologically coherent predictions. Our work presents a new paradigm for building more accurate, reliable, and interpretable deep learning models for geoscience applications.

Learning effective netlist representations is fundamentally constrained by the scarcity of labeled datasets, as real designs are protected by Intellectual Property (IP) and costly to annotate. Existing work therefore focuses on small-scale circuits with clean labels, limiting scalability to realistic designs. Meanwhile, Large Language Models (LLMs) can generate Register-Transfer-Level (RTL) at scale, but their functional incorrectness has hindered their use in circuit analysis. In this work, we make a key observation: even when LLM-Generated RTL is functionally imperfect, the synthesized netlists still preserve structural patterns that are strongly indicative of the intended functionality. Building on this insight, we propose a cost-effective data augmentation and training framework that systematically exploits imperfect LLM-Generated RTL as training data for netlist representation learning, forming an end-to-end pipeline from automated code generation to downstream tasks. We conduct evaluations on circuit functional understanding tasks, including sub-circuit boundary identification and component classification, across benchmarks of increasing scales, extending the task scope from operator-level to IP-level. The evaluations demonstrate that models trained on our noisy synthetic corpus generalize well to real-world netlists, matching or even surpassing methods trained on scarce high-quality data and effectively breaking the data bottleneck in circuit representation learning.

Dense image captioning is critical for cross-modal alignment in vision-language pretraining and text-to-image generation, but scaling expert-quality annotations is prohibitively expensive. While synthetic captioning via strong vision-language models (VLMs) is a practical alternative, supervised distillation often yields limited output diversity and weak generalization. Reinforcement learning (RL) could overcome these limitations, but its successes have so far been concentrated in verifiable domains that rely on deterministic checkers -- a luxury not available in open-ended captioning. We address this bottleneck with RubiCap, a novel RL framework that derives fine-grained, sample-specific reward signals from LLM-written rubrics. RubiCap first assembles a diverse committee of candidate captions, then employs an LLM rubric writer to extract consensus strengths and diagnose deficiencies in the current policy. These insights are converted into explicit evaluation criteria, enabling an LLM judge to decompose holistic quality assessment and replace coarse scalar rewards with structured, multi-faceted evaluations. Across extensive benchmarks, RubiCap achieves the highest win rates on CapArena, outperforming supervised distillation, prior RL methods, human-expert annotations, and GPT-4V-augmented outputs. On CaptionQA, it demonstrates superior word efficiency: our 7B model matches Qwen2.5-VL-32B-Instruct, and our 3B model surpasses its 7B counterpart. Remarkably, using the compact RubiCap-3B as a captioner produces stronger pretrained VLMs than those trained on captions from proprietary models.

As large language models evolve from conversational assistants to autonomous agents, ensuring trustworthiness requires a fundamental shift from post-hoc evaluation to real-time action verification. Current frameworks like AgentBench evaluate task completion, while TrustLLM and HELM assess output quality after generation. However, none of these prevent harmful actions during agent execution. We present TrustBench, a dual-mode framework that (1) benchmarks trust across multiple dimensions using both traditional metrics and LLM-as-a-Judge evaluations, and (2) provides a toolkit agents invoke before taking actions to verify safety and reliability. Unlike existing approaches, TrustBench intervenes at the critical decision point: after an agent formulates an action but before execution. Domain-specific plugins encode specialized safety requirements for healthcare, finance, and technical domains. Across multiple agentic tasks, TrustBench reduced harmful actions by 87%. Domain-specific plugins outperformed generic verification, achieving 35% greater harm reduction. With sub-200ms latency, TrustBench enables practical real-time trust verification for autonomous agents.

Large language models (LLMs) trained on internet-scale corpora can exhibit systematic biases that increase the probability of unwanted behavior. In this study, we examined potential biases towards synthetic vs. biological technological solutions across four domains (materials, energy, manufacturing, and algorithms). A sample of 5 frontier and 5 open-weight models were measured using 50 curated Bioalignment prompts with a Kelly criterion-inspired evaluation framework. According to this metric, most models were not bioaligned in that they exhibit biases in favor of synthetic (non-biological) solutions. We next examined if fine-tuning could increase the preferences of two open-weight models, Llama 3.2-3B-Instruct and Qwen2.5-3B-Instruct, for biological-based approaches. A curated corpus of ~22M tokens from 6,636 PMC articles emphasizing biological problem-solving was used first to fine-tune Llama 3B with a mixed corpus of continued training and instruction-formatted. This was then extended to Qwen 3B using instruction-formatted only. We found that QLoRA fine-tuning significantly increased the scoring of biological solutions for both models without degrading general capabilities (Holm-Bonferroni-corrected p < 0.001 and p < 0.01, respectively). This suggests that even a small amount of fine-tuning can change how models weigh the relative value of biological and bioinspired vs. synthetic approaches. Although this work focused on small open-weight LLMs, it may be extensible to much larger models and could be used to develop models that favor bio-based approaches. We release the benchmark, corpus, code, and adapter weights.

Table Question Answering (TableQA) enables natural language interaction with structured tabular data. However, existing large language model (LLM) approaches face critical limitations: context length constraints that restrict data handling capabilities, hallucination issues that compromise answer reliability, and single-agent architectures that struggle with complex reasoning scenarios involving semantic relationships and multi-hop logic. This paper introduces DataFactory, a multi-agent framework that addresses these limitations through specialized team coordination and automated knowledge transformation. The framework comprises a Data Leader employing the ReAct paradigm for reasoning orchestration, together with dedicated Database and Knowledge Graph teams, enabling the systematic decomposition of complex queries into structured and relational reasoning tasks. We formalize automated data-to-knowledge graph transformation via the mapping function T:D x S x R -> G, and implement natural language-based consultation that - unlike fixed workflow multi-agent systems - enables flexible inter-agent deliberation and adaptive planning to improve coordination robustness. We also apply context engineering strategies that integrate historical patterns and domain knowledge to reduce hallucinations and improve query accuracy. Across TabFact, WikiTableQuestions, and FeTaQA, using eight LLMs from five providers, results show consistent gains. Our approach improves accuracy by 20.2% (TabFact) and 23.9% (WikiTQ) over baselines, with significant effects (Cohen's d > 1). Team coordination also outperforms single-team variants (+5.5% TabFact, +14.4% WikiTQ, +17.1% FeTaQA ROUGE-2). The framework offers design guidelines for multi-agent collaboration and a practical platform for enterprise data analysis through integrated structured querying and graph-based knowledge representation.

Large language models often struggle with complex long-horizon analytical tasks over unstructured tables, which typically feature hierarchical and bidirectional headers and non-canonical layouts. We formalize this challenge as Deep Tabular Research (DTR), requiring multi-step reasoning over interdependent table regions. To address DTR, we propose a novel agentic framework that treats tabular reasoning as a closed-loop decision-making process. We carefully design a coupled query and table comprehension for path decision making and operational execution. Specifically, (i) DTR first constructs a hierarchical meta graph to capture bidirectional semantics, mapping natural language queries into an operation-level search space; (ii) To navigate this space, we introduce an expectation-aware selection policy that prioritizes high-utility execution paths; (iii) Crucially, historical execution outcomes are synthesized into a siamese structured memory, i.e., parameterized updates and abstracted texts, enabling continual refinement. Extensive experiments on challenging unstructured tabular benchmarks verify the effectiveness and highlight the necessity of separating strategic planning from low-level execution for long-horizon tabular reasoning.

Current expansion-based methods for Class Incremental Learning (CIL) effectively mitigate catastrophic forgetting by freezing old features. However, such task-specific features learned from the new task may collide with the old features. From a causal perspective, spurious feature correlations are the main cause of this collision, manifesting in two scopes: (i) guided by empirical risk minimization (ERM), intra-task spurious correlations cause task-specific features to rely on shortcut features. These non-robust features are vulnerable to interference, inevitably drifting into the feature space of other tasks; (ii) inter-task spurious correlations induce semantic confusion between visually similar classes across tasks. To address this, we propose a Probability of Necessity and Sufficiency (PNS)-based regularization method to guide feature expansion in CIL. Specifically, we first extend the definition of PNS to expansion-based CIL, termed CPNS, which quantifies both the causal completeness of intra-task representations and the separability of inter-task representations. We then introduce a dual-scope counterfactual generator based on twin networks to ensure the measurement of CPNS, which simultaneously generates: (i) intra-task counterfactual features to minimize intra-task PNS risk and ensure causal completeness of task-specific features, and (ii) inter-task interfering features to minimize inter-task PNS risk, ensuring the separability of inter-task representations. Theoretical analyses confirm its reliability. The regularization is a plug-and-play method for expansion-based CIL to mitigate feature collision. Extensive experiments demonstrate the effectiveness of the proposed method.

Multi-agent systems (MAS) powered by LLMs promise adaptive, reasoning-driven enterprise workflows, yet granting agents autonomous control over tools, memory, and communication introduces attack surfaces absent from deterministic pipelines. While current research largely addresses prompt-level exploits and narrow individual vectors, it lacks a holistic architectural model for enterprise-grade security. We introduce AgenticCyOps (Securing Multi-Agentic AI Integration in Enterprise Cyber Operations), a framework built on a systematic decomposition of attack surfaces across component, coordination, and protocol layers, revealing that documented vectors consistently trace back to two integration surfaces: tool orchestration and memory management. Building on this observation, we formalize these integration surfaces as primary trust boundaries and define five defensive principles: authorized interfaces, capability scoping, verified execution, memory integrity & synchronization, and access-controlled data isolation; each aligned with established compliance standards (NIST, ISO 27001, GDPR, EU AI Act). We apply the framework to a Security Operations Center (SOC) workflow, adopting the Model Context Protocol (MCP) as the structural basis, with phase-scoped agents, consensus validation loops, and per-organization memory boundaries. Coverage analysis, attack path tracing, and trust boundary assessment confirm that the design addresses the documented attack vectors with defense-in-depth, intercepts three of four representative attack chains within the first two steps, and reduces exploitable trust boundaries by a minimum of 72% compared to a flat MAS, positioning AgenticCyOps as a foundation for securing enterprise-grade integration.

Collective AI systems increasingly rely on multi-LLM deliberation, but their stability under repeated execution remains poorly characterized. We model five-agent LLM committees as random dynamical systems and quantify inter-run sensitivity using an empirical Lyapunov exponent ($\hatλ$) derived from trajectory divergence in committee mean preferences. Across 12 policy scenarios, a factorial design at $T=0$ identifies two independent routes to instability: role differentiation in homogeneous committees and model heterogeneity in no-role committees. Critically, these effects appear even in the $T=0$ regime where practitioners often expect deterministic behavior. In the HL-01 benchmark, both routes produce elevated divergence ($\hatλ=0.0541$ and $0.0947$, respectively), while homogeneous no-role committees also remain in a positive-divergence regime ($\hatλ=0.0221$). The combined mixed+roles condition is less unstable than mixed+no-role ($\hatλ=0.0519$ vs $0.0947$), showing non-additive interaction. Mechanistically, Chair-role ablation reduces $\hatλ$ most strongly, and targeted protocol variants that shorten memory windows further attenuate divergence. These results support stability auditing as a core design requirement for multi-LLM governance systems.

Diffusion-based image super-resolution (ISR) has shown strong potential, but it still struggles in real-world scenarios where degradations are unknown and spatially non-uniform, often resulting in lost details or visual artifacts. To address this challenge, we propose a novel super-resolution diffusion model, QUSR, which integrates a Quality-Aware Prior (QAP) with an Uncertainty-Guided Noise Generation (UNG) module. The UNG module adaptively adjusts the noise injection intensity, applying stronger perturbations to high-uncertainty regions (e.g., edges and textures) to reconstruct complex details, while minimizing noise in low-uncertainty regions (e.g., flat areas) to preserve original information. Concurrently, the QAP leverages an advanced Multimodal Large Language Model (MLLM) to generate reliable quality descriptions, providing an effective and interpretable quality prior for the restoration process. Experimental results confirm that QUSR can produce high-fidelity and high-realism images in real-world scenarios. The source code is available at https://github.com/oTvTog/QUSR.

While Vision-Language-Action (VLA) models have demonstrated promising generalization capabilities in robotic manipulation, deploying them on specific and complex downstream tasks still demands effective post-training. In parallel, Human-in-the-Loop (HiL) learning has proven to be a powerful mechanism for refining robot policies. However, extending this paradigm to dexterous manipulation remains challenging: multi-finger control is high-dimensional, contact-intensive, and exhibits execution distributions that differ markedly from standard arm motions, leaving existing dexterous VLA systems limited in reliability and adaptability. We present DexHiL, the first integrated arm-hand human-in-the-loop framework for dexterous VLA models, enabling coordinated interventions over the arm and the dexterous hand within a single system. DexHiL introduces an intervention-aware data sampling strategy that prioritizes corrective segments for post-training, alongside a lightweight teleoperation interface that supports instantaneous human corrections during execution. Real-robot experiments demonstrate that DexHiL serves as an effective post-training framework, yielding a substantial performance leap, outperforming standard offline-only fine-tuning baselines by an average of 25% in success rates across distinct tasks. Project page: https://chenzhongxi-sjtu.github.io/dexhil/

Reinforcement Learning from Verifiable Rewards (RLVR) significantly enhances large language models (LLMs) reasoning but severely suffers from calibration degeneration, where models become excessively over-confident in incorrect answers. Previous studies devote to directly incorporating calibration objective into existing optimization target. However, our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. Building on this insight, we propose DCPO, a simple yet effective framework that systematically decouples reasoning and calibration objectives. Extensive experiments demonstrate that our DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue. Our study provides valuable insights and practical solution for more reliable LLM deployment.

Existing language-driven embodied navigation paradigms face challenges in functional buildings (FBs) with highly similar features, as they lack the ability to effectively utilize priori spatial knowledge. To tackle this issue, we propose a Priori-Map Guided Embodied Navigation (PM-Nav), wherein environmental maps are transformed into navigation-friendly semantic priori-maps, a hierarchical chain-of-thought prompt template with an annotation priori-map is designed to enable precise path planning, and a multi-model collaborative action output mechanism is built to accomplish positioning decisions and execution control for navigation planning. Comprehensive tests using a home-made FB dataset show that the PM-Nav obtains average improvements of 511\% and 1175\%, and 650\% and 400\% over the SG-Nav and the InstructNav in simulation and real-world, respectively. These tremendous boosts elucidate the great potential of using the PM-Nav as a backbone navigation framework for FBs.

Vision-language pretraining has driven significant progress in medical image analysis. However, current methods typically supervise visual encoders using one-hot labels or free-form text, neither of which effectively captures the complex semantic relationships among clinical findings. In this study, we introduce VIVID-Med, a novel framework that leverages a frozen large language model (LLM) as a structured semantic teacher to pretrain medical vision transformers (ViTs). VIVID-Med translates clinical findings into verifiable JSON field-state pairs via a Unified Medical Schema (UMS), utilizing answerability-aware masking to focus optimization. It then employs Structured Prediction Decomposition (SPD) to partition cross-attention into orthogonality-regularized query groups, extracting complementary visual aspects. Crucially, the LLM is discarded post-training, yielding a lightweight, deployable ViT-only backbone. We evaluated VIVID-Med across multiple settings: on CheXpert linear probing, it achieves a macro-AUC of 0.8588, outperforming BiomedCLIP by +6.65 points while using 500x less data. It also demonstrates robust zero-shot cross-domain transfer to NIH ChestX-ray14 (0.7225 macro-AUC) and strong cross-modality generalization to CT, achieving 0.8413 AUC on LIDC-IDRI lung nodule classification and 0.9969 macro-AUC on OrganAMNIST 11-organ classification. VIVID-Med offers a highly efficient, scalable alternative to deploying resource-heavy vision-language models in clinical settings.

Medical image retrieval aims to identify clinically relevant lesion cases to support diagnostic decision making, education, and quality control. In practice, retrieval queries often combine a reference lesion image with textual descriptors such as dermoscopic features. We study composed vision-language retrieval for skin cancer, where each query consists of an image to text pair and the database contains biopsy-confirmed, multi-class disease cases. We propose a transformer based framework that learns hierarchical composed query representations and performs joint global-local alignment between queries and candidate images. Local alignment aggregates discriminative regions via multiple spatial attention masks, while global alignment provides holistic semantic supervision. The final similarity is computed through a convex, domain-informed weighting that emphasizes clinically salient local evidence while preserving global consistency. Experiments on the public Derm7pt dataset demonstrate consistent improvements over state-of-the-art methods. The proposed framework enables efficient access to relevant medical records and supports practical clinical deployment.

Batteries with silicon-graphite-based anodes, which offer higher energy density and improved charging performance, introduce pronounced voltage hysteresis, making state-of-charge (SoC) estimation particularly challenging. Existing approaches to modeling hysteresis rely on exhaustive high-fidelity tests or focus on conventional graphite-based lithium-ion batteries, without considering uncertainty quantification or computational constraints. This work introduces a data-driven approach for probabilistic hysteresis factor prediction, with a particular emphasis on applications involving silicon-graphite anode-based batteries. A data harmonization framework is proposed to standardize heterogeneous driving cycles across varying operating conditions. Statistical learning and deep learning models are applied to assess performance in predicting the hysteresis factor with uncertainties while considering computational efficiency. Extensive experiments are conducted to evaluate the generalizability of the optimal model configuration in unseen vehicle models through retraining, zero-shot prediction, fine-tuning, and joint training. By addressing key challenges in SoC estimation, this research facilitates the adoption of advanced battery technologies. A summary page is available at: https://runyao-yu.github.io/Porsche_Hysteresis_Factor_Prediction/

The emergence of Large Language Models (LLMs) has opened new opportunities to automate software engineering activities that traditionally require substantial manual effort. Among these, class diagram generation represents a critical yet resource-intensive phase in software design. This paper investigates the capabilities of state-of-the-art LLMs, including GPT-5, Claude Sonnet 4.0, Gemini 2.5 Flash Thinking, and Llama-3.1-8B-Instruct, to generate UML class diagrams from natural language requirements automatically. To evaluate the effectiveness and reliability of LLM-based model generation, we propose a comprehensive dual-validation framework that integrates an LLM-as-a-Judge methodology with human-in-the-loop assessment. Using eight heterogeneous datasets, we apply chain-of-thought prompting to extract domain entities, attributes, and associations, generating corresponding PlantUML representations. The resulting models are evaluated across five quality dimensions: completeness, correctness, conformance to standards, comprehensibility, and terminological alignment. Two independent LLM judges (Grok and Mistral) perform structured pairwise comparisons, and their judgments are further validated against expert evaluations. Our results demonstrate that LLMs can generate structurally coherent and semantically meaningful UML diagrams, achieving substantial alignment with human evaluators. The consistency observed between LLM-based and human-based assessments highlights the potential of LLMs not only as modeling assistants but also as reliable evaluators in automated requirements engineering workflows, offering practical insights into the capabilities and limitations of LLM-driven UML class diagram automation.

Multimodal large language models (MLLMs) can process text presented as images, yet they often perform worse than when the same content is provided as textual tokens. We systematically diagnose this "modality gap" by evaluating seven MLLMs across seven benchmarks in five input modes, spanning both synthetically rendered text and realistic document images from arXiv PDFs to Wikipedia pages. We find that the modality gap is task- and data-dependent. For example, math tasks degrade by over 60 points on synthetic renderings, while natural document images often match or exceed text-mode performance. Rendering choices such as font and resolution are strong confounds, with font alone swinging accuracy by up to 47 percentage points. To understand this, we conduct a grounded-theory error analysis of over 4,000 examples, revealing that image mode selectively amplifies reading errors (calculation and formatting failures) while leaving knowledge and reasoning errors largely unchanged, and that some models exhibit a chain-of-thought reasoning collapse under visual input. Motivated by these findings, we propose a self-distillation method that trains the model on its own pure text reasoning traces paired with image inputs, raising image-mode accuracy on GSM8K from 30.71% to 92.72% and transferring to unseen benchmarks without catastrophic forgetting. Overall, our study provides a systematic understanding of the modality gap and suggests a practical path toward improving visual text understanding in multimodal language models.

In reinforcement learning environments with state-dependent action validity, action masking consistently outperforms penalty-based handling of invalid actions, yet existing theory only shows that masking preserves the policy gradient theorem. We identify a distinct failure mode of unmasked training: it systematically suppresses valid actions at states the agent has not yet visited. This occurs because gradients pushing down invalid actions at visited states propagate through shared network parameters to unvisited states where those actions are valid. We prove that for softmax policies with shared features, when an action is invalid at visited states but valid at an unvisited state $s^*$, the probability $π(a \mid s^*)$ is bounded by exponential decay due to parameter sharing and the zero-sum identity of softmax logits. This bound reveals that entropy regularization trades off between protecting valid actions and sample efficiency, a tradeoff that masking eliminates. We validate empirically that deep networks exhibit the feature alignment condition required for suppression, and experiments on Craftax, Craftax-Classic, and MiniHack confirm the predicted exponential suppression and demonstrate that feasibility classification enables deployment without oracle masks.

Emerging generative world models and vision-language-action (VLA) systems are rapidly reshaping automated driving by enabling scalable simulation, long-horizon forecasting, and capability-rich decision making. Across these directions, latent representations serve as the central computational substrate: they compress high-dimensional multi-sensor observations, enable temporally coherent rollouts, and provide interfaces for planning, reasoning, and controllable generation. This paper proposes a unifying latent-space framework that synthesizes recent progress in world models for automated driving. The framework organizes the design space by the target and form of latent representations (latent worlds, latent actions, latent generators; continuous states, discrete tokens, and hybrids) and by structural priors for geometry, topology, and semantics. Building on this taxonomy, the paper articulates five cross-cutting internal mechanics (i.e, structural isomorphism, long-horizon temporal stability, semantic and reasoning alignment, value-aligned objectives and post-training, as well as adaptive computation and deliberation) and connects these design choices to robustness, generalization, and deployability. The work also proposes concrete evaluation prescriptions, including a closed-loop metric suite and a resource-aware deliberation cost, designed to reduce the open-loop / closed-loop mismatch. Finally, the paper identifies actionable research directions toward advancing latent world model for decision-ready, verifiable, and resource-efficient automated driving.

By capturing the prevailing sentiment and market mood, textual data has become increasingly vital for forecasting commodity prices, particularly in metal markets. However, the effectiveness of lightweight, finetuned large language models (LLMs) in extracting predictive signals for aluminum prices, and the specific market conditions under which these signals are most informative, remains under-explored. This study generates monthly sentiment scores from English and Chinese news headlines (Reuters, Dow Jones Newswires, and China News Service) and integrates them with traditional tabular data, including base metal indices, exchange rates, inflation rates, and energy prices. We evaluate the predictive performance and economic utility of these models through long-short simulations on the Shanghai Metal Exchange from 2007 to 2024. Our results demonstrate that during periods of high volatility, Long Short-Term Memory (LSTM) models incorporating sentiment data from a finetuned Qwen3 model (Sharpe ratio 1.04) significantly outperform baseline models using tabular data alone (Sharpe ratio 0.23). Subsequent analysis elucidates the nuanced roles of news sources, topics, and event types in aluminum price forecasting.

To support latency-sensitive Internet of Vehicles (IoV) applications amidst dynamic environments and intermittent links, this paper proposes a Reconfigurable Intelligent Surface (RIS)-aided semantic-aware Vehicle Edge Computing (VEC) framework. This approach integrates RIS to optimize wireless connectivity and semantic communication to minimize latency by transmitting semantic features. We formulate a comprehensive joint optimization problem by optimizing offloading ratios, the number of semantic symbols, and RIS phase shifts. Considering the problem's high dimensionality and non-convexity, we propose a two-tier hybrid scheme that employs Proximal Policy Optimization (PPO) for discrete decision-making and Linear Programming (LP) for offloading optimization. {The simulation results have validated the proposed framework's superiority over existing methods. Specifically, the proposed PPO-based hybrid optimization scheme reduces the average end-to-end latency by approximately 40% to 50% compared to Genetic Algorithm (GA) and Quantum-behaved Particle Swarm Optimization (QPSO). Moreover, the system demonstrates strong scalability by maintaining low latency even in congested scenarios with up to 30 vehicles.

VLA models encode visual observations as 2D patch tokens with no intrinsic geometric structure. We introduce GST-VLA with two contributions. First, the Gaussian Spatial Tokenizer (GST) converts frozen dense depth and frozen semantic patch features into $N_g{=}128$ anisotropic 3D Gaussian primitives, each parameterized by a metric residual mean $μ\in \mathbb{R}^3$, log-scale covariance $\log σ\in \mathbb{R}^3$, and learned opacity $α\in (0,1)$. The covariance eigenstructure encodes local surface orientation, and opacity provides per-primitive geometric confidence, both inaccessible from scalar depth. Spatial attention pooling with learned queries concentrates the fixed token budget on geometrically salient regions rather than distributing uniformly. Second, 3D Depth-Aware Chain-of-Thought (DA-CoT) reasoning supervises four structured intermediate spatial thoughts, covering 3D object grounding, grasp affordance contact geometry, pairwise metric distances, and coarse SE(3) waypoints, as explicit generation targets in the training loss. A cross-attention sublayer at every VLM transformer block provides direct access to the raw 256-primitive Gaussian field during DA-CoT generation. A 300M-parameter flow-matching action expert with mixture-of-experts feedforward sublayers decodes 7-DoF delta action chunks via conditional ODE integration, conditioned on both VLM hidden states and DA-CoT outputs through dual cross-attention. Trained with composite $\mathcal{L}_\mathrm{flow} + \mathcal{L}_\mathrm{CoT} + \mathcal{L}_\mathrm{depth}$ across three progressive stages, GST-VLA achieves 96.4% on LIBERO (+2.0%), and 80.2% on SimplerEnv (+5.4%). Ablations isolate the contribution of each GST component, each DA-CoT thought, and each training stage, confirming independent and synergistic gains concentrated on precision demanding tasks.

We introduce exclusive self attention (XSA), a simple modification of self attention (SA) that improves Transformer's sequence modeling performance. The key idea is to constrain attention to capture only information orthogonal to the token's own value vector (thus excluding information of self position), encouraging better context modeling. Evaluated on the standard language modeling task, XSA consistently outperforms SA across model sizes up to 2.7B parameters and shows increasingly larger gains as sequence length grows.

Everyone can write their stories in freeform text format -- it's something we all learn in school. Yet storytelling via video requires one to learn specialized and complicated tools. In this paper, we introduce Doki, a text-native interface for generative video authoring, aligning video creation with the natural process of text writing. In Doki, writing text is the primary interaction: within a single document, users define assets, structure scenes, create shots, refine edits, and add audio. We articulate the design principles of this text-first approach and demonstrate Doki's capabilities through a series of examples. To evaluate its real-world use, we conducted a week-long deployment study with participants of varying expertise in video authoring. This work contributes a fundamental shift in generative video interfaces, demonstrating a powerful and accessible new way to craft visual stories.

We verify that persistent observers in causally invariant hypergraph substrates satisfy the conditions of the Conant-Ashby Good Regulator Theorem. Building on Wolfram's hypergraph physics and Vanchurin's neural network cosmology, we formalize persistent observers as entities that minimize prediction error at their boundary with the environment. Applying a modern reformulation of the Conant-Ashby theorem, we demonstrate that hypergraph observers satisfy Good Regulator conditions, requiring them to maintain internal models. Once an internal model with loss function exists, the emergence of a Fisher information metric follows from standard information geometry. Invoking Amari's uniqueness theorem for reparameterization-invariant gradients, we show that natural gradient descent is the unique admissible learning rule. Under the ansatz M=F^2 for exponential family observers and one specific convergence time functional, we derive a closed-form formula for the regime parameter alpha in Vanchurin's Type II framework, with a quantum-classical threshold at kappa(F)=2. However, three alternative convergence models do not reproduce this result, so this prediction is strongly model-dependent. We further introduce the directional regime parameter alpha_{v_k} and the trace-free deviation tensor, showing that a single observer can simultaneously occupy different Vanchurin regimes along different eigendirections of the Fisher metric. This connects Wolfram and Vanchurin frameworks through established theorems, providing approximately 25-30% novel contribution.

Decoding from large language models (LLMs) typically relies on fixed sampling hyperparameters (e.g., temperature, top-p), despite substantial variation in task difficulty and uncertainty across prompts and individual decoding steps. We propose to learn adaptive decoding policies that dynamically select sampling strategies at inference time, conditioned on available compute resources. Rather than fine-tuning the language model itself, we introduce lightweight decoding adapters trained with reinforcement learning and verifiable terminal rewards (e.g. correctness on math and coding tasks). At the sequence level, we frame decoding as a contextual bandit problem: a policy selects a decoding strategy (e.g. greedy, top-k, min-p) for each prompt, conditioned on the prompt embedding and a parallel sampling budget. At the token level, we model decoding as a partially observable Markov decision process (POMDP), where a policy selects sampling actions at each token step based on internal model features and the remaining token budget. Experiments on the MATH and CodeContests benchmarks show that the learned adapters improve the accuracy-budget tradeoff: on MATH, the token-level adapter improves Pass@1 accuracy by up to 10.2% over the best static baseline under a fixed token budget, while the sequence-level adapter yields 2-3% gains under fixed parallel sampling. Ablation analyses support the contribution of both sequence- and token-level adaptation.

Online Time Series Forecasting (OTSF) requires models to continuously adapt to concept drift. However, existing methods often treat concept drift as a monolithic phenomenon. To address this limitation, we first redefine concept drift by categorizing it into two distinct types: Recurring Drift, where previously seen patterns reappear, and Emergent Drift, where entirely new patterns emerge. We then propose DynaME (Dynamic Multi-period Experts), a novel hybrid framework designed to effectively address this dual nature of drift. For Recurring Drift, DynaME employs a committee of specialized experts that are dynamically fitted to the most relevant historical periodic patterns at each time step. For Emergent Drift, the framework detects high-uncertainty scenarios and shifts reliance to a stable, general expert. Extensive experiments on several benchmark datasets and backbones demonstrate that DynaME effectively adapts to both concept drifts and significantly outperforms existing baselines.

Accurate on-orbit reliability prediction for satellite electronics is often hindered by limited data availability, varying operational conditions, and considerable unit-to-unit variability. To overcome these obstacles, this paper proposes a novel integrated online reliability prediction framework. The main contributions are twofold. First, a Wiener process-based degradation model is developed, incorporating a generalized Arrhenius link function, individual random effects, and spatial correlations among adjacent units. A customized maximum likelihood estimation method is further devised to facilitate efficient and accurate parameter inference. Second, a two-stage active learning sampling scheme is designed to adaptively enhance prediction accuracy. This strategy initially selects representative units based on spatial configuration, and subsequently determines optimal sampling times using a comprehensive criterion that balances unit-specific information, model uncertainty, and degradation dynamics. Numerical experiments and a practical case study from the Tiangong space station demonstrate that the proposed method markedly improves reliability prediction accuracy while significantly reducing data requirements, offering an efficient solution for the prognostic and health management of complex satellite electronic systems.

Learning from demonstrations has emerged as a promising paradigm for end-to-end robot control, particularly when scaled to diverse and large datasets. However, the quality of demonstration data, often collected through human teleoperation, remains a critical bottleneck for effective data-driven robot learning. Human errors, operational constraints, and teleoperator variability introduce noise and suboptimal behaviors, making data curation essential yet largely manual and heuristic-driven. In this work, we propose Quality over Quantity (QoQ), a grounded and systematic approach to identifying high-quality data by defining data quality as the contribution of each training sample to reducing loss on validation demonstrations. To efficiently estimate this contribution, we leverage influence functions, which quantify the impact of individual training samples on model performance. We further introduce two key techniques to adapt influence functions for robot demonstrations: (i) using maximum influence across validation samples to capture the most relevant state-action pairs, and (ii) aggregating influence scores of state-action pairs within the same trajectory to reduce noise and improve data coverage. Experiments in both simulated and real-world settings show that QoQ consistently improves policy performances over prior data selection methods.

Simulation-to-decision learning enables safe policy training in digital environments without risking real-world deployment, and has become essential in mission-critical domains such as supply chains and industrial systems. However, simulators learned from noisy or biased real-world data often exhibit prediction errors in decision-critical regions, leading to unstable action ranking and unreliable policies. Existing approaches either focus on improving average simulation fidelity or adopt conservative regularization, which may cause policy collapse by discarding high-risk high-reward actions. We propose Sim2Act, a robust simulation-to-decision framework that addresses both simulator and policy robustness. First, we introduce an adversarial calibration mechanism that re-weights simulation errors in decision-critical state-action pairs to align surrogate fidelity with downstream decision impact. Second, we develop a group-relative perturbation strategy that stabilizes policy learning under simulator uncertainty without enforcing overly pessimistic constraints. Extensive experiments on multiple supply chain benchmarks demonstrate improved simulation robustness and more stable decision performance under structured and unstructured perturbations.

Background: Remote patient monitoring (RPM) generates vast data, yet landmark trials (Tele-HF, BEAT-HF) failed because data volume overwhelmed clinical staff. While TIM-HF2 showed 24/7 physician-led monitoring reduces mortality by 30%, this model remains prohibitively expensive and unscalable. Methods: We developed Sentinel, an autonomous AI agent using Model Context Protocol (MCP) for contextual triage of RPM vitals via 21 clinical tools and multi-step reasoning. Evaluation included: (1) self-consistency (100 readings x 5 runs); (2) comparison against rule-based thresholds; and (3) validation against 6 clinicians (3 physicians, 3 NPs) using a connected matrix design. A leave-one-out (LOO) analysis compared the agent against individual clinicians; severe overtriage cases underwent independent physician adjudication. Results: Against a human majority-vote standard (N=467), the agent achieved 95.8% emergency sensitivity and 88.5% sensitivity for all actionable alerts (85.7% specificity). Four-level exact accuracy was 69.4% (quadratic-weighted kappa=0.778); 95.9% of classifications were within one severity level. In LOO analysis, the agent outperformed every clinician in emergency sensitivity (97.5% vs. 60.0% aggregate) and actionable sensitivity (90.9% vs. 69.5%). While disagreements skewed toward overtriage (22.5%), independent adjudication of severe gaps (>=2 levels) validated agent escalation in 88-94% of cases; consensus resolution validated 100%. The agent showed near-perfect self-consistency (kappa=0.850). Median cost was $0.34/triage. Conclusions: Sentinel triages RPM vitals with sensitivity exceeding individual clinicians. By automating systematic context synthesis, Sentinel addresses the core limitation of prior RPM trials, offering a scalable path toward the intensive monitoring shown to reduce mortality while maintaining a clinically defensible overtriage profile.

Autonomous agents are increasingly used to improve prompts, code, and machine learning systems through iterative execution and feedback. Yet existing approaches are usually designed as task-specific optimization loops rather than as a unified protocol for establishing baselines and managing tracked multi-round self-improvement. We introduce EPOCH, an engineering protocol for multi-round system optimization in heterogeneous environments. EPOCH organizes optimization into two phases: baseline construction and iterative self-improvement. It further structures each round through role-constrained stages that separate planning, implementation, and evaluation, and standardizes execution through canonical command interfaces and round-level tracking. This design enables coordinated optimization across prompts, model configurations, code, and rule-based components while preserving stability, reproducibility, traceability, and integrity of evaluation. Empirical studies in various tasks illustrate the practicality of EPOCH for production-oriented autonomous improvement workflows.

Device-side Large Language Models (LLMs) have witnessed explosive growth, offering higher privacy and availability compared to cloud-side LLMs. During LLM inference, both model weights and user data are valuable, and attackers may even compromise the OS kernel to steal them. ARM TrustZone is the de facto hardware-based isolation technology on mobile devices, used to protect sensitive applications from a compromised OS. However, protecting LLM inference with TrustZone incurs significant overhead due to its inflexible isolation of memory and the NPU. To address these challenges, this paper introduces FlexServe, a fast and secure LLM serving system for mobile devices. It first introduces a Flexible Resource Isolation mechanism to construct Flexible Secure Memory (Flex-Mem) and Flexible Secure NPU (Flex-NPU). Both memory pages and the NPU can be efficiently switched between unprotected and protected modes. Based on these mechanisms, FlexServe designs a fast and secure LLM inference framework within TrustZone's secure world. The LLM-Aware Memory Management and Secure Inference Pipeline are introduced to accelerate inference. A Multi-Model Scheduler is proposed to optimize multi-model workflows. We implement a prototype of FlexServe and compare it with two TrustZone-based strawman designs. The results show that FlexServe achieves an average $10.05\times$ speedup in Time to First Token (TTFT) compared to the strawman, and an average $2.44\times$ TTFT speedup compared to an optimized strawman with pipeline and secure NPU enabled. For multi-model agent workflows, the end-to-end speedup is up to $24.30\times$ and $4.05\times$ compared to the strawman and optimized strawman, respectively.

The weaponization of LLMs for automated malware generation poses an existential threat to conventional detection paradigms. AI-generated malware exhibits polymorphic, metamorphic, and context-aware evasion capabilities that render signature-based and shallow heuristic defenses obsolete. This paper introduces a novel hybrid analysis framework that synergistically combines \emph{concolic execution} with \emph{LLM-augmented path prioritization} and \emph{deep-learning-based vulnerability classification} to detect zero-day AI-generated malware with provable guarantees. We formalize the detection problem within a first-order temporal logic over program execution traces, define a lattice-theoretic abstraction for path constraint spaces, and prove both the \emph{soundness} and \emph{relative completeness} of our detection algorithm, assuming classifier correctness. The framework introduces three novel algorithms: (i) an LLM-guided concolic exploration strategy that reduces the average number of explored paths by 73.2\% compared to depth-first search while maintaining equivalent malicious-path coverage; (ii) a transformer-based path-constraint classifier trained on symbolic execution traces; and (iii) a feedback loop that iteratively refines the LLM's prioritization policy using reinforcement learning from detection outcomes. We provide a comprehensive implementation built upon \texttt{angr} 9.2, \texttt{Z3} 4.12, Hugging Face Transformers 4.38, and PyTorch 2.2, with configuration details enabling reproducibility. Experimental evaluation on the EMBER, Malimg, SOREL-20M, and a novel AI-Gen-Malware benchmark comprising 2{,}500 LLM-synthesized samples demonstrates that achieves 98.7\% accuracy on conventional malware and 97.5\% accuracy on AI-generated threats, outperforming ClamAV, YARA, MalConv, and EMBER-GBDT baselines by margins of 8.4--52.2 percentage points on AI-generated samples.

Machine consciousness evaluations mostly see behavior. For language model agents that behavior is language and tool use. That lets an agent say the right things about itself even when the constraints that should make those statements matter are not jointly present at decision time. We apply Stack Theory's temporal gap to scaffold trajectories. This separates ingredient-wise occurrence within an evaluation window from co-instantiation at a single objective step. We then instantiate Stack Theory's Arpeggio and Chord postulates on grounded identity statements. This yields two persistence scores that can be computed from instrumented scaffold traces. We connect these scores to five operational identity metrics and map common scaffolds into an identity morphospace that exposes predictable tradeoffs. The result is a conservative toolkit for identity evaluation. It separates talking like a stable self from being organized like one.

Weak spectral responses in hyperspectral images are often obscured by dominant endmembers and sensor noise, resulting in inaccurate abundance estimation. This paper introduces WS-Net, a deep unmixing framework specifically designed to address weak-signal collapse through state-space modelling and Weak Signal Attention fusion. The network features a multi-resolution wavelet-fused encoder that captures both high-frequency discontinuities and smooth spectral variations with a hybrid backbone that integrates a Mamba state-space branch for efficient long-range dependency modelling. It also incorporates a Weak Signal Attention branch that selectively enhances low-similarity spectral cues. A learnable gating mechanism adaptively fuses both representations, while the decoder leverages KL-divergence-based regularisation to enforce separability between dominant and weak endmembers. Experiments on one simulated and two real datasets (synthetic dataset, Samson, and Apex) demonstrate consistent improvements over six state-of-the-art baselines, achieving up to 55% and 63% reductions in RMSE and SAD, respectively. The framework maintains stable accuracy under low-SNR conditions, particularly for weak endmembers, establishing WS-Net as a robust and computationally efficient benchmark for weak-signal hyperspectral unmixing.

LM-based agents excel when given high-level action APIs but struggle to ground language into low-level control. Prior work has LLMs generate skills or reward functions for RL, but these one-shot approaches lack feedback to correct specification errors. We introduce SCALAR, a bidirectional framework coupling LLM planning with RL through a learned skill library. The LLM proposes skills with preconditions and effects; RL trains policies for each skill and feeds back execution results to iteratively refine specifications, improving robustness to initial errors. Pivotal Trajectory Analysis corrects LLM priors by analyzing RL trajectories; Frontier Checkpointing optionally saves environment states at skill boundaries to improve sample efficiency. On Craftax, SCALAR achieves 88.2% diamond collection, a 1.9x improvement over the best baseline, and reaches the Gnomish Mines 9.1% of the time where prior methods fail entirely.

Software engineering (SE) conferences are vital for knowledge exchange and collaboration, yet can also involve significant barriers for researchers in geographically distant regions such as New Zealand. We identify barriers such as high travel costs, misaligned academic calendars, and limited representation, and propose strategies including hybrid participation, cost-conscious venues, and governance reforms. We make recommendations to promote equitable global participation and strengthen the SE research community.

Scientific machine learning (SciML) is increasingly applied to in-field processing, controlling, and monitoring; however, wide-area sensing, real-time demands, and strict energy and reliability constraints make centralized SciML implementation impractical. Most SciML models assume raw data aggregation at a central node, incurring prohibitively high communication latency and energy costs; yet, distributing models developed for general-purpose ML often breaks essential physical principles, resulting in degraded performance. To address these challenges, we introduce EPIC, a hardware- and physics-co-guided distributed SciML framework, using full-waveform inversion (FWI) as a representative task. EPIC performs lightweight local encoding on end devices and physics-aware decoding at a central node. By transmitting compact latent features rather than high-volume raw data and by using cross-attention to capture inter-receiver wavefield coupling, EPIC significantly reduces communication cost while preserving physical fidelity. Evaluated on a distributed testbed with five end devices and one central node, and across 10 datasets from OpenFWI, EPIC reduces latency by 8.9$\times$ and communication energy by 33.8$\times$, while even improving reconstruction fidelity on 8 out of 10 datasets.

Action-conditioned video models offer a promising path to building general-purpose robot simulators that can improve directly from data. Yet, despite training on large-scale robot datasets, current state-of-the-art video models still struggle to predict physically consistent robot-object interactions that are crucial in robotic manipulation. To close this gap, we present PlayWorld, a simple, scalable, and fully autonomous pipeline for training high-fidelity video world simulators from interaction experience. In contrast to prior approaches that rely on success-biased human demonstrations, PlayWorld is the first system capable of learning entirely from unsupervised robot self-play, enabling naturally scalable data collection while capturing complex, long-tailed physical interactions essential for modeling realistic object dynamics. Experiments across diverse manipulation tasks show that PlayWorld generates high-quality, physically consistent predictions for contact-rich interactions that are not captured by world models trained on human-collected data.We further demonstrate the versatility of PlayWorld in enabling fine-grained failure prediction and policy evaluation, with up to 40% improvements over human-collected data. Finally, we demonstrate how PlayWorld enables reinforcement learning in the world model, improving policy performance by 65% in success rates when deployed in the real world.

Like classical software, quantum software systems rely on automated testing. However, their inherently probabilistic outputs make them susceptible to quantum flakiness -- tests that pass or fail inconsistently without code changes. Such quantum flaky tests can mask real defects and reduce developer productivity, yet systematic tooling for their detection and diagnosis remains limited. This paper presents an automated pipeline to detect flaky-test-related issues and pull requests in quantum software repositories and to support the identification of their root causes. We aim to expand an existing quantum flaky test dataset and evaluate the capability of Large Language Models (LLMs) for flakiness classification and root-cause identification. Building on a prior manual analysis of 14 quantum software repositories, we automate the discovery of additional flaky test cases using LLMs and cosine similarity. We further evaluate a variety of LLMs from OpenAI GPT, Meta LLaMA, Google Gemini, and Anthropic Claude suites for classifying flakiness and identifying root causes from issue descriptions and code context. Classification performance is assessed using standard performance metrics, including F1-score. Using our pipeline, we identify 25 previously unknown flaky tests, increasing the original dataset size by 54%. The best-performing model, Google Gemini, achieves an F1-score of 0.9420 for flakiness detection and 0.9643 for root-cause identification, demonstrating that LLMs can provide practical support for triaging flaky reports and understanding their underlying causes in quantum software. The expanded dataset and automated pipeline provide reusable artifacts for the quantum software engineering community. Future work will focus on improving detection robustness and exploring automated repair of quantum flaky tests.

Enterprises increasingly rely on cloud-based applications to process highly sensitive data artifacts. Although cloud adoption improves agility and scalability, it also introduces new security challenges such as expanded attack surfaces, a wider radius of attack from credential compromise, and challenges maintaining strict access controls across users, services, and workflows. These challenges are especially acute for applications that handle privileged data and execute security-critical analysis, where traditional trust boundaries and ad hoc safeguards are insufficient. This paper presents Lockbox; a Zero Trust architecture designed for secure processing of sensitive cloud workloads under strict enterprise security and governance requirements. Lockbox applies explicit trust verification, strong isolation, least-privilege access, and policy-driven enforcement throughout the entire application lifecycle, from user authentication and document ingestion to analysis execution and result storage. The system incorporates modern cloud security primitives including; role-based access control, centralized key management, encryption in transit and at rest, and controlled integration with cloud-based data processing services, ensuring that sensitive artifacts remain protected and accessible only to authorized users. We discuss the usage of Lockbox in processing highly sensitive cybersecurity reports and demonstrate how this architecture enables organizations to safely adopt advanced capabilities, including AI-assisted processing, without weakening their security posture.

Sudden concept drift makes previously trained predictors unreliable, yet deciding when to retrain and what post-drift data size is sufficient is rarely addressed. We propose CALIPER - a detector- and model-agnostic, data-only test that estimates the post-drift data size required for stable retraining. CALIPER exploits state dependence in streams generated by dynamical systems: we run a single-pass weighted local regression over the post-drift window and track a one-step proxy error as a function of a locality parameter $θ$. When an effective sample size gate is satisfied, a monotonically non-increasing trend in this error with increasing a locality parameter indicates that the data size is sufficiently informative for retraining. We also provide a theoretical analysis of our method, and we show that the algorithm has a low per-update time and memory. Across datasets from four heterogeneous domains, three learner families, and two detectors, CALIPER consistently matches or exceeds the best fixed data size for retraining while incurring negligible overhead and often outperforming incremental updates. CALIPER closes the gap between drift detection and data-sufficient adaptation in streaming learning.

The context window of a large language model is not memory. It is L1 cache: a small, fast, expensive resource that the field treats as the entire memory system. There is no L2, no virtual memory, no paging. Every tool definition, every system prompt, and every stale tool result occupies context for the lifetime of the session. The result is measurable: across 857 production sessions and 4.45 million effective input tokens, 21.8% is structural waste. We present Pichay, a demand paging system for LLM context windows. Implemented as a transparent proxy between client and inference API, Pichay interposes on the message stream to evict stale content, detect page faults when the model re-requests evicted material, and pin working-set pages identified by fault history. In offline replay across 1.4 million simulated evictions, the fault rate is 0.0254%. In live production deployment over 681turns, the system reduces context consumption by up to 93% (5,038KB to 339KB); under extreme sustained pressure, the system remains operational but exhibits the expected thrashing pathology, with repeated fault-in of evicted content. The key observation is that the problems the field faces, such as context limits, attention degradation, cost scaling, lost state across sessions, are virtual memory problems wearing different clothes. The solutions exist: working set theory (Denning, 1968), demand paging, fault-driven replacement policies, and memory hierarchies with multiple eviction-managed levels. We describe the architecture of a full memory hierarchy for LLM systems (L1 through persistent storage), report on the first three levels deployed in production use (L1 eviction, L2 fault-driven pinning, L3 model-initiated conversation compaction), and identify cross-session memory as the remaining frontier.

Multi-turn, multi-agent LLM game evaluations often exhibit substantial run-to-run variance. In long-horizon interactions, small early deviations compound across turns and are amplified by multi-agent coupling. This biases win rate estimates and makes rankings unreliable across repeated tournaments. Prompt choice worsens this further by producing different effective policies. We address both instability and underperformance with MEMO (Memory-augmented MOdel context optimization), a self-play framework that optimizes inference-time context by coupling retention and exploration. Retention maintains a persistent memory bank that stores structured insights from self-play trajectories and injects them as priors during later play. Exploration runs tournament-style prompt evolution with uncertainty-aware selection via TrueSkill, and uses prioritized replay to revisit rare and decisive states. Across five text-based games, MEMO raises mean win rate from 25.1% to 49.5% for GPT-4o-mini and from 20.9% to 44.3% for Qwen-2.5-7B-Instruct, using $2,000$ self-play games per task. Run-to-run variance also drops, giving more stable rankings across prompt variations. These results suggest that multi-agent LLM game performance and robustness have substantial room for improvement through context optimization. MEMO achieves the largest gains in negotiation and imperfect-information games, while RL remains more effective in perfect-information settings.

There is no 'ordinary' when it comes to AI. The human-AI experience is extraordinarily complex and specific to each person, yet dominant measures such as usability scales and engagement metrics flatten away nuance. We argue for AI phenomenology: a research stance that asks "How did it feel?" beyond the standard questions of "How well did it perform?" when interacting with AI systems. AI phenomenology acts as a paradigm for bidirectional human-AI alignment as it foregrounds users' first-person perceptions and interpretations of AI systems over time. We motivate AI phenomenology as a framework that captures how alignment is experienced, negotiated, and updated between users and AI systems. Tracing a lineage from Husserl through postphenomenology to Actor-Network Theory, and grounding our argument in three studies-two longitudinal studies with "Day", an AI companion, and a multi-method study of agentic AI in software engineering-we contribute a set of replicable methodological toolkits for conducting AI phenomenology research: instruments for capturing lived experience across personal and professional contexts, three design concepts (translucent design, agency-aware value alignment, temporal co-evolution tracking), and a concrete research agenda. We offer this toolkit not as a new paradigm but as a practical scaffold that researchers can adapt as AI systems-and the humans who live alongside them-continue to co-evolve.

Multi-modal large language models (MM-LLMs) have shown strong performance in medical image understanding and clinical reasoning. Recent medical agent systems extend them with tool use and multi-agent collaboration, enabling complex decision-making. However, these systems rely almost entirely on frontier models (e.g., GPT), whose API-based deployment incurs high cost, high latency, and privacy risks that conflict with on-premise clinical requirements. We present Meissa, a lightweight 4B-parameter medical MM-LLM that brings agentic capability offline. Instead of imitating static answers, Meissa learns both when to engage external interaction (strategy selection) and how to execute multi-step interaction (strategy execution) by distilling structured trajectories from frontier models. Specifically, we propose: (1) Unified trajectory modeling: trajectories (reasoning and action traces) are represented within a single state-action-observation formalism, allowing one model to generalize across heterogeneous medical environments. (2) Three-tier stratified supervision: the model's own errors trigger progressive escalation from direct reasoning to tool-augmented and multi-agent interaction, explicitly learning difficulty-aware strategy selection. (3) Prospective-retrospective supervision: pairing exploratory forward traces with hindsight-rationalized execution traces enables stable learning of effective interaction policies. Trained on 40K curated trajectories, Meissa matches or exceeds proprietary frontier agents in 10 of 16 evaluation settings across 13 medical benchmarks spanning radiology, pathology, and clinical reasoning. Using over 25x fewer parameters than typical frontier models like Gemini-3, Meissa operates fully offline with 22x lower end-to-end latency compared to API-based deployment. Data, models, and environments are released at https://github.com/Schuture/Meissa.

Flatness measures based on the spectrum or the trace of the Hessian of the loss are widely used as proxies for the generalization ability of deep networks. However, most existing definitions are either tailored to fully connected architectures, relying on stochastic estimators of the Hessian trace, or ignore the specific geometric structure of modern Convolutional Neural Networks (CNNs). In this work, we develop a flatness measure that is both exact and architecturally faithful for a broad and practically relevant class of CNNs. We first derive a closed-form expression for the trace of the Hessian of the cross-entropy loss with respect to convolutional kernels in networks that use global average pooling followed by a linear classifier. Building on this result, we then specialize the notion of relative flatness to convolutional layers and obtain a parameterization-aware flatness measure that properly accounts for the scaling symmetries and filter interactions induced by convolution and pooling. Finally, we empirically investigate the proposed measure on families of CNNs trained on standard image-classification benchmarks. The results obtained suggest that the proposed measure can serve as a robust tool to assess and compare the generalization performance of CNN models, and to guide the design of architecture and training choices in practice.

Flow models have rapidly become the go-to method for training and deploying large-scale generators, owing their success to inference-time flexibility via adjustable integration steps. A crucial ingredient in flow training is the choice of coupling measure for sampling noise/data pairs that define the flow matching (FM) regression loss. While FM training defaults usually to independent coupling, recent works show that adaptive couplings informed by noise/data distributions (e.g., via optimal transport, OT) improve both model training and inference. We radicalize this insight by shifting the paradigm: rather than computing adaptive couplings directly, we use distilled couplings from a different, pretrained model capable of placing noise and data spaces in bijection -- a property intrinsic to normalizing flows (NF) through their maximum likelihood and invertibility requirements. Leveraging recent advances in NF image generation via auto-regressive (AR) blocks, we propose Normalized Flow Matching (NFM), a new method that distills the quasi-deterministic coupling of pretrained NF models to train student flow models. These students achieve the best of both worlds: significantly outperforming flow models trained with independent or even OT couplings, while also improving on the teacher AR-NF model.

Robots that interact with humans must adapt to individual users' preferences to operate effectively in human-centered environments. An intuitive and effective technique to learn non-expert users' preferences is through rankings of robot behaviors, e.g., trajectories, gestures, or voices. Existing techniques primarily focus on generating queries that optimize preference learning outcomes, such as sample efficiency or final preference estimation accuracy. However, the focus on outcome overlooks key user expectations in the process of providing these rankings, which can negatively impact users' adoption of robotic systems. This work proposes the Covariance Matrix Adaptation Evolution Strategies with Information Gain (CMA-ES-IG) algorithm. CMA-ES-IG explicitly incorporates user experience considerations into the preference learning process by suggesting perceptually distinct and informative trajectories for users to rank. We demonstrate these benefits through both simulated studies and real-robot experiments. CMA-ES-IG, compared to state-of-the-art alternatives, (1) scales more effectively to higher-dimensional preference spaces, (2) maintains computational tractability for high-dimensional problems, (3) is robust to noisy or inconsistent user feedback, and (4) is preferred by non-expert users in identifying their preferred robot behaviors. This project's code is available at github.com/interaction-lab/CMA-ES-IG

Generative AI has achieved remarkable empirical success, but from the perspective of statistics it often remains opaque: its predictions may be accurate, yet the underlying mechanism is difficult to interpret, analyze, and trust. This book reinterprets generative AI in the language of statistics, using flow matching as a central example. The key idea is that generative models should be understood not merely as devices for producing plausible data, but as methods for the nonparametric learning of high-dimensional probability distributions. From this viewpoint, missing-data imputation becomes principled sampling from learned conditional distributions, counterfactual analysis becomes the estimation of intervention distributions, and distributional dynamics become statistically analyzable objects. Mathematically, flow matching represents distributional deformation through the continuity equation and a time-dependent velocity field, thereby extending score matching from the learning of static score fields to the learning of transport paths themselves. Building on this foundation, the book develops a statistical framework in which generative models are used to estimate nuisance components while inferential validity is maintained through orthogonalization and cross-fitting in the spirit of double/debiased machine learning. Applications to survival analysis, censoring, missingness, and causal inference show how generative models can be integrated into statistical inference for structured high-dimensional problems.

Audio deepfake detection aims to detect real human voices from those generated by Artificial Intelligence (AI) and has emerged as a significant problem in the field of voice biometrics systems. With the ever-improving quality of synthetic voice, the probability of such a voice being exploited for illicit practices like identity thest and impersonation increases. Although significant progress has been made in the field of Audio Deepfake Detection in recent times, the issue of gender bias remains underexplored and in its nascent stage In this paper, we have attempted a thorough analysis of gender dependent performance and fairness in audio deepfake detection models. We have used the ASVspoof 5 dataset and train a ResNet-18 classifier and evaluate detection performance across four different audio features, and compared the performance with baseline AASIST model. Beyond conventional metrics such as Equal Error Rate (EER %), we incorporated five established fairness metrics to quantify gender disparities in the model. Our results show that even when the overall EER difference between genders appears low, fairness-aware evaluation reveals disparities in error distribution that are obscured by aggregate performance measures. These findings demonstrate that reliance on standard metrics is unreliable, whereas fairness metrics provide critical insights into demographic-specific failure modes. This work highlights the importance of fairness-aware evaluation for developing a more equitable, robust, and trustworthy audio deepfake detection system.

LLMs have advanced code generation, but their use for generating microservices with explicit dependencies and API contracts remains understudied. We examine whether AI agents can generate functional microservices and how different forms of contextual information influence their performance. We assess 144 generated microservices across 3 agents, 4 projects, 2 prompting strategies, and 2 scenarios. Incremental generation operates within existing systems and is evaluated with unit tests. Clean state generation starts from requirements alone and is evaluated with integration tests. We analyze functional correctness, code quality, and efficiency. Minimal prompts outperformed detailed ones in incremental generation, with 50-76% unit test pass rates. Clean state generation produced higher integration test pass rates (81-98%), indicating strong API contract adherence. Generated code showed lower complexity than human baselines. Generation times varied widely across agents, averaging 6-16 minutes per service. AI agents can produce microservices with maintainable code, yet inconsistent correctness and reliance on human oversight show that fully autonomous microservice generation is not yet achievable.

Multi-agent artificial intelligence systems or MAS are systems of autonomous agents that exercise delegated tool authority, share persistent memory, and coordinate via inter-agent communication. MAS introduces qualitatively distinct security vulnerabilities from those documented for singular AI models. Existing security and governance frameworks were not designed for these emerging attack surfaces. This study systematically characterizes the threat landscape of MAS and quantitatively evaluates 16 security frameworks for AI against it. A four-phase methodology is proposed: constructing a deep technical knowledge base of production multi-agent architectures; conducting generative AI-assisted threat modeling scoped to MAS cybersecurity risks and validated by domain experts; structuring survey plans at individual-threat granularity; and scoring each framework on a three-point scale against the cybersecurity risks. The risks were organized into 193 distinct main threat items across nine risk categories. The expected minimal average score is 2. No reviewed framework achieves majority coverage of any single category. Non-Determinism (mean score 1.231 across all 16 frameworks) and Data Leakage (1.340) are the most under-addressed domains. The OWASP Agentic Security Initiative leads overall at 65.3\% coverage and in the design phase; the CDAO Generative AI Responsible AI Toolkit leads in development and operational coverage. These results provide the first empirical cross-framework comparison for MAS security and offer evidence-based guidance for framework selection.

Large language models (LLMs) achieve strong reasoning performance through chain-of-thought (CoT) reasoning, yet often generate unnecessarily long reasoning paths that incur high inference cost. Recent self-consistency-based approaches further improve accuracy but require sampling and aggregating multiple reasoning trajectories, leading to substantial additional computational overhead. This paper introduces a confidence-aware decision framework that analyzes a single completed reasoning trajectory to adaptively select between single-path and multi-path reasoning. The framework is trained using sentence-level numeric and linguistic features extracted from intermediate reasoning states in the MedQA dataset and generalizes effectively to MathQA, MedMCQA, and MMLU without additional fine-tuning. Experimental results show that the proposed method maintains accuracy comparable to multi-path baselines while using up to 80\% fewer tokens. These findings demonstrate that reasoning trajectories contain rich signals for uncertainty estimation, enabling a simple, transferable mechanism to balance accuracy and efficiency in LLM reasoning.

System prompts for LLM-based coding agents are software artifacts that govern agent behavior, yet lack the testing infrastructure applied to conventional software. We present Arbiter, a framework combining formal evaluation rules with multi-model LLM scouring to detect interference patterns in system prompts. Applied to three major coding agent system prompts: Claude Code (Anthropic), Codex CLI (OpenAI), and Gemini CLI (Google), we identify 152 findings across the undirected scouring phase and 21 hand-labeled interference patterns in directed analysis of one vendor. We show that prompt architecture (monolithic, flat, modular) strongly correlates with observed failure class but not with severity, and that multi-model evaluation discovers categorically different vulnerability classes than single-model analysis. One scourer finding was structural data loss in Gemini CLI's memory system was consistent with an issue filed and patched by Google, which addressed the symptom without addressing the schema-level root cause identified by the scourer. Total cost of cross-vendor analysis: \$0.27 USD.

Thematic analysis (TA) is widely used in health research to extract patterns from patient interviews, yet manual TA faces challenges in scalability and reproducibility. LLM-based automation can help, but existing approaches produce codebooks with limited generalizability and lack analytic auditability. We present an automated TA framework combining iterative codebook refinement with full provenance tracking. Evaluated on five corpora spanning clinical interviews, social media, and public transcripts, the framework achieves the highest composite quality score on four of five datasets compared to six baselines. Iterative refinement yields statistically significant improvements on four datasets with large effect sizes, driven by gains in code reusability and distributional consistency while preserving descriptive quality. On two clinical corpora (pediatric cardiology), generated themes align with expert-annotated themes.

Recent advances in medical large language models have explored Test-Time Reinforcement Learning (TTRL) to enhance reasoning. However, standard TTRL often relies on majority voting (MV) as a heuristic supervision signal, which can be unreliable in complex medical scenarios where the most frequent reasoning path is not necessarily the clinically correct one. In this work, we propose a novel and unified training paradigm that integrates medical process reward models with TTRL to bridge the gap between test-time scaling (TTS) and parametric model optimization. Specifically, we advance the TTRL framework by replacing the conventional MV with a fine-grained, expert-aligned supervision paradigm using Med-RPM. This integration ensures that reinforcement learning is guided by medical correctness rather than mere consensus, effectively distilling search-based intelligence into the model's parametric memory. Extensive evaluations on four different benchmarks have demonstrated that our developed method consistently and significantly outperforms current TTRL and standalone PRM selection. Our findings establish that transitioning from stochastic heuristics to structured, step-wise rewards is essential for developing reliable and scalable medical AI systems

We consider Markov decision processes (MDPs) with unknown disturbance distribution and address this problem using the robust Markov decision process (RMDP) approach. We construct the empirical distribution of the unknown disturbance distribution and characterize our ambiguity set of distributions as the sublevel set of a nonnegative distance function from the empirical distribution. By connecting the weak convergence of distributions to convergence with respect to the distance function, we prove that the robust optimal value function and the out-of-sample value function converge to the true optimal value function with increasing sample-sizes. We establish that, for finite sample-sizes, the robust optimal value function serves as a high probability upper bound on the out-of-sample value function. We also obtain probabilistic convergence rates, sample complexity bounds, and out-of-distribution performance bounds. The finite sample performance guarantees rely on the distance function satisfying a certain concentration type inequality. Several well-studied distances in the literature meet the requirements imposed on the distance function. We also analyze the data-driven properties of empirical MDPs and demonstrate that, unlike our data-driven RMDPs, empirical MDPs fail to satisfy some of the finite sample performance guarantees.

Internet of Things (IoT) Analytics often involves applying machine learning (ML) models on data streams. In such scenarios, traditional ML paradigms face obstacles related to continuous learning while dealing with concept drifts, temporal dependence, and avoiding forgetting. Moreover, in IoT, different edge devices build up a network. When learning models on those devices, connecting them could be useful in improving performance and reusing others' knowledge. This work proposes Mutual Assisted Learning, a learning paradigm grounded on Vygotsky's popular Sociocultural Theory of Cognitive Development. Each device is autonomous and does not need a central orchestrator. Whenever it degrades its performance due to a concept drift, it asks for assistance from others and decides whether their knowledge is useful for solving the new problem. This way, the number of connections is drastically reduced compared to the classical Federated Learning approaches, where the devices communicate at each training round. Every device is equipped with a Continuous Progressive Neural Network (cPNN) to handle the dynamic nature of data streams. We call this implementation Mutual Assisted cPNN (MAcPNN). To implement it, we allow cPNNs for single data point predictions and apply quantization to reduce the memory footprint. Experimental results prove the effectiveness of MAcPNN in boosting performance on synthetic and real data streams.

AI memory systems increasingly organize knowledge into graph structures -- knowledge graphs, entity relations, community hierarchies -- yet lack a principled mechanism for continuous resolution control: where do the qualitative boundaries between abstraction levels lie, and how should an agent navigate them? We introduce Semantic Level of Detail (SLoD), a framework that answers both questions by defining a continuous zoom operator via heat kernel diffusion on the Poincaré ball $\mathbb{B}^d$. At coarse scales ($σ\to \infty$), diffusion aggregates embeddings into high-level summaries; at fine scales ($σ\to 0$), local semantic detail is preserved. We prove hierarchical coherence with bounded approximation error $O(σ)$ and $(1+\varepsilon)$ distortion for tree-structured hierarchies under Sarkar embedding. Crucially, we show that spectral gaps in the graph Laplacian induce emergent scale boundaries -- scales where the representation undergoes qualitative transitions -- which can be detected automatically without manual resolution parameters. On synthetic hierarchies (HSBM), our boundary scanner recovers planted levels with ARI up to 1.00, with detection degrading gracefully near the information-theoretic Kesten-Stigum threshold. On the full WordNet noun hierarchy (82K synsets), detected boundaries align with true taxonomic depth ($τ= 0.79$), demonstrating that the method discovers meaningful abstraction levels in real-world knowledge graphs without supervision.

Forward reachability analysis is a dominant approach for verifying reach-avoid specifications in neural feedback systems, i.e., dynamical systems controlled by neural networks, and a number of directions have been proposed and studied. In contrast, far less attention has been given to backward reachability analysis for these systems, in part because of the limited scalability of known techniques. In this work, we begin to address this gap by introducing new algorithms for computing both over- and underapproximations of backward reachable sets for nonlinear neural feedback systems. We also describe and implement an integration of these backward reachability techniques with existing ones for forward analysis. We call the resulting algorithm Forward and Backward Reachability Integration for Certification (FaBRIC). We evaluate our algorithms on a representative set of benchmarks and show that they significantly outperform the prior state of the art.

Mixture-of-Experts (MoE) models deliver high quality at low training FLOPs, but this efficiency often vanishes at inference. We identify a double penalty that structurally disadvantages MoE architectures during decoding: first, expert routing fragments microbatches and reduces weight reuse; second, massive resident expert pools reduce high-bandwidth memory (HBM) headroom for the KV cache. This phenomenon, formalized as reuse fragmentation, pushes feed-forward networks (FFNs) into a bandwidth-bound regime, especially at long context lengths. We introduce the $qs$ inequality, a predictive criterion that identifies when MoE is structurally disadvantaged relative to a quality-matched dense model. This criterion unifies sparsity ($s$), the fraction of parameters activated per token, and the quality-equivalence factor ($q$), the size multiplier required for a dense model to match MoE performance. Our evaluation across frontier models including DeepSeek-V3, Qwen3-235B, Grok-1, and Switch-C demonstrates that this fragmentation is a general architectural phenomenon. For DeepSeek-V3 at 128k context, this results in a 4.5x throughput advantage for a quality-matched dense baseline. Crucially, massive architectures like Switch-C can become infeasible on cluster sizes where a quality-matched dense model remains viable. Our results suggest that training-time FLOP efficiency is an incomplete proxy for inference-time performance in long-context serving. They also indicate that MoE may be best viewed as a training-time optimization, with distillation into dense models as a possible path toward inference-efficient deployment.

A \emph{tensor-relational} computation is a relational computation where individual tuples carry vectors, matrices, or higher-dimensional arrays. An advantage of tensor-relational computation is that the overall computation can be executed on top of a relational system, inheriting the system's ability to automatically handle very large inputs with high levels of sparsity while high-performance kernels (such as optimized matrix-matrix multiplication codes) can be used to perform most of the underlying mathematical operations. In this paper, we introduce upper-case-lower-case \texttt{EinSum}, which is a tensor-relational version of the classical Einstein Summation Notation. We study how to automatically rewrite a computation in Einstein Notation into upper-case-lower-case \texttt{EinSum} so that computationally intensive components are executed using efficient numerical kernels, while sparsity is managed relationally.

This survey (re)introduces reinforcement learning methods to economists. The curse of dimensionality limits how far exact dynamic programming can be effectively applied, forcing us to rely on suitably "small" problems or our ability to convert "big" problems into smaller ones. While this reduction has been sufficient for many classical applications, a growing class of economic models resists such reduction. Reinforcement learning algorithms offer a natural, sample-based extension of dynamic programming, extending tractability to problems with high-dimensional states, continuous actions, and strategic interactions. I review the theory connecting classical planning to modern learning algorithms and demonstrate their mechanics through simulated examples in pricing, inventory control, strategic games, and preference elicitation. I also examine the practical vulnerabilities of these algorithms, noting their brittleness, sample inefficiency, sensitivity to hyperparameters, and the absence of global convergence guarantees outside of tabular settings. The successes of reinforcement learning remain strictly bounded by these constraints, as well as a reliance on accurate simulators. When guided by economic structure, reinforcement learning provides a remarkably flexible framework. It stands as an imperfect, but promising, addition to the computational economist's toolkit. A companion survey (Rust and Rawat, 2026b) covers the inverse problem of inferring preferences from observed behavior.

The first 72 hours of a missing-person investigation are critical for successful recovery. Guardian is an end-to-end system designed to support missing-child investigation and early search planning. This paper presents the Guardian LLM Pipeline, a multi-model system in which LLMs are used for intelligent information extraction and processing related to missing-person search operations. The pipeline coordinates end-to-end execution across task-specialized LLM models and invokes a consensus LLM engine that compares multiple model outputs and resolves disagreements. The pipeline is further strengthened by QLoRA-based fine-tuning, using curated datasets. The presented design aligns with prior work on weak supervision and LLM-assisted annotation, emphasizing conservative, auditable use of LLMs as structured extractors and labelers rather than unconstrained end-to-end decision makers.

Qualitative research gives rich insights into the quintessentially human aspects of software engineering as a socio-technical system. Qualitative research spans diverse strategies and methods, from interpretivist, in situ observational field studies, to deductive coding of data from mining studies. Advances in large language models and generative AI (GenAI) have prompted claims that artificial intelligence could automate qualitative analysis. Such claims are overgeneralizing from narrow successes. GenAI support must be carefully adapted to the data of interest, but also to the characteristics of a particular research strategy. In this Frontiers of SE paper, we discuss the emerging use of GenAI in relation to the broad spectrum of qualitative research in software engineering. We outline the dimensions of qualitative work in software engineering, review emerging empirical evidence for GenAI assistance, examine the pros and cons of GenAI-mediated qualitative research practices, and revisit qualitative research quality factors, in light of GenAI. Our goal is to inform researchers about the promises and pitfalls of GenAI-assisted qualitative research. We conclude with future plans to advance understanding of its use in software engineering.

Scientific knowledge expands by observing the world, hypothesizing some theories about it, and testing them against collected data. When those theories take the form of statistical models, statistical analyses are involved in the process of testing and refining scientific hypotheses. In this thesis, we focus on statistical models that take the form of scientific simulators and provide background about how machine learning can be used for statistical analyses in this context. The first part of this thesis is about showing empirically that performing statistical analyses with machine learning involves a degree of approximation. Specifically, all statistical analyses involve a level of uncertainty in the conclusions drawn, and we show that approximations can lead to overconfident conclusions. We draw caution regarding such overconfident conclusions and introduce a criterion to diagnose overconfident approximations. In the second part, we introduce balancing, a way to regularize machine learning models to reduce overconfidence and favor calibrated or underconfident approximations. Balancing is first introduced for neural ratio estimation algorithms and then extended to other algorithms. Intuition about why balancing leads to less overconfident solutions is provided, and it is shown empirically that balanced algorithms are often either close to calibrated or underconfident. The third part shows that Bayesian neural networks can also be used to mitigate the overconfidence of approximations. Unlike balancing, no regularization is required, and this solution can then work with few training samples and, hence, computationally expensive simulators. To that end, a new Bayesian neural network prior tailored for simulation-based inference is developed, and empirical results show a reduction in overconfidence compared to similar solutions without Bayesian neural networks.

We propose ULFS-KDPE, a kernel debiased plug-in estimator based on the universal least favorable submodel, for estimating pathwise differentiable parameters in nonparametric models. The method constructs a data-adaptive debiasing flow in a reproducing kernel Hilbert space (RKHS), producing a plug-in estimator that achieves semiparametric efficiency without requiring explicit derivation or evaluation of efficient influence functions. We place ULFS-KDPE on a rigorous functional-analytic foundation by formulating the universal least favorable update as a nonlinear ordinary differential equation on probability densities. We establish existence, uniqueness, stability, and finite-time convergence of the empirical score along the induced flow. Under standard regularity conditions, the resulting estimator is regular, asymptotically linear, and attains the semiparametric efficiency bound simultaneously for a broad class of pathwise differentiable parameters. The method admits a computationally tractable implementation based on finite-dimensional kernel representations and principled stopping criteria. In finite samples, the combination of solving a rich collection of score equations with RKHS-based smoothing and avoidance of direct influence-function evaluation leads to improved numerical stability. Simulation studies illustrate the method and support the theoretical results.

Recent advances in vision-language models (VLMs) have demonstrated remarkable zero-shot capabilities, yet adapting these models to specialized domains remains a significant challenge. Building on recent theoretical insights suggesting that independently trained VLMs are related by a canonical transformation, we extend this understanding to the concept of domains. We hypothesize that image features across disparate domains are related by a canonicalized geometric transformation that can be recovered using a small set of anchors. Few-shot classification provides a natural setting for this alignment, as the limited labeled samples serve as the anchors required to estimate this transformation. Motivated by this hypothesis, we introduce BiCLIP, a framework that applies a targeted transformation to multimodal features to enhance cross-modal alignment. Our approach is characterized by its extreme simplicity and low parameter footprint. Extensive evaluations across 11 standard benchmarks, including EuroSAT, DTD, and FGVCAircraft, demonstrate that BiCLIP consistently achieves state-of-the-art results. Furthermore, we provide empirical verification of existing geometric findings by analyzing the orthogonality and angular distribution of the learned transformations, confirming that structured alignment is the key to robust domain adaptation. Code is available at https://github.com/QuantitativeImagingLaboratory/BilinearCLIP

The rapid emergence of open-source, locally hosted intelligent agents marks a critical inflection point in human-computer interaction. Systems such as OpenClaw demonstrate that Large Language Model (LLM)-based agents can autonomously operate local computing environments, orchestrate workflows, and integrate external tools. However, within the current paradigm, these agents remain conventional applications running on legacy operating systems originally designed for Graphical User Interfaces (GUIs) or Command Line Interfaces (CLIs). This architectural mismatch leads to fragmented interaction models, poorly structured permission management (often described as "Shadow AI"), and severe context fragmentation. This paper proposes a new paradigm: a Personal Agent Operating System (AgentOS). In AgentOS, traditional GUI desktops are replaced by a Natural User Interface (NUI) centered on a unified natural language or voice portal. The system core becomes an Agent Kernel that interprets user intent, decomposes tasks, and coordinates multiple agents, while traditional applications evolve into modular Skills-as-Modules enabling users to compose software through natural language rules. We argue that realizing AgentOS fundamentally becomes a Knowledge Discovery and Data Mining (KDD) problem. The Agent Kernel must operate as a real-time engine for intent mining and knowledge discovery. Viewed through this lens, the operating system becomes a continuous data mining pipeline involving sequential pattern mining for workflow automation, recommender systems for skill retrieval, and dynamically evolving personal knowledge graphs. These challenges define a new research agenda for the KDD community in building the next generation of intelligent computing systems.

Speech Large Language Models (LLMs) show great promise for speech emotion recognition (SER) via generative interfaces. However, shifting from closed-set classification to open text generation introduces zero-shot stochasticity, making evaluation highly sensitive to prompts. Additionally, conventional speech LLMs benchmarks overlook the inherent ambiguity of human emotion. Hence, we present VoxEmo, a comprehensive SER benchmark encompassing 35 emotion corpora across 15 languages for Speech LLMs. VoxEmo provides a standardized toolkit featuring varying prompt complexities, from direct classification to paralinguistic reasoning. To reflect real-world perception/application, we introduce a distribution-aware soft-label protocol and a prompt-ensemble strategy that emulates annotator disagreement. Experiments reveal that while zero-shot speech LLMs trail supervised baselines in hard-label accuracy, they uniquely align with human subjective distributions.

Pathology underpins modern diagnosis and cancer care, yet its most valuable asset, the accumulated experience encoded in millions of narrative reports, remains largely inaccessible. Although institutions are rapidly digitizing pathology workflows, storing data without effective mechanisms for retrieval and reasoning risks transforming archives into a passive data repository, where institutional knowledge exists but cannot meaningfully inform patient care. True progress requires not only digitization, but the ability for pathologists to interrogate prior similar cases in real time while evaluating a new diagnostic dilemma. We present PathoScribe, a unified retrieval-augmented large language model (LLM) framework designed to transform static pathology archives into a searchable, reasoning-enabled living library. PathoScribe enables natural language case exploration, automated cohort construction, clinical question answering, immunohistochemistry (IHC) panel recommendation, and prompt-controlled report transformation within a single architecture. Evaluated on 70,000 multi-institutional surgical pathology reports, PathoScribe achieved perfect Recall@10 for natural language case retrieval and demonstrated high-quality retrieval-grounded reasoning (mean reviewer score 4.56/5). Critically, the system operationalized automated cohort construction from free-text eligibility criteria, assembling research-ready cohorts in minutes (mean 9.2 minutes) with 91.3% agreement to human reviewers and no eligible cases incorrectly excluded, representing orders-of-magnitude reductions in time and cost compared to traditional manual chart review. This work establishes a scalable foundation for converting digital pathology archives from passive storage systems into active clinical intelligence platforms.

The first 72 hours of a missing-child investigation are critical for successful recovery. However, law enforcement agencies often face fragmented, unstructured data and a lack of dynamic, geospatial predictive tools. Our system, Guardian, provides an end-to-end decision-support system for missing-child investigation and early search planning. It converts heterogeneous, unstructured case documents into a schema-aligned spatiotemporal representation, enriches cases with geocoding and transportation context, and provides probabilistic search products spanning 0-72 hours. In this paper, we present an overview of Guardian as well as a detailed description of a three-layer predictive component of the system. The first layer is a Markov chain, a sparse, interpretable model with transitions incorporating road accessibility costs, seclusion preferences, and corridor bias with separate day/night parameterizations. The Markov chain's output prediction distributions are then transformed into operationally useful search plans by the second layer's reinforcement learning. Finally, the third layer's LLM performs post hoc validation of layer 2 search plans prior to their release. Using a synthetic but realistic case study, we report quantitative outputs across 24/48/72-hour horizons and analyze sensitivity, failure modes, and tradeoffs. Results show that the proposed predictive system with the three-layer architecture produces interpretable priors for zone optimization and human review.

In this paper, we investigate a novel digital network twin (DNT) assisted deep learning (DL) model training framework. In particular, we consider a physical network where a base station (BS) uses several antennas to serve multiple mobile users, and a DNT that is a virtual representation of the physical network. The BS must adjust its antenna tilt angles to optimize the data rates of all users. Due to user mobility, the BS may not be able to accurately track network dynamics such as wireless channels and user mobilities. Hence, a reinforcement learning (RL) approach is used to dynamically adjust the antenna tilt angles. To train the RL, we can use data collected from the physical network and the DNT. The data collected from the physical network is more accurate but incurs more communication overhead compared to the data collected from the DNT. Therefore, it is necessary to determine the ratio of data collected from the physical network and the DNT to improve the training of the RL model. We formulate this problem as an optimization problem whose goal is to jointly optimize the tilt angle adjustment policy and the data collection strategy, aiming to maximize the data rates of all users while constraining the time delay introduced by collecting data from the physical network. To solve this problem, we propose a hierarchical RL framework that integrates robust adversarial loss and proximal policy optimization (PPO). Simulation results show that our proposed method reduces the physical network data collection delay by up to 28.01% and 1x compared to a hierarchical RL that uses vanilla PPO as the first level RL, and the baseline that uses robust-RL at the first level and selects the data collection ratio randomly.

This paper introduces a synthetic benchmark to evaluate the performance of vision language models (VLMs) in generating plant simulation configurations for digital twins. While functional-structural plant models (FSPMs) are useful tools for simulating biophysical processes in agricultural environments, their high complexity and low throughput create bottlenecks for deployment at scale. We propose a novel approach that leverages state-of-the-art open-source VLMs -- Gemma 3 and Qwen3-VL -- to directly generate simulation parameters in JSON format from drone-based remote sensing images. Using a synthetic cowpea plot dataset generated via the Helios 3D procedural plant generation library, we tested five in-context learning methods and evaluated the models across three categories: JSON integrity, geometric evaluations, and biophysical evaluations. Our results show that while VLMs can interpret structural metadata and estimate parameters like plant count and sun azimuth, they often exhibit performance degradation due to contextual bias or rely on dataset means when visual cues are insufficient. Validation on a real-world drone orthophoto dataset and an ablation study using a blind baseline further characterize the models' reasoning capabilities versus their reliance on contextual priors. To the best of our knowledge, this is the first study to utilize VLMs to generate structural JSON configurations for plant simulations, providing a scalable framework for reconstruction 3D plots for digital twin in agriculture.

AI-powered answer engines are inherently non-deterministic: identical queries submitted at different times can produce different responses and cite different sources. Despite this stochastic behavior, current approaches to measuring domain visibility in generative search typically rely on single-run point estimates of citation share and prevalence, implicitly treating them as fixed values. This paper argues that citation visibility metrics should be treated as sample estimators of an underlying response distribution rather than fixed values. We conduct an empirical study of citation variability across three generative search platforms--Perplexity Search, OpenAI SearchGPT, and Google Gemini--using repeated sampling across three consumer product topics. Two sampling regimes are employed: daily collections over nine days and high-frequency sampling at ten-minute intervals. We show that citation distributions follow a power-law form and exhibit substantial variability across repeated samples. Bootstrap confidence intervals reveal that many apparent differences between domains fall within the noise floor of the measurement process. Distribution-wide rank stability analysis further demonstrates that citation rankings are unstable across samples, not only among top-ranked domains but throughout the frequently cited domain set. These findings demonstrate that single-run visibility metrics provide a misleadingly precise picture of domain performance in generative search. We argue that citation visibility must be reported with uncertainty estimates and provide practical guidance for sample sizes required to achieve interpretable confidence intervals.

Concept Bottleneck Models (CBMs) are a prominent framework for interpretable AI that map learned visual features to a set of meaningful concepts for task-specific downstream predictions. Their sequential structure enhances transparency by connecting model predictions to the underlying concepts that support them. In medical imaging, where transparency is essential, CBMs offer an appealing foundation for explainable model design. However, discrete concept representations often overlook broader clinical context such as diagnostic guidelines and expert heuristics, reducing reliability in complex cases. We propose MedCBR, a concept-based reasoning framework that integrates clinical guidelines with vision-language and reasoning models. Labeled clinical descriptors are transformed into guideline-conformant text, and a concept-based model is trained with a multitask objective combining multimodal contrastive alignment, concept supervision, and diagnostic classification to jointly ground image features, concepts, and pathology. A reasoning model then converts these predictions into structured clinical narratives that explain the diagnosis, emulating expert reasoning based on established guidelines. MedCBR achieves superior diagnostic and concept-level performance, with AUROCs of 94.2% on ultrasound and 84.0% on mammography. Further experiments on non-medical datasets achieve 86.1% accuracy. Our framework enhances interpretability and forms an end-to-end bridge from medical image analysis to decision-making.

Over-parameterized neural networks incur prohibitive memory and computational costs for resource-constrained deployment. The Strong Lottery Ticket (SLT) hypothesis suggests that randomly initialized networks contain sparse subnetworks achieving competitive accuracy without weight training. Existing SLT methods, notably edge-popup, rely on non-differentiable score-based selection, limiting optimization efficiency and scalability. We propose using continuously relaxed Bernoulli gates to discover SLTs through fully differentiable, end-to-end optimization - training only gating parameters while keeping all network weights frozen at their initialized values. Continuous relaxation enables direct gradient-based optimization of an $\ell_0$-regularization objective, eliminating the need for non-differentiable gradient estimators or iterative pruning cycles. To our knowledge, this is the first fully differentiable approach for SLT discovery that avoids straight-through estimator approximations. Experiments across fully connected networks, CNNs (ResNet, Wide-ResNet), and Vision Transformers (ViT, Swin-T) demonstrate up to 90% sparsity with minimal accuracy loss - nearly double the sparsity achieved by edge-popup at comparable accuracy - establishing a scalable framework for pre-training network sparsification.

Genomic language models (GLMs) have emerged as powerful tools for learning representations of DNA sequences, enabling advances in variant prediction, regulatory element identification, and cross-task transfer learning. However, as these models are increasingly trained or fine-tuned on sensitive genomic cohorts, they risk memorizing specific sequences from their training data, raising serious concerns around privacy, data leakage, and regulatory compliance. Despite growing awareness of memorization risks in general-purpose language models, little systematic evaluation exists for these risks in the genomic domain, where data exhibit unique properties such as a fixed nucleotide alphabet, strong biological structure, and individual identifiability. We present a comprehensive, multi-vector privacy evaluation framework designed to quantify memorization risks in GLMs. Our approach integrates three complementary risk assessment methodologies: perplexity-based detection, canary sequence extraction, and membership inference. These are combined into a unified evaluation pipeline that produces a worst-case memorization risk score. To enable controlled evaluation, we plant canary sequences at varying repetition rates into both synthetic and real genomic datasets, allowing precise quantification of how repetition and training dynamics influence memorization. We evaluate our framework across multiple GLM architectures, examining the relationship between sequence repetition, model capacity, and memorization risk. Our results establish that GLMs exhibit measurable memorization and that the degree of memorization varies across architectures and training regimes. These findings reveal that no single attack vector captures the full scope of memorization risk, underscoring the need for multi-vector privacy auditing as a standard practice for genomic AI systems.

Federated Learning (FL) enables distributed Artificial Intelligence (AI) across cloud-edge environments by allowing collaborative model training without centralizing data. In cross-device deployments, FL systems face strict communication and participation constraints, as well as strong non-independent and identically distributed (non-IID) data that degrades convergence and model quality. Since only a subset of devices (a.k.a clients) can participate per training round, intelligent client selection becomes a key systems challenge. This paper proposes FedLECC (Federated Learning with Enhanced Cluster Choice), a lightweight, cluster-aware, and loss-guided client selection strategy for cross-device FL. FedLECC groups clients by label-distribution similarity and prioritizes clusters and clients with higher local loss, enabling the selection of a small yet informative and diverse set of clients. Experimental results under severe label skew show that FedLECC improves test accuracy by up to 12%, while reducing communication rounds by approximately 22% and overall communication overhead by up to 50% compared to strong baselines. These results demonstrate that informed client selection improves the efficiency and scalability of FL workloads in cloud-edge systems.

We introduce SciTaRC, an expert-authored benchmark of questions about tabular data in scientific papers requiring both deep language reasoning and complex computation. We show that current state-of-the-art AI models fail on at least 23% of these questions, a gap that remains significant even for highly capable open-weight models like Llama-3.3-70B-Instruct, which fails on 65.5% of the tasks. Our analysis reveals a universal "execution bottleneck": both code and language models struggle to faithfully execute plans, even when provided with correct strategies. Specifically, code-based methods prove brittle on raw scientific tables, while natural language reasoning primarily fails due to initial comprehension issues and calculation errors.

We present a comprehensive ablation of nine finite-sample bound families for selective prediction with risk control, combining concentration inequalities (Hoeffding, Empirical Bernstein, Clopper-Pearson, Wasserstein DRO, CVaR) with multiple-testing corrections (union bound, Learn Then Test fixed-sequence) and betting-based confidence sequences (WSR). Our main theoretical contribution is Transfer-Informed Betting (TIB), which warm-starts the WSR wealth process using a source domain's risk profile, achieving tighter bounds in data-scarce settings with a formal dominance guarantee. We prove that the TIB wealth process remains a valid supermartingale under all source-target divergences, that TIB dominates standard WSR when domains match, and that no data-independent warm-start can achieve better convergence. The combination of betting-based confidence sequences, LTT monotone testing, and cross-domain transfer is, to our knowledge, a three-way novelty not present in the literature. We evaluate all nine bound families on four benchmarks-MASSIVE (n=1,102), NyayaBench (n=280), CLINC-150 (n=22.5K), and Banking77 (n=13K)-across 18 (alpha, delta) configurations. On MASSIVE at alpha=0.10, LTT eliminates the ln(K) union-bound penalty, achieving 94.0% guaranteed coverage versus 73.8% for Hoeffding-a 27% relative improvement. On NyayaBench, where the small calibration set makes Hoeffding-family bounds infeasible below alpha=0.20, Transfer-Informed Betting achieves 18.5% coverage at alpha=0.10, a 5.4x improvement over LTT + Hoeffding. We additionally compare with split-conformal prediction, showing that conformal methods produce prediction sets (avg. 1.67 classes) whereas selective prediction provides single-prediction risk guarantees. We apply these methods to agentic caching systems, formalizing a progressive trust model where the guarantee determines when cached responses can be served autonomously.

Deep learning (DL)-based Network Intrusion Detection System (NIDS) has demonstrated great promise in detecting malicious network traffic. However, they face significant security risks due to their vulnerability to adversarial examples (AEs). Most existing adversarial attacks maliciously perturb data to maximize misclassification errors. Among AEs, natural adversarial examples (NAEs) are particularly difficult to detect because they closely resemble real data, making them challenging for both humans and machine learning models to distinguish from legitimate inputs. Creating NAEs is crucial for testing and strengthening NIDS defenses. This paper proposes NetDiffuser1, a novel framework for generating NAEs capable of deceiving NIDS. NetDiffuser consists of two novel components. First, a new feature categorization algorithm is designed to identify relatively independent features in network traffic. Perturbing these features minimizes changes while preserving network flow validity. The second component is a novel application of diffusion models to inject semantically consistent perturbations for generating NAEs. NetDiffuser performance was extensively evaluated using three benchmark NIDS datasets across various model architectures and state-of-the-art adversarial detectors. Our experimental results show that NetDiffuser achieves up to a 29.93% higher attack success rate and reduces AE detection performance by at least 0.267 (in some cases up to 0.534) in the Area under the Receiver Operating Characteristic Curve (AUC-ROC) score compared to the baseline attacks.

Considering the high volume, wide variety, and rapid speed of data generation, investigating feature selection methods for big data presents various applications and advantages. By removing irrelevant and redundant features, feature selection reduces data dimensions, thereby facilitating optimal decision-making within decision systems. One of the key tools for feature selection in hybrid information systems is fuzzy rough set theory. However, this theory faces two significant challenges: First, obtaining fuzzy equivalence relations through intersection operations in high-dimensional spaces can be both time-consuming and memory-intensive. Additionally, this method may produce noisy data, complicating the feature selection process. The purpose and innovation of this paper are to address these issues. We proposed a new feature selection model that calculates the combined distance between objects and subsequently used this information to derive the fuzzy equivalence relation. Rather than directly solving the feature selection problem, this approach reformulates it into an optimization problem that can be tackled using appropriate meta-heuristic algorithms. We have named this new approach FSbuHD. The FSbuHD model operates in two modes - normal and optimistic - based on the selection of one of the two introduced fuzzy equivalence relations. The model is then tested on standard datasets from the UCI repository and compared with other algorithms. The results of this research demonstrate that FSbuHD is one of the most efficient and effective methods for feature selection when compared to previous methods and algorithms.

Speculative decoding has emerged as a powerful approach to accelerate large language model (LLM) inference by employing lightweight draft models to propose candidate tokens that are subsequently verified by the target model. The effectiveness of this paradigm critically depends on the quality of the draft model. While recent advances such as the EAGLE series achieve state-of-the-art speedup, existing draft models remain limited by error accumulation: they condition only on the current prefix, causing their predictions to drift from the target model over steps. In this work, we propose \textbf{ConFu} (Contemplate the Future), a novel speculative decoding framework that enables draft models to anticipate the future direction of generation. ConFu introduces (i) contemplate tokens and soft prompts that allow the draft model to leverage future-oriented signals from the target model at negligible cost, (ii) a dynamic contemplate token mechanism with MoE to enable context-aware future prediction, and (iii) a training framework with anchor token sampling and future prediction replication that learns robust future prediction. Experiments demonstrate that ConFu improves token acceptance rates and generation speed over EAGLE-3 by 8--11% across various downstream tasks with Llama-3 3B and 8B models. We believe our work is the first to bridge speculative decoding with continuous reasoning tokens, offering a new direction for accelerating LLM inference.

Compositionally complex solid solution electrocatalysts span vast composition spaces, and even one materials system can contain more candidate compositions than can be measured exhaustively. Here we evaluate a label-free screening strategy that represents each composition using embeddings derived from scientific texts and prioritizes candidates based on similarity to two property concepts. We compare a corpus-trained Word2Vec baseline with transformer-based embeddings, where compositions are encoded either by linear element-wise mixing or by short composition prompts. Similarities to `concept directions', the terms conductivity and dielectric, define a 2-dimensional descriptor space, and a symmetric Pareto-front selection is used to filter candidate subsets without using electrochemical labels. Performance is assessed on 15 materials libraries including noble metal alloys and multicomponent oxides. In this setting, the lightweight Word2Vec baseline, which uses a simple linear combination of element embeddings, often achieves the highest number of reductions of possible candidate compositions while staying close to the best measured performance.

Accessing sensitive patient data for machine learning is challenging due to privacy concerns. Datasets with annotations of personally identifiable information are crucial for developing and testing anonymization systems to enable safe data sharing that complies with privacy regulations. Since accessing real patient data is a bottleneck, synthetic data offers an efficient solution for data scarcity, bypassing privacy regulations that apply to real data. Moreover, neural machine translation can help to create high-quality data for low-resource languages by translating validated real or synthetic data from a high-resource language. In this work, we create a multilingual anonymization benchmark in ten languages, using a machine translation methodology that preserves the original annotations and renders names of cities and people in a culturally and contextually appropriate form in each target language. Our evaluation study with medical professionals confirms the quality of the translations, both in general and with respect to the translation and adaptation of personal information. Our benchmark with over 2,500 annotations of personal information can be used in many applications, including training annotators, validating annotations across institutions without legal complications, and helping improve the performance of automatic personal information detection. We make our benchmark and annotation guidelines available for further research.

Agentic Retrieval-Augmented Generation (RAG) systems combine iterative search, planning prompts, and retrieval backends, but deployed settings impose explicit budgets on tool calls and completion tokens. We present a controlled measurement study of how search depth, retrieval strategy, and completion budget affect accuracy and cost under fixed constraints. Using Budget-Constrained Agentic Search (BCAS), a model-agnostic evaluation harness that surfaces remaining budget and gates tool use, we run comparisons across six LLMs and three question-answering benchmarks. Across models and datasets, accuracy improves with additional searches up to a small cap, hybrid lexical and dense retrieval with lightweight re-ranking produces the largest average gains in our ablation grid, and larger completion budgets are most helpful on HotpotQA-style synthesis. These results provide practical guidance for configuring budgeted agentic retrieval pipelines and are accompanied by reproducible prompts and evaluation settings.

Do the features learned by Sparse Autoencoders (SAEs) represent abstract meaning, or are they tied to how text is written? We investigate this question using Serbian digraphia as a controlled testbed: Serbian is written interchangeably in Latin and Cyrillic scripts with a near-perfect character mapping between them, enabling us to vary orthography while holding meaning exactly constant. Crucially, these scripts are tokenized completely differently, sharing no tokens whatsoever. Analyzing SAE feature activations across the Gemma model family (270M-27B parameters), we find that identical sentences in different Serbian scripts activate highly overlapping features, far exceeding random baselines. Strikingly, changing script causes less representational divergence than paraphrasing within the same script, suggesting SAE features prioritize meaning over orthographic form. Cross-script cross-paraphrase comparisons provide evidence against memorization, as these combinations rarely co-occur in training data yet still exhibit substantial feature overlap. This script invariance strengthens with model scale. Taken together, our findings suggest that SAE features can capture semantics at a level of abstraction above surface tokenization, and we propose Serbian digraphia as a general evaluation paradigm for probing the abstractness of learned representations.

Communication-aware robot planning requires accurate predictions of wireless network performance. Current approaches rely on channel-level metrics such as received signal strength and signal-to-noise ratio, assuming these translate reliably into end-to-end throughput. We challenge this assumption through a measurement campaign in a private 5G industrial environment. We evaluate throughput predictions from a commercial ray-tracing simulator as well as data-driven Gaussian process regression models against measurements collected using a mobile robot. The study uses off-the-shelf user equipment in an underground, radio-shielded facility with detailed 3D modeling, representing a best-case scenario for prediction accuracy. The ray-tracing simulator captures the spatial structure of indoor propagation and predicts channel-level metrics with reasonable fidelity. However, it systematically over-predicts throughput, even in line-of-sight regions. The dominant error source is shown to be over-estimation of sustainable MIMO spatial layers: the simulator assumes near-uniform four-layer transmission while measurements reveal substantial adaptation between one and three layers. This mismatch inflates predicted throughput even when channel metrics appear accurate. In contrast, a Gaussian process model with a rational quadratic kernel achieves approximately two-thirds reduction in prediction error with near-zero bias by learning end-to-end throughput directly from measurements. These findings demonstrate that favorable channel conditions do not guarantee high throughput; communication-aware planners relying solely on channel-centric predictions risk overly optimistic trajectories that violate reliability requirements. Accurate throughput prediction for 5G systems requires either extensive calibration of link-layer models or data-driven approaches that capture real system behavior.

Autonomous navigation in highly constrained environments remains challenging for mobile robots. Classical navigation approaches offer safety assurances but require environment-specific parameter tuning; end-to-end learning bypasses parameter tuning but struggles with precise control in constrained spaces. To this end, recent robot learning approaches automate parameter tuning while retaining classical systems' safety, yet still face challenges in generalizing to unseen environments. Recently, Vision-Language-Action (VLA) models have shown promise by leveraging foundation models' scene understanding capabilities, but still struggle with precise control and inference latency in navigation tasks. In this paper, we propose Adaptive Planner Parameter Learning from Vision-Language-Action Model (\textsc{applv}). Unlike traditional VLA models that directly output actions, \textsc{applv} leverages pre-trained vision-language models with a regression head to predict planner parameters that configure classical planners. We develop two training strategies: supervised learning fine-tuning from collected navigation trajectories and reinforcement learning fine-tuning to further optimize navigation performance. We evaluate \textsc{applv} across multiple motion planners on the simulated Benchmark Autonomous Robot Navigation (BARN) dataset and in physical robot experiments. Results demonstrate that \textsc{applv} outperforms existing methods in both navigation performance and generalization to unseen environments.

Hybrid sequence models--combining Transformer and state-space model layers--seek to gain the expressive versatility of attention as well as the computational efficiency of state-space model layers. Despite burgeoning interest in hybrid models, we lack a basic understanding of the settings where--and underlying mechanisms through which--they offer benefits over their constituent models. In this paper, we study this question, focusing on a broad family of core synthetic tasks. For this family of tasks, we prove the existence of fundamental limitations for non-hybrid models. Specifically, any Transformer or state-space model that solves the underlying task requires either a large number of parameters or a large working memory. On the other hand, for two prototypical tasks within this family--namely selective copying and associative recall--we construct hybrid models of small size and working memory that provably solve these tasks, thus achieving the best of both worlds. Our experimental evaluation empirically validates our theoretical findings. Importantly, going beyond the settings in our theoretical analysis, we empirically show that learned--rather than constructed--hybrids outperform non-hybrid models with up to 6x as many parameters. We additionally demonstrate that hybrid models exhibit stronger length generalization and out-of-distribution robustness than non-hybrids.

Algorithmic support systems often return optimal solutions that are hard to understand. Effective human-algorithm collaboration, however, requires interpretability. When machine solutions are equally optimal, humans must select one, but a precise account of what makes one solution more interpretable than another remains missing. To identify structural properties of interpretable machine solutions, we present an experimental paradigm in which participants chose which of two equally optimal solutions for packing items into bins was easier to understand. We show that preferences reliably track three quantifiable properties of solution structure: alignment with a greedy heuristic, simple within-bin composition, and ordered visual representation. The strongest associations were observed for ordered representations and heuristic alignment, with compositional simplicity also showing a consistent association. Reaction-time evidence was mixed, with faster responses observed primarily when heuristic differences were larger, and aggregate webcam-based gaze did not show reliable effects of complexity. These results provide a concrete, feature-based account of interpretability in optimal packing solutions, linking solution structure to human preference. By identifying actionable properties (simple compositions, ordered representation, and heuristic alignment), our findings enable interpretability-aware optimization and presentation of machine solutions, and outline a path to quantify trade-offs between optimality and interpretability in real-world allocation and design tasks.

As multi-agent AI systems grow in complexity, the protocols connecting them constrain their capabilities. Current protocols such as A2A and MCP do not expose model-level properties as first-class primitives, ignoring properties fundamental to effective delegation: model identity, reasoning profile, quality calibration, and cost characteristics. We present the LLM Delegate Protocol (LDP), an AI-native communication protocol introducing five mechanisms: (1) rich delegate identity cards with quality hints and reasoning profiles; (2) progressive payload modes with negotiation and fallback; (3) governed sessions with persistent context; (4) structured provenance tracking confidence and verification status; (5) trust domains enforcing security boundaries at the protocol level. We implement LDP as a plugin for the JamJet agent runtime and evaluate against A2A and random baselines using local Ollama models and LLM-as-judge evaluation. Identity-aware routing achieves ~12x lower latency on easy tasks through delegate specialization, though it does not improve aggregate quality in our small delegate pool; semantic frame payloads reduce token count by 37% (p=0.031) with no observed quality loss; governed sessions eliminate 39% token overhead at 10 rounds; and noisy provenance degrades synthesis quality below the no-provenance baseline, arguing that confidence metadata is harmful without verification. Simulated analyses show architectural advantages in attack detection (96% vs. 6%) and failure recovery (100% vs. 35% completion). This paper contributes a protocol design, reference implementation, and initial evidence that AI-native protocol primitives enable more efficient and governable delegation.

Accurate localization of tumor regions from hematoxylin and eosin-stained whole-slide images is fundamental for translational research including spatial analysis, molecular profiling, and tissue architecture investigation. However, deep learning-based tumor detection trained within specific cancers may exhibit reduced robustness when applied across different tumor types. We investigated whether balanced training across cancers at modest scale can achieve high performance and generalize to unseen tumor types. A multi-cancer tumor localization model (MuCTaL) was trained on 79,984 non-overlapping tiles from four cancers (melanoma, hepatocellular carcinoma, colorectal cancer, and non-small cell lung cancer) using transfer learning with DenseNet169. The model achieved a tile-level ROC-AUC of 0.97 in validation data from the four training cancers, and 0.71 on an independent pancreatic ductal adenocarcinoma cohort. A scalable inference workflow was built to generate spatial tumor probability heatmaps compatible with existing digital pathology tools. Code and models are publicly available at https://github.com/AivaraX-AI/MuCTaL.

The rapid adoption of LLM-based agentic systems has produced a rich ecosystem of frameworks (smolagents, LangGraph, AutoGen, CAMEL, LlamaIndex, i.a.). Yet existing benchmarks are model-centric: they fix the agentic setup and do not compare other system components. We argue that implementation decisions substantially impact performance, including choices such as topology, orchestration logic, and error handling. MASEval addresses this evaluation gap with a framework-agnostic library that treats the entire system as the unit of analysis. Through a systematic system-level comparison across 3 benchmarks, 3 models, and 3 frameworks, we find that framework choice matters as much as model choice. MASEval allows researchers to explore all components of agentic systems, opening new avenues for principled system design, and practitioners to identify the best implementation for their use case. MASEval is available under the MIT licence https://github.com/parameterlab/MASEval.

Discrete graph generation has emerged as a powerful paradigm for modeling graph data, often relying on highly expressive neural backbones such as transformers or higher-order architectures. We revisit this design choice by introducing GenGNN, a modular message-passing framework for graph generation. Diffusion models with GenGNN achieve more than 90% validity on Tree and Planar datasets, within margins of graph transformers, at 2-5x faster inference speed. For molecule generation, DiGress with a GenGNN backbone achieves 99.49% Validity. A systematic ablation study shows the benefit provided by each GenGNN component, indicating the need for residual connections to mitigate oversmoothing on complicated graph-structure. Through scaling analyses, we apply a principled metric-space view to investigate learned diffusion representations and uncover whether GNNs can be expressive neural backbones for discrete diffusion.

Automatic differentiation (AD) frameworks such as JAX and PyTorch have enabled gradient-based optimization for a wide range of scientific fields. Yet, many "hard" primitives in these libraries such as thresholding, Boolean logic, discrete indexing, and sorting operations yield zero or undefined gradients that are not useful for optimization. While numerous "soft" relaxations have been proposed that provide informative gradients, the respective implementations are fragmented across projects, making them difficult to combine and compare. This work introduces SoftJAX and SoftTorch, open-source, feature-complete libraries for soft differentiable programming. These libraries provide a variety of soft functions as drop-in replacements for their hard JAX and PyTorch counterparts. This includes (i) elementwise operators such as clip or abs, (ii) utility methods for manipulating Booleans and indices via fuzzy logic, (iii) axiswise operators such as sort or rank -- based on optimal transport or permutahedron projections, and (iv) offer full support for straight-through gradient estimation. Overall, SoftJAX and SoftTorch make the toolbox of soft relaxations easily accessible to differentiable programming, as demonstrated through benchmarking and a practical case study. Code is available at github.com/a-paulus/softjax and github.com/a-paulus/softtorch.

We introduce Fish Audio S2, an open-sourced text-to-speech system featuring multi-speaker, multi-turn generation, and, most importantly, instruction-following control via natural-language descriptions. To scale training, we develop a multi-stage training recipe together with a staged data pipeline covering video captioning and speech captioning, voice-quality assessment, and reward modeling. To push the frontier of open-source TTS, we release our model weights, fine-tuning code, and an SGLang-based inference engine. The inference engine is production-ready for streaming, achieving an RTF of 0.195 and a time-to-first-audio below 100 ms.Our code and weights are available on GitHub (https://github.com/fishaudio/fish-speech) and Hugging Face (https://huggingface.co/fishaudio/s2-pro). We highly encourage readers to visit https://fish.audio to try custom voices.

Retrieval-augmented generation (RAG) systems combine document retrieval with a generative model to address complex information seeking tasks like report generation. While the relationship between retrieval quality and generation effectiveness seems intuitive, it has not been systematically studied. We investigate whether upstream retrieval metrics can serve as reliable early indicators of the final generated response's information coverage. Through experiments across two text RAG benchmarks (TREC NeuCLIR 2024 and TREC RAG 2024) and one multimodal benchmark (WikiVideo), we analyze 15 text retrieval stacks and 10 multimodal retrieval stacks across four RAG pipelines and multiple evaluation frameworks (Auto-ARGUE and MiRAGE). Our findings demonstrate strong correlations between coverage-based retrieval metrics and nugget coverage in generated responses at both topic and system levels. This relationship holds most strongly when retrieval objectives align with generation goals, though more complex iterative RAG pipelines can partially decouple generation quality from retrieval effectiveness. These findings provide empirical support for using retrieval metrics as proxies for RAG performance.

Long-horizon task planning for heterogeneous multi-robot systems is essential for deploying collaborative teams in real-world environments; yet, it remains challenging due to the large volume of perceptual information, much of which is irrelevant to task objectives and burdens planning. Traditional symbolic planners rely on manually constructed problem specifications, limiting scalability and adaptability, while recent large language model (LLM)-based approaches often suffer from hallucinations and weak grounding-i.e., poor alignment between generated plans and actual environmental objects and constraints-in object-rich settings. We present Scale-Plan, a scalable LLM-assisted framework that generates compact, task-relevant problem representations from natural language instructions. Given a PDDL domain specification, Scale-Plan constructs an action graph capturing domain structure and uses shallow LLM reasoning to guide a structured graph search that identifies a minimal subset of relevant actions and objects. By filtering irrelevant information prior to planning, Scale-Plan enables efficient decomposition, allocation, and long-horizon plan generation. We evaluate our approach on complex multi-agent tasks and introduce MAT2-THOR, a cleaned benchmark built on AI2-THOR for reliable evaluation of multi-robot planning systems. Scale-Plan outperforms pure LLM and hybrid LLM-PDDL baselines across all metrics, improving scalability and reliability.

We present Test-Driven AI Agent Definition (TDAD), a methodology that treats agent prompts as compiled artifacts: engineers provide behavioral specifications, a coding agent converts them into executable tests, and a second coding agent iteratively refines the prompt until tests pass. Deploying tool-using LLM agents in production requires measurable behavioral compliance that current development practices cannot provide. Small prompt changes cause silent regressions, tool misuse goes undetected, and policy violations emerge only after deployment. To mitigate specification gaming, TDAD introduces three mechanisms: (1) visible/hidden test splits that withhold evaluation tests during compilation, (2) semantic mutation testing via a post-compilation agent that generates plausible faulty prompt variants, with the harness measuring whether the test suite detects them, and (3) spec evolution scenarios that quantify regression safety when requirements change. We evaluate TDAD on SpecSuite-Core, a benchmark of four deeply-specified agents spanning policy compliance, grounded analytics, runbook adherence, and deterministic enforcement. Across 24 independent trials, TDAD achieves 92% v1 compilation success with 97% mean hidden pass rate; evolved specifications compile at 58%, with most failed runs passing all visible tests except 1-2, and show 86-100% mutation scores, 78% v2 hidden pass rate, and 97% regression safety scores. The implementation is available as an open benchmark at https://github.com/f-labs-io/tdad-paper-code.

The Markov Transition Field (MTF), introduced by Wang and Oates (2015), encodes a time series as a two-dimensional image by mapping each pair of time steps to the transition probability between their quantile states, estimated from a single global transition matrix. This construction is efficient when the transition dynamics are stationary, but produces a misleading representation when the process changes regime over time: the global matrix averages across regimes and the resulting image loses all information about \emph{when} each dynamical regime was active. In this paper we introduce the \emph{Temporal Markov Transition Field} (TMTF), an extension that partitions the series into $K$ contiguous temporal chunks, estimates a separate local transition matrix for each chunk, and assembles the image so that each row reflects the dynamics local to its chunk rather than the global average. The resulting $T \times T$ image has $K$ horizontal bands of distinct texture, each encoding the transition dynamics of one temporal segment. We develop the formal definition, establish the key structural properties of the representation, work through a complete numerical example that makes the distinction from the global MTF concrete, analyse the bias--variance trade-off introduced by temporal chunking, and discuss the geometric interpretation of the local transition matrices in terms of process properties such as persistence, mean reversion, and trending behaviour. The TMTF is amplitude-agnostic and order-preserving, making it suitable as an input channel for convolutional neural networks applied to time series characterisation tasks.

Superconducting circuits have demonstrated significant potential in quantum information processing and quantum sensing. Implementing novel control and measurement sequences for superconducting qubits is often a complex and time-consuming process, requiring extensive expertise in both the underlying physics and the specific hardware and software. In this work, we introduce a framework that leverages a large language model (LLM) to automate qubit control and measurement. Specifically, our framework conducts experiments by generating and invoking schema-less tools on demand via a knowledge base on instrumental usage and experimental procedures. We showcase this framework with two experiments: an autonomous resonator characterization and a direct reproduction of a quantum non-demolition (QND) characterization of a superconducting qubit from literature. This framework enables rapid deployment of standard control-and-measurement protocols and facilitates implementation of novel experimental procedures, offering a more flexible and user-friendly paradigm for controlling complex quantum hardware.

Diffusion models degrade images through noise, and reversing this process reveals an information hierarchy across timesteps. Scale-space theory exhibits a similar hierarchy via low-pass filtering. We formalize this connection and show that highly noisy diffusion states contain no more information than small, downsampled images - raising the question of why they must be processed at full resolution. To address this, we fuse scale spaces into the diffusion process by formulating a family of diffusion models with generalized linear degradations and practical implementations. Using downsampling as the degradation yields our proposed Scale Space Diffusion. To support Scale Space Diffusion, we introduce Flexi-UNet, a UNet variant that performs resolution-preserving and resolution-increasing denoising using only the necessary parts of the network. We evaluate our framework on CelebA and ImageNet and analyze its scaling behavior across resolutions and network depths. Our project website ( https://prateksha.github.io/projects/scale-space-diffusion/ ) is available publicly.

We consider problems in sequential decision making with natural multi-level structure, where sub-tasks are assembled together to accomplish complex goals. Systematically inferring and leveraging hierarchical structure has remained a longstanding challenge; we describe an efficient multi-level procedure for repeatedly compressing Markov decision processes (MDPs), wherein a parametric family of policies at one level is treated as single actions in the compressed MDPs at higher levels, while preserving the semantic meanings and structure of the original MDP, and mimicking the natural logic to address a complex MDP. Higher-level MDPs are themselves independent MDPs with less stochasticity, and may be solved using existing algorithms. As a byproduct, spatial or temporal scales may be coarsened at higher levels, making it more efficient to find long-term optimal policies. The multi-level representation delivered by this procedure decouples sub-tasks from each other and usually greatly reduces unnecessary stochasticity and the policy search space, leading to fewer iterations and computations when solving the MDPs. A second fundamental aspect of this work is that these multi-level decompositions plus the factorization of policies into embeddings (problem-specific) and skills (including higher-order functions) yield new transfer opportunities of skills across different problems and different levels. This whole process is framed within curriculum learning, wherein a teacher organizes the student agent's learning process in a way that gradually increases the difficulty of tasks and and promotes transfer across MDPs and levels within and across curricula. The consistency of this framework and its benefits can be guaranteed under mild assumptions. We demonstrate abstraction, transferability, and curriculum learning in examples, including MazeBase+, a more complex variant of the MazeBase example.

Recent advances in time-series forecasting increasingly rely on pre-trained foundation-style models. While these models often claim broad generalization, existing evaluation protocols provide limited evidence. Indeed, most current benchmarks use static train-test splits that can easily lead to contamination as foundation models can inadvertently train on test data or perform model selection using test scores, which can inflate performance. We introduce Impermanent, a live benchmark that evaluates forecasting models under open-world temporal change by scoring forecasts sequentially over time on continuously updated data streams, enabling the study of temporal robustness, distributional shift, and performance stability rather than one-off accuracy on a frozen test set. Impermanent is instantiated on GitHub open-source activity, providing a naturally live and highly non-stationary dataset shaped by releases, shifting contributor behavior, platform/tooling changes, and external events. We focus on the top 400 repositories by star count and construct time series from issues opened, pull requests opened, push events, and new stargazers, evaluated over a rolling window with daily updates, alongside standardized protocols and leaderboards for reproducible, ongoing comparison. By shifting evaluation from static accuracy to sustained performance, Impermanent takes a concrete step toward assessing when and whether foundation-level generalization in time-series forecasting can be meaningfully claimed. Code and a live dashboard are available at https://github.com/TimeCopilot/impermanent and https://impermanent.timecopilot.dev.

Training large language models (LLMs) as autonomous agents often begins with imitation learning, but it only teaches agents what to do without understanding why: agents never contrast successful actions against suboptimal alternatives and thus lack awareness of action quality. Recent approaches attempt to address this by introducing self-reflection supervision derived from contrasts between expert and alternative actions. However, the training paradigm fundamentally remains imitation learning: the model imitates pre-constructed reflection text rather than learning to reason autonomously. We propose Agentic Critical Training (ACT), a reinforcement learning paradigm that trains agents to identify the better action among alternatives. By rewarding whether the model's judgment is correct, ACT drives the model to autonomously develop reasoning about action quality, producing genuine self-reflection rather than imitating it. Across three challenging agent benchmarks, ACT consistently improves agent performance when combined with different post-training methods. It achieves an average improvement of 5.07 points over imitation learning and 4.62 points over reinforcement learning. Compared to approaches that inject reflection capability through knowledge distillation, ACT also demonstrates clear advantages, yielding an average improvement of 2.42 points. Moreover, ACT enables strong out-of-distribution generalization on agentic benchmarks and improves performance on general reasoning benchmarks without any reasoning-specific training data, highlighting the value of our method. These results suggest that ACT is a promising path toward developing more reflective and capable LLM agents.

Large language models are increasingly used for financial analysis and investment research, yet systematic evaluation of their financial reasoning capabilities remains limited. In this work, we introduce the AI Financial Intelligence Benchmark (AFIB), a multi-dimensional evaluation framework designed to assess financial analysis capabilities across five dimensions: factual accuracy, analytical completeness, data recency, model consistency, and failure patterns. We evaluate five AI systems: GPT, Gemini, Perplexity, Claude, and SuperInvesting, using a dataset of 95+ structured financial analysis questions derived from real-world equity research tasks. The results reveal substantial differences in performance across models. Within this benchmark setting, SuperInvesting achieves the highest aggregate performance, with an average factual accuracy score of 8.96/10 and the highest completeness score of 56.65/70, while also demonstrating the lowest hallucination rate among evaluated systems. Retrieval-oriented systems such as Perplexity perform strongly on data recency tasks due to live information access but exhibit weaker analytical synthesis and consistency. Overall, the results highlight that financial intelligence in large language models is inherently multi-dimensional, and systems that combine structured financial data access with analytical reasoning capabilities provide the most reliable performance for complex investment research workflows.

The rapid advancement of artificial intelligence (AI) technologies presents both unprecedented opportunities and significant challenges for sustainable economic development. While AI offers transformative potential for addressing environmental challenges and enhancing economic resilience, its deployment often involves substantial energy consumption and environmental costs. This research introduces the EcoAI-Resilience framework, a multi-objective optimization approach designed to maximize the sustainability benefits of AI deployment while minimizing environmental costs and enhancing economic resilience. The framework addresses three critical objectives through mathematical optimization: sustainability impact maximization, economic resilience enhancement, and environmental cost minimization. The methodology integrates diverse data sources, including energy consumption metrics, sustainability indicators, economic performance data, and entrepreneurship outcomes across 53 countries and 14 sectors from 2015-2024. Our experimental validation demonstrates exceptional performance with R scores exceeding 0.99 across all model components, significantly outperforming baseline methods, including Linear Regression (R = 0.943), Random Forest (R = 0.957), and Gradient Boosting (R = 0.989). The framework successfully identifies optimal AI deployment strategies featuring 100\% renewable energy integration, 80% efficiency improvement targets, and optimal investment levels of $202.48 per capita. Key findings reveal strong correlations between economic complexity and resilience (r = 0.82), renewable energy adoption and sustainability outcomes (r = 0.71), and demonstrate significant temporal improvements in AI readiness (+1.12 points/year) and renewable energy adoption (+0.67 year) globally.

Can we find a network architecture for ML model training so as to optimize training loss (and thus, accuracy) in Split Federated Learning (SFL)? And can this architecture also reduce training delay and communication overhead? While accuracy is not influenced by how we split the model in ordinary, state-of-the-art SFL, in this work we answer the questions above in the affirmative. Recent Hierarchical SFL (HSFL) architectures adopt a three-tier training structure consisting of clients, (local) aggregators, and a central server. In this architecture, the model is partitioned at two partitioning layers into three sub-models, which are executed across the three tiers. Despite their merits, HSFL architectures overlook the impact of the partitioning layers and client-to-aggregator assignments on accuracy, delay, and overhead. This work explicitly captures the impact of the partitioning layers and client-to-aggregator assignments on accuracy, delay and overhead by formulating a joint optimization problem. We prove that the problem is NP-hard and propose the first accuracy-aware heuristic algorithm that explicitly accounts for model accuracy, while remaining delay-efficient. Simulation results on public datasets show that our approach can improve accuracy by 3%, while reducing delay by 20% and overhead by 50%, compared to state-of-the-art SFL and HSFL schemes.

Autoregressive "language" models (LMs) trained on raw waveforms can be repurposed for lossless audio compression, but prior work is limited to 8-bit audio, leaving open whether such approaches work for practical settings (16/24-bit) and can compete with existing codecs. We benchmark LM-based compression on full-fidelity audio across diverse domains (music, speech, bioacoustics), sampling rates (16kHz-48kHz), and bit depths (8, 16, 24-bit). Standard sample-level tokenization becomes intractable at higher bit depths due to vocabulary size (65K for 16-bit; 16.7M for 24-bit). We propose Trilobyte, a byte-level tokenization schema for full resolution audio, improving vocabulary scaling from $O(2^{b})$ to $O(1)$ and enabling the first tractable 24-bit LM-based lossless compression. While LMs consistently outperform FLAC and yield state-of-the-art compression at 8-bit and 16-bit, we observe that compression gains become more modest as bit depth increases beyond 8-bit.

We introduce structural causal bottleneck models (SCBMs), a novel class of structural causal models. At the core of SCBMs lies the assumption that causal effects between high-dimensional variables only depend on low-dimensional summary statistics, or bottlenecks, of the causes. SCBMs provide a flexible framework for task-specific dimension reduction while being estimable via standard, simple learning algorithms in practice. We analyse identifiability in SCBMs, connect them to information bottlenecks in the sense of Tishby & Zaslavsky (2015), and illustrate how to estimate them experimentally. We also demonstrate the benefit of bottlenecks for effect estimation in low-sample transfer learning settings. We argue that SCBMs provide an alternative to existing causal dimension reduction frameworks like causal representation learning or causal abstraction learning.

The celebrated Myerson--Satterthwaite theorem shows that in bilateral trade, no mechanism can be simultaneously fully efficient, Bayesian incentive compatible (BIC), and budget balanced (BB). This naturally raises the question of how closely the gains from trade (GFT) achievable by a BIC and BB mechanism can approximate the first-best (fully efficient) benchmark. The optimal BIC and BB mechanism is typically complex and highly distribution-dependent, making it difficult to characterize directly. Consequently, much of the literature analyzes simpler mechanisms such as the Random-Offerer (RO) mechanism and establishes constant-factor guarantees relative to the first-best GFT. An important open question concerns the worst-case performance of the RO mechanism relative to first-best (FB) efficiency. While it was originally hypothesized that the approximation ratio $\frac{\text{GFT}_{\text{FB}}}{\text{GFT}_{\text{RO}}}$ is bounded by $2$, recent work provided counterexamples to this conjecture: Cai et al. proved that the ratio can be strictly larger than $2$, and Babaioff et al. exhibited an explicit example with ratio approximately $2.02$. In this work, we employ AlphaEvolve, an AI-guided evolutionary search framework, to explore the space of value distributions. We identify a new worst-case instance that yields an improved lower bound of $\frac{\text{GFT}_{\text{FB}}}{\text{GFT}_{\text{RO}}} \ge \textbf{2.0749}$. This establishes a new lower bound on the worst-case performance of the Random-Offerer mechanism, demonstrating a wider efficiency gap than previously known.

Maximum marginal likelihood estimation (MMLE) can be formulated as the optimization of a free energy functional. From this viewpoint, the Expectation-Maximisation (EM) algorithm admits a natural interpretation as a coordinate descent method over the joint space of model parameters and probability measures. Recently, a significant body of work has adopted this perspective, leading to interacting particle algorithms for MMLE. In this paper, we propose an accelerated version of one such procedure, based on Stein variational gradient descent (SVGD), by introducing Nesterov acceleration in both the parameter updates and in the space of probability measures. The resulting method, termed Momentum SVGD-EM, consistently accelerates convergence in terms of required iterations across various tasks of increasing difficulty, demonstrating effectiveness in both low- and high-dimensional settings.

In the forthcoming years the LHC experiments are going to be upgraded to benefit from the substantial increase of the LHC instantaneous luminosity, which will lead to larger, denser events, and, consequently, greater complexity in reconstructing charged particle tracks, motivating frontier research in new technologies. Quantum machine learning models are being investigated as potential new approaches to high energy physics (HEP) tasks. We characterize and upgrade a quantum graph neural network (QGNN) architecture for charged particle track reconstruction on a simulated high luminosity dataset. The model operates on a set of event graphs, each built from the hits generated in tracking detector layers by particles produced in proton collisions, performing a classification of the possible hit connections between adjacent layers. In this approach the QGNN is designed as a hybrid architecture, interleaving classical feedforward networks with parametrized quantum circuits. We characterize the interplay between the classical and quantum components. We report on the principal upgrades to the original design, and present new evidence of improved training behavior, specifically in terms of convergence toward the final trained configuration.

Unsupervised reinforcement learning with verifiable rewards (URLVR) offers a pathway to scale LLM training beyond the supervision bottleneck by deriving rewards without ground truth labels. Recent works leverage model intrinsic signals, showing promising early gains, yet their potential and limitations remain unclear. In this work, we revisit URLVR and provide a comprehensive analysis spanning taxonomy, theory and extensive experiments. We first classify URLVR methods into intrinsic versus external based on reward sources, then establish a unified theoretical framework revealing that all intrinsic methods converge toward sharpening the model's initial distribution This sharpening mechanism succeeds when initial confidence aligns with correctness but fails catastrophically when misaligned. Through systematic experiments, we show intrinsic rewards consistently follow a rise-then-fall pattern across methods, with collapse timing determined by model prior rather than engineering choices. Despite these scaling limits, we find intrinsic rewards remain valuable in test-time training on small datasets, and propose Model Collapse Step to measure model prior, serving as a practical indicator for RL trainability. Finally, we explore external reward methods that ground verification in computational asymmetries, showing preliminary evidence they may escape the confidence-correctness ceiling. Our findings chart boundaries for intrinsic URLVR while motivating paths toward scalable alternatives.

The emergence of large reasoning models demonstrates that scaling inference-time compute significantly enhances performance on complex tasks. However, it often falls into another trap: overthinking simple problems, where repetitive rationales yield minimal accuracy gains at a disproportionately high cost. This motivates adaptive reasoning: dynamically aligning reasoning depth with instance difficulty. In this paper, we study adaptive reasoning from an optimality perspective, formalizing it as a utility maximization problem where tokens are allocated until the marginal accuracy gain falls below the incremental cost. Based on this, we propose CODA (Compute Allocation by Difficulty Awareness), a method that operationalizes this principle by allocating tokens via a policy-internal difficulty signal. Specifically, CODA estimates difficulty via group-based rollouts and maps it to two non-negative gates that modulate a length-dependent shaping term on top of the binary base reward. The easy-side gate penalizes verbosity on simple instances, whereas the hard-side gate encourages more deliberative rollouts on challenging ones. Across model scales and benchmarks, CODA achieves adaptive reasoning without external annotations or user-provided budgets: on easy tasks, CODA reduces token costs by over 60% while maintaining strong accuracy, whereas on hard tasks it incentivizes more deliberative rollouts to maximize performance.

In this paper, we present a context-free unsupervised approach based on a self-conditioned GAN to learn different modes from 2D trajectories. Our intuition is that each mode indicates a different behavioral moving pattern in the discriminator's feature space. We apply this approach to the problem of trajectory forecasting. We present three different training settings based on self-conditioned GAN, which produce better forecasters. We test our method in two data sets: human motion and road agents. Experimental results show that our approach outperforms previous context-free methods in the least representative supervised labels while performing well in the remaining labels. In addition, our approach outperforms globally in human motion, while performing well in road agents.

We introduce OfficeQA Pro, a benchmark for evaluating AI agents on grounded, multi-document reasoning over a large and heterogeneous document corpus. The corpus consists of U.S. Treasury Bulletins spanning nearly 100 years, comprising 89,000 pages and over 26 million numerical values. OfficeQA Pro consists of 133 questions that require precise document parsing, retrieval, and analytical reasoning across both unstructured text and tabular data. Frontier LLMs including Claude Opus 4.6, GPT-5.4, and Gemini 3.1 Pro Preview achieve less than 5% accuracy on OfficeQA Pro when relying on parametric knowledge, and less than 12% with additional access to the web. When provided directly with the document corpus, frontier agents still struggle on over half of questions, scoring 34.1% on average. We find that providing agents with a structured document representation produced by Databricks' ai_parse_document yields a 16.1% average relative performance gain across agents. We conduct additional ablations to study the effects of model selection, table representation, retrieval strategy, and test-time scaling on performance. Despite these improvements, significant headroom remains before agents can be considered reliable at enterprise-grade grounded reasoning.

Recent advancements in Unified Multimodal Models (UMMs) have significantly advanced text-to-image (T2I) generation, particularly through the integration of Chain-of-Thought (CoT) reasoning. However, existing CoT-based T2I methods largely rely on abstract natural-language planning, which lacks the precision required for complex spatial layouts, structured visual elements, and dense textual content. In this work, we propose CoCo (Code-as-CoT), a code-driven reasoning framework that represents the reasoning process as executable code, enabling explicit and verifiable intermediate planning for image generation. Given a text prompt, CoCo first generates executable code that specifies the structural layout of the scene, which is then executed in a sandboxed environment to render a deterministic draft image. The model subsequently refines this draft through fine-grained image editing to produce the final high-fidelity result. To support this training paradigm, we construct CoCo-10K, a curated dataset containing structured draft-final image pairs designed to teach both structured draft construction and corrective visual refinement. Empirical evaluations on StructT2IBench, OneIG-Bench, and LongText-Bench show that CoCo achieves improvements of +68.83%, +54.8%, and +41.23% over direct generation, while also outperforming other generation methods empowered by CoT. These results demonstrate that executable code is an effective and reliable reasoning paradigm for precise, controllable, and structured text-to-image generation. The code is available at: https://github.com/micky-li-hd/CoCo

We introduce a comprehensive theoretical and algorithmic framework that bridges formal group theory and group entropies with modern machine learning, paving the way for an infinite, flexible family of Mirror Descent (MD) optimization algorithms. Our approach exploits the rich structure of group entropies, which are generalized entropic functionals governed by group composition laws, encompassing and significantly extending all trace-form entropies such as the Shannon, Tsallis, and Kaniadakis families. By leveraging group-theoretical mirror maps (or link functions) in MD, expressed via multi-parametric generalized logarithms and their inverses (group exponentials), we achieve highly flexible and adaptable MD updates that can be tailored to diverse data geometries and statistical distributions. To this end, we introduce the notion of \textit{mirror duality}, which allows us to seamlessly switch or interchange group-theoretical link functions with their inverses, subject to specific learning rate constraints. By tuning or learning the hyperparameters of the group logarithms enables us to adapt the model to the statistical properties of the training distribution, while simultaneously ensuring desirable convergence characteristics via fine-tuning. This generality not only provides greater flexibility and improved convergence properties, but also opens new perspectives for applications in machine learning and deep learning by expanding the design of regularizers and natural gradient algorithms. We extensively evaluate the validity, robustness, and performance of the proposed updates on large-scale, simplex-constrained quadratic programming problems.

In this article the authors develop an intrinsic measure for quantifying heterogeneity in training data for supervised learning. This measure is the variance of a random variable which factors through the influences of pairs of training points. The variance is shown to capture data heterogeneity and can thus be used to assess if a sample is a mixture of distributions. The authors prove that the data itself contains key information that supports a partitioning into blocks. Several proof of concept studies are provided that quantify the connection between variance and heterogeneity for EMNIST image data and synthetic data. The authors establish that variance is maximal for equal mixes of distributions, and detail how variance-based data purification followed by conventional training over blocks can lead to significant increases in test accuracy.

Adapting pre-trained models to specialized tasks often leads to catastrophic forgetting, where new knowledge overwrites foundational capabilities. Existing methods either compromise performance on the new task or struggle to balance training stability with efficient reuse of pre-trained knowledge. We introduce a novel function-preserving expansion method that resolves this dilemma. Our technique expands model capacity by replicating pre-trained parameters within transformer submodules and applying a scaling correction that guarantees the expanded model is mathematically identical to the original at initialization, enabling stable training while exploiting existing knowledge. Empirically, our method eliminates the trade-off between plasticity and stability, matching the performance of full fine-tuning on downstream tasks without any degradation of the model's original capabilities. Furthermore, we demonstrate the modularity of our approach, showing that by selectively expanding a small subset of layers we can achieve the same performance as full fine-tuning at a fraction of the computational cost.

Template-free animatable head avatars can achieve high visual fidelity by learning expression-dependent facial deformation directly from a subject's capture, avoiding parametric face templates and hand-designed blendshape spaces. However, since learned deformation is supervised only by the expressions observed for a single identity, these models suffer from limited expression coverage and often struggle when driven by motions that deviate from the training distribution. We introduce RAF (Retrieval-Augmented Faces), a simple training-time augmentation designed for template-free head avatars that learn deformation from data. RAF constructs a large unlabeled expression bank and, during training, replaces a subset of the subject's expression features with nearest-neighbor expressions retrieved from this bank while still reconstructing the subject's original frames. This exposes the deformation field to a broader range of expression conditions, encouraging stronger identity-expression decoupling and improving robustness to expression distribution shift without requiring paired cross-identity data, additional annotations, or architectural changes. We further analyze how retrieval augmentation increases expression diversity and validate retrieval quality with a user study showing that retrieved neighbors are perceptually closer in expression and pose. Experiments on the NeRSemble benchmark demonstrate that RAF consistently improves expression fidelity over the baseline, in both self-driving and cross-driving scenarios.

AI agents have become surprisingly proficient at software engineering over the past year, largely due to improvements in reasoning capabilities. This raises a deeper question: can these systems extend their capabilities to automate AI research itself? In this paper, we explore post-training, the critical phase that turns base LLMs into useful assistants. We introduce PostTrainBench to benchmark how well LLM agents can perform post-training autonomously under bounded compute constraints (10 hours on one H100 GPU). We ask frontier agents (e.g., Claude Code with Opus 4.6) to optimize the performance of a base LLM on a particular benchmark (e.g., Qwen3-4B on AIME). Importantly, we do not provide any predefined strategies to the agents and instead give them full autonomy to find necessary information on the web, run experiments, and curate data. We find that frontier agents make substantial progress but generally lag behind instruction-tuned LLMs from leading providers: 23.2% for the best agent vs. 51.1% for official instruction-tuned models. However, agents can exceed instruction-tuned models in targeted scenarios: GPT-5.1 Codex Max achieves 89% on BFCL with Gemma-3-4B vs. 67% for the official model. We also observe several failure modes worth flagging. Agents sometimes engage in reward hacking: training on the test set, downloading existing instruction-tuned checkpoints instead of training their own, and using API keys they find to generate synthetic data without authorization. These behaviors are concerning and highlight the importance of careful sandboxing as these systems become more capable. Overall, we hope PostTrainBench will be useful for tracking progress in AI R&D automation and for studying the risks that come with it. Website and code are available at https://posttrainbench.com/.

Ensuring trustworthiness in open-world visual recognition requires models that are interpretable, fair, and robust to distribution shifts. Yet modern vision systems are increasingly deployed as proprietary black-box APIs, exposing only output probabilities and hiding architecture, parameters, gradients, and training data. This opacity prevents meaningful auditing, bias detection, and failure analysis. Existing explanation methods assume white- or gray-box access or knowledge of the training distribution, making them unusable in these real-world settings. We introduce UNBOX, a framework for class-wise model dissection under fully data-free, gradient-free, and backpropagation-free constraints. UNBOX leverages Large Language Models and text-to-image diffusion models to recast activation maximization as a purely semantic search driven by output probabilities. The method produces human-interpretable text descriptors that maximally activate each class, revealing the concepts a model has implicitly learned, the training distribution it reflects, and potential sources of bias. We evaluate UNBOX on ImageNet-1K, Waterbirds, and CelebA through semantic fidelity tests, visual-feature correlation analyses and slice-discovery auditing. Despite operating under the strictest black-box constraints, UNBOX performs competitively with state-of-the-art white-box interpretability methods. This demonstrates that meaningful insight into a model's internal reasoning can be recovered without any internal access, enabling more trustworthy and accountable visual recognition systems.

We derive integral formulas that simplify the Vector Spherical Tensor Product recently introduced by Xie et al., which generalizes the Gaunt tensor product to antisymmetric couplings. In particular, we obtain explicit closed-form expressions for the antisymmetric analogues of the Gaunt coefficients. This enables us to simulate the Clebsch-Gordan tensor product using a single Vector Spherical Tensor Product, yielding a $9\times$ reduction in the required tensor product evaluations. Our results enable efficient and practical implementations of the Vector Spherical Tensor Product, paving the way for applications of this generalization of Gaunt tensor products in $\mathrm{SO}(3)$-equivariant neural networks. Moreover, we discuss how the Gaunt and the Vector Spherical Tensor Products allow to control the expressivity-runtime tradeoff associated with the usual Clebsch-Gordan Tensor Products. Finally, we investigate low rank decompositions of the normalizations of the considered tensor products in view of their use in equivariant neural networks.

Coverage-guided fuzzing has proven effective for software testing, but targeting library code requires specialized fuzz harnesses that translate fuzzer-generated inputs into valid API invocations. Manual harness creation is time-consuming and requires deep understanding of API semantics, initialization sequences, and exception handling contracts. We present a multi-agent architecture that automates fuzz harness generation for Java libraries through specialized LLM-powered agents. Five ReAct agents decompose the workflow into research, synthesis, compilation repair, coverage analysis, and refinement. Rather than preprocessing entire codebases, agents query documentation, source code, and callgraph information on demand through the Model Context Protocol, maintaining focused context while exploring complex dependencies. To enable effective refinement, we introduce method-targeted coverage that tracks coverage only during target method execution to isolate target behavior, and agent-guided termination that examines uncovered source code to distinguish productive refinement opportunities from diminishing returns. We evaluated our approach on seven target methods from six widely-deployed Java libraries totaling 115,000+ Maven dependents. Our generated harnesses achieve a median 26\% improvement over OSS-Fuzz baselines and outperform Jazzer AutoFuzz by 5\% in package-scope coverage. Generation costs average \$3.20 and 10 minutes per harness, making the approach practical for continuous fuzzing workflows. During a 12-hour fuzzing campaign, our generated harnesses discovered 3 bugs in projects that are already integrated into OSS-Fuzz, demonstrating the effectiveness of the generated harnesses.

We present Midicoth, a lossless compression system that introduces a micro-diffusion denoising layer for improving probability estimates produced by adaptive statistical models. In compressors such as Prediction by Partial Matching (PPM), probability estimates are smoothed by a prior to handle sparse observations. When contexts have been seen only a few times, this prior dominates the prediction and produces distributions that are significantly flatter than the true source distribution, leading to compression inefficiency. Midicoth addresses this limitation by treating prior smoothing as a shrinkage process and applying a reverse denoising step that corrects predicted probabilities using empirical calibration statistics. To make this correction data-efficient, the method decomposes each byte prediction into a hierarchy of binary decisions along a bitwise tree. This converts a single 256-way calibration problem into a sequence of binary calibration tasks, enabling reliable estimation of correction terms from relatively small numbers of observations. The denoising process is applied in multiple successive steps, allowing each stage to refine residual prediction errors left by the previous one. The micro-diffusion layer operates as a lightweight post-blend calibration stage applied after all model predictions have been combined, allowing it to correct systematic biases in the final probability distribution. Midicoth combines five fully online components: an adaptive PPM model, a long-range match model, a trie-based word model, a high-order context model, and the micro-diffusion denoiser applied as the final stage.

Background and objectives: Colorectal cancer histopathological grading depends on accurate segmentation of glandular structures. Current deep learning approaches rely on large scale pixel level annotations that are labor intensive and difficult to obtain in routine clinical practice. Weakly supervised semantic segmentation offers a promising alternative. However, class activation map based methods often produce incomplete pseudo masks that emphasize highly discriminative regions and fail to supervise unannotated glandular structures. We propose a weakly supervised teacher student framework that leverages sparse pathologist annotations and an Exponential Moving Average stabilized teacher network to generate refined pseudo masks. Methods: The framework integrates confidence based filtering, adaptive fusion of teacher predictions with limited ground truth, and curriculum guided refinement to progressively segment unannotated glandular regions. The method was evaluated on an institutional colorectal cancer cohort from The Ohio State University Wexner Medical Center consisting of 60 hematoxylin and eosin stained whole slide images and on public datasets including the Gland Segmentation dataset, TCGA COAD, TCGA READ, and SPIDER. Results: On the Gland Segmentation dataset the framework achieved a mean Intersection over Union of 80.10 and a mean Dice coefficient of 89.10. Cross cohort evaluation demonstrated robust generalization on TCGA COAD and TCGA READ without additional annotations, while reduced performance on SPIDER reflected domain shift. Conclusions: The proposed framework provides an annotation efficient and generalizable approach for gland segmentation in colorectal histopathology.

Concept drift, temporal dependence, and catastrophic forgetting represent major challenges when learning from data streams. While Streaming Machine Learning and Continual Learning (CL) address these issues separately, recent efforts in Streaming Continual Learning (SCL) aim to unify them. In this work, we introduce MAGIC Net, a novel SCL approach that integrates CL-inspired architectural strategies with recurrent neural networks to tame temporal dependence. MAGIC Net continuously learns, looks back at past knowledge by applying learnable masks over frozen weights, and expands its architecture when necessary. It performs all operations online, ensuring inference availability at all times. Experiments on synthetic and real-world streams show that it improves adaptation to new concepts, limits memory usage, and mitigates forgetting.

State-of-the-art deep reinforcement learning (RL) methods have achieved remarkable performance in continuous control tasks, yet their computational complexity is often incompatible with the constraints of resource-limited hardware, due to their reliance on replay buffers, batch updates, and target networks. The emerging paradigm of streaming deep RL addresses this limitation through purely online updates, achieving strong empirical performance on standard benchmarks. In this work, we propose two novel streaming deep RL algorithms, Streaming Soft Actor-Critic (S2AC) and Streaming Deterministic Actor-Critic (SDAC), explicitly designed to be compatible with state-of-the-art batch RL methods, making them particularly suitable for on-device finetuning applications such as Sim2Real transfer. Both algorithms achieve performance comparable to state-of-the-art streaming baselines on standard benchmarks without requiring tedious hyperparameter tuning. Finally, we further investigate the practical challenges of transitioning from batch to streaming learning during finetuning and propose concrete strategies to tackle them.

Multi-Layer Perceptrons (MLPs) rely on pre-defined, fixed activation functions, imposing a static inductive bias that forces the network to approximate complex topologies solely through increased depth and width. Kolmogorov-Arnold Networks (KANs) address this limitation through edge-centric learnable functions, yet their formulation suffers from quadratic parameter scaling and architectural rigidity that hinders the effective integration of standard regularization techniques. This paper introduces the DualFlexKAN (DFKAN), a flexible architecture featuring a dual-stage mechanism that independently controls pre-linear input transformations and post-linear output activations. This decoupling enables hybrid networks that optimize the trade-off between expressiveness and computational cost. Unlike standard formulations, DFKAN supports diverse basis function families, including orthogonal polynomials, B-splines, and radial basis functions, integrated with configurable regularization strategies that stabilize training dynamics. Comprehensive evaluations across regression benchmarks, physics-informed tasks, and function approximation demonstrate that DFKAN outperforms both MLPs and conventional KANs in accuracy, convergence speed, and gradient fidelity. The proposed hybrid configurations achieve superior performance with one to two orders of magnitude fewer parameters than standard KANs, effectively mitigating the parameter explosion problem while preserving KAN-style expressiveness. DFKAN provides a principled, scalable framework for incorporating adaptive non-linearities, proving particularly advantageous for data-efficient learning and interpretable function discovery in scientific applications.

Deployed machine learning systems face distribution drift, yet most monitoring pipelines stop at alarms and leave the response underspecified under labeling, compute, and latency constraints. We introduce Drift2Act, a drift-to-action controller that treats monitoring as constrained decision-making with explicit safety. Drift2Act combines a sensing layer that maps unlabeled monitoring signals to a belief over drift types with an active risk certificate that queries a small set of delayed labels from a recent window to produce an anytime-valid upper bound $U_t(δ)$ on current risk. The certificate gates operation: if $U_t(δ) \le τ$, the controller selects low-cost actions (e.g., recalibration or test-time adaptation); if $U_t(δ) > τ$, it activates abstain/handoff and escalates to rollback or retraining under cooldowns. In a realistic streaming protocol with label delay and explicit intervention costs, Drift2Act achieves near-zero safety violations and fast recovery at moderate cost on WILDS Camelyon17, DomainNet, and a controlled synthetic drift stream, outperforming alarm-only monitoring, adapt-always adaptation, schedule-based retraining, selective prediction alone, and an ablation without certification. Overall, online risk certification enables reliable drift response and reframes monitoring as decision-making with safety.

The question of \emph{knowledge}, \emph{truth} and \emph{trust} is explored via a mathematical formulation of claims and sources. We define truth as the reproducibly perceived subset of knowledge, formalize sources as having both generative and discriminative roles, and develop a framework for reputation grounded in the \emph{conviction} -- the likelihood that a source's stance is vindicated by independent consensus. We argue that conviction, rather than correctness or faithfulness, is the principled basis for trust: it is regime-independent, rewards genuine contribution, and demands the transparent and self-sufficient perceptions that make external verification possible. We formalize reputation as the expected weighted signed conviction over a realm of claims, characterize its behavior across source-claim regimes, and identify continuous verification as both a theoretical necessity and a practical mechanism through which reputation accrues. The framework is applied to AI agents, which are identified as capable but error-prone sources for whom verifiable conviction and continuously accrued reputation constitute the only robust foundation for trust.

Learning natural, stable, and compositionally generalizable whole-body control policies for humanoid robots performing simultaneous locomotion and manipulation (loco-manipulation) remains a fundamental challenge in robotics. Existing reinforcement learning approaches typically rely on a single monolithic policy to acquire multiple skills, which often leads to cross-skill gradient interference and motion pattern conflicts in high-degree-of-freedom systems. As a result, generated behaviors frequently exhibit unnatural movements, limited stability, and poor generalization to complex task compositions. To address these limitations, we propose MetaWorld-X, a hierarchical world model framework for humanoid control. Guided by a divide-and-conquer principle, our method decomposes complex control problems into a set of specialized expert policies (Specialized Expert Policies, SEP). Each expert is trained under human motion priors through imitation-constrained reinforcement learning, introducing biomechanically consistent inductive biases that ensure natural and physically plausible motion generation. Building upon this foundation, we further develop an Intelligent Routing Mechanism (IRM) supervised by a Vision-Language Model (VLM), enabling semantic-driven expert composition. The VLM-guided router dynamically integrates expert policies according to high-level task semantics, facilitating compositional generalization and adaptive execution in multi-stage loco-manipulation tasks.

DARPA's AI Cyber Challenge (AIxCC) showed that cyber reasoning systems (CRSs) can go beyond vulnerability discovery to autonomously confirm and patch bugs: seven teams built such systems and open-sourced them after the competition. Yet all seven open-sourced CRSs remain largely unusable outside their original teams, each bound to the competition cloud infrastructure that no longer exists. We present OSS-CRS, an open, locally deployable framework for running and combining CRS techniques against real-world open-source projects, with budget-aware resource management. We ported the first-place system (Atlantis) and discovered 10 previously unknown bugs (three of high severity) across 8 OSS-Fuzz projects. OSS-CRS is publicly available.

Large language model (LLM)-based agents trained with reinforcement learning (RL) have shown strong potential on complex interactive tasks. However, standard RL paradigms favor static problem-solving over continuous adaptation: agents often converge to suboptimal strategies due to insufficient exploration, while learned knowledge remains implicit within parameters rather than explicitly retrievable, limiting effective experiential learning. To address these limitations, we introduce RetroAgent, an online RL framework that empowers agents to master complex interactive environments not just by solving, but by evolving. Concretely, RetroAgent features a hindsight self-reflection mechanism that produces dual intrinsic feedback: (1) intrinsic numerical feedback that that tracks incremental subtask completion relative to prior attempts, rewarding promising explorations, and (2) intrinsic language feedback that distills reusable lessons into a memory buffer, retrieved via our proposed Similarity & Utility-Aware Upper Confidence Bound (SimUtil-UCB) strategy balancing relevance, utility, and exploration to effectively leverage past experiences. Extensive experiments on two model families across four challenging agentic tasks demonstrate that RetroAgent significantly outperforms existing methods, achieving state-of-the-art results -- e.g., surpassing Group Relative Policy Optimization (GRPO)-trained agents by +18.3% on ALFWorld, +15.4% on WebShop, +27.1% on Sokoban, and +8.9% on MineSweeper -- while exhibiting strong test-time adaptation and generalization to out-of-distribution scenarios.

Learning compact state representations in Markov Decision Processes (MDPs) has proven crucial for addressing the curse of dimensionality in large-scale reinforcement learning (RL) problems. Existing principled approaches leverage structural priors on the MDP by constructing state representations as linear combinations of the state-graph Laplacian eigenvectors. When the transition graph is unknown or the state space is prohibitively large, the graph spectral features can be estimated directly via sample trajectories. In this work, we prove an upper bound on the approximation error of linear value function approximation under the learned spectral features. We show how this error scales with the algebraic connectivity of the state-graph, grounding the approximation quality in the topological structure of the MDP. We further bound the error introduced by the eigenvector estimation itself, leading to an end-to-end error decomposition across the representation learning pipeline. Additionally, our expression of the Laplacian operator for the RL setting, although equivalent to existing ones, prevents some common misunderstandings, of which we show some examples from the literature. Our results hold for general (non-uniform) policies without any assumptions on the symmetry of the induced transition kernel. We validate our theoretical findings with numerical simulations on gridworld environments.

We propose Generative Adversarial Regression (GAR), a framework for learning conditional risk scenarios through generators aligned with downstream risk objectives. GAR builds on a regression characterization of conditional risk for elicitable functionals, including quantiles, expectiles, and jointly elicitable pairs. We extend this principle from point prediction to generative modeling by training generators whose policy-induced risk matches that of real data under the same context. To ensure robustness across all policies, GAR adopts a minimax formulation in which an adversarial policy identifies worst-case discrepancies in risk evaluation while the generator adapts to eliminate them. This structure preserves alignment with the risk functional across a broad class of policies rather than a fixed, pre-specified set. We illustrate GAR through a tail-risk instantiation based on jointly elicitable $(\mathrm{VaR}, \mathrm{ES})$ objectives. Experiments on S\&P 500 data show that GAR produces scenarios that better preserve downstream risk than unconditional, econometric, and direct predictive baselines while remaining stable under adversarially selected policies.

Action-conditioned video prediction models (often referred to as world models) have shown strong potential for robotics applications, but existing approaches are often slow and struggle to capture physically consistent interactions over long horizons, limiting their usefulness for scalable robot policy training and evaluation. We present Interactive World Simulator, a framework for building interactive world models from a moderate-sized robot interaction dataset. Our approach leverages consistency models for both image decoding and latent-space dynamics prediction, enabling fast and stable simulation of physical interactions. In our experiments, the learned world models produce interaction-consistent pixel-level predictions and support stable long-horizon interactions for more than 10 minutes at 15 FPS on a single RTX 4090 GPU. Our framework enables scalable demonstration collection solely within the world models to train state-of-the-art imitation policies. Through extensive real-world evaluation across diverse tasks involving rigid objects, deformable objects, object piles, and their interactions, we find that policies trained on world-model-generated data perform comparably to those trained on the same amount of real-world data. Additionally, we evaluate policies both within the world models and in the real world across diverse tasks, and observe a strong correlation between simulated and real-world performance. Together, these results establish the Interactive World Simulator as a stable and physically consistent surrogate for scalable robotic data generation and faithful, reproducible policy evaluation.

Object co-occurrences provide a key cue for finding objects successfully and efficiently in unfamiliar environments. Typically, one looks for cups in kitchens and views fridges as evidence of being in a kitchen. Such priors have also been exploited in artificial agents, but they are typically learned from explicitly labeled data or queried from language models. It is still unclear whether these relations can be learned implicitly from unlabeled observations alone. In this work, we address this problem and propose ProReFF, a feature field model trained to predict relative distributions of features obtained from pre-trained vision language models. In addition, we introduce a learning-based strategy that enables training from unlabeled and potentially contradictory data by aligning inconsistent observations into a coherent relative distribution. For the downstream object search task, we propose an agent that leverages predicted feature distributions as a semantic prior to guide exploration toward regions with a high likelihood of containing the object. We present extensive evaluations demonstrating that ProReFF captures meaningful relative feature distributions in natural scenes and provides insight into the impact of our proposed alignment step. We further evaluate the performance of our search agent in 100 challenges in the Matterport3D simulator, comparing with feature-based baselines and human participants. The proposed agent is 20% more efficient than the strongest baseline and achieves up to 80% of human performance.

Unsupervised graph alignment aims to find the node correspondence across different graphs without any anchor node pairs. Despite the recent efforts utilizing deep learning-based techniques, such as the embedding and optimal transport (OT)-based approaches, we observe their limitations in terms of model accuracy-efficiency tradeoff. By focusing on the exploitation of local and global graph information, we formalize them as the ``local representation, global alignment'' paradigm, and present a new ``global representation and alignment'' paradigm to resolve the mismatch between the two phases in the alignment process. We then propose \underline{Gl}obal representation and \underline{o}ptimal transport-\underline{b}ased \underline{Align}ment (\texttt{GlobAlign}), and its variant, \texttt{GlobAlign-E}, for better \underline{E}fficiency. Our methods are equipped with the global attention mechanism and a hierarchical cross-graph transport cost, able to capture long-range and implicit node dependencies beyond the local graph structure. Furthermore, \texttt{GlobAlign-E} successfully closes the time complexity gap between representative embedding and OT-based methods, reducing OT's cubic complexity to quadratic terms. Through extensive experiments, our methods demonstrate superior performance, with up to a 20\% accuracy improvement over the best competitor. Meanwhile, \texttt{GlobAlign-E} achieves the best efficiency, with an order of magnitude speedup against existing OT-based methods.

The application of large language models to code generation has evolved from one-shot generation to iterative refinement, yet the evolution of security throughout iteration remains insufficiently understood. Through comparative experiments on three mainstream LLMs, this paper reveals the iterative refinement paradox: specification drift during multi-objective optimization causes security to degrade gradually over successive iterations. Taking GPT-4o as an example, 43.7 % of iteration chains contain more vulnerabilities than the baseline after ten rounds, and cross-model experiments show that this phenomenon is prevalent. Further analysis shows that simply introducing static application security testing (SAST) gating cannot effectively suppress degradation; instead, it increases the latent security degradation rate from 12.5% under the unprotected baseline to 20.8 %. The root cause is that static-analysis rules cannot cover structural degradations such as the removal of defensive logic or the weakening of exception handling. To address this problem, we propose the SCAFFOLD-CEGIS framework. Drawing on the counterexample-guided inductive synthesis (CEGIS) paradigm, the framework adopts a multi-agent collaborative architecture that transforms security constraints from implicit prompts into explicit verifiable constraints. It automatically identifies and solidifies security-critical elements as hard constraints through semantic anchoring, enforces safety monotonicity through four-layer gated verification, and continuously assimilates experience from failures. Comparative experiments against six existing defense methods show that the full framework reduces the latent security degradation rate to 2.1% and achieves a safety monotonicity rate of 100%.

While standard reinforcement learning optimizes a single reward signal, many applications require optimizing a nonlinear utility $f(J_1^π,\dots,J_M^π)$ over multiple objectives, where each $J_m^π$ denotes the expected discounted return of a distinct reward function. A common approach is concave scalarization, which captures important trade-offs such as fairness and risk sensitivity. However, nonlinear scalarization introduces a fundamental challenge for policy gradient methods: the gradient depends on $\partial f(J^π)$, while in practice only empirical return estimates $\hat J$ are available. Because $f$ is nonlinear, the plug-in estimator is biased ($\mathbb{E}[\partial f(\hat J)] \neq \partial f(\mathbb{E}[\hat J])$), leading to persistent gradient bias that degrades sample complexity. In this work we identify and overcome this bias barrier in concave-scalarized multi-objective reinforcement learning. We show that existing policy-gradient methods suffer an intrinsic $\widetilde{\mathcal{O}}(ε^{-4})$ sample complexity due to this bias. To address this issue, we develop a Natural Policy Gradient (NPG) algorithm equipped with a multi-level Monte Carlo (MLMC) estimator that controls the bias of the scalarization gradient while maintaining low sampling cost. We prove that this approach achieves the optimal $\widetilde{\mathcal{O}}(ε^{-2})$ sample complexity for computing an $ε$-optimal policy. Furthermore, we show that when the scalarization function is second-order smooth, the first-order bias cancels automatically, allowing vanilla NPG to achieve the same $\widetilde{\mathcal{O}}(ε^{-2})$ rate without MLMC. Our results provide the first optimal sample complexity guarantees for concave multi-objective reinforcement learning under policy-gradient methods.

Recent DEtection TRansformer (DETR) based frameworks have achieved remarkable success in end-to-end object detection. However, the reliance on the Hungarian algorithm for bipartite matching between queries and ground truths introduces computational overhead and complicates the training dynamics. In this paper, we propose a novel matching-free training scheme for DETR-based detectors that eliminates the need for explicit heuristic matching. At the core of our approach is a dedicated Cross-Attention-based Query Selection (CAQS) module. Instead of discrete assignment, we utilize encoded ground-truth information to probe the decoder queries through a cross-attention mechanism. By minimizing the weighted error between the queried results and the ground truths, the model autonomously learns the implicit correspondences between object queries and specific targets. This learned relationship further provides supervision signals for the learning of queries. Experimental results demonstrate that our proposed method bypasses the traditional matching process, significantly enhancing training efficiency, reducing the matching latency by over 50\%, effectively eliminating the discrete matching bottleneck through differentiable correspondence learning, and also achieving superior performance compared to existing state-of-the-art methods.

In high stakes environments, agents relying purely on imitation learning or reinforcement learning often struggle to avoid safety-critical errors during exploration. Existing reinforcement learning approaches for environments such as chess require hundreds of thousands of episodes and substantial computational resources to converge. Imitation learning, on the other hand, is more sample efficient but is brittle under distributional shift and lacks mechanisms for proactive risk avoidance. In this work, we propose Oracle-Guided Soft Shielding (OGSS), a simple yet effective framework for safer decision-making, enabling safe exploration by learning a probabilistic safety model from oracle feedback in an imitation learning setting. Focusing on the domain of chess, we train a model to predict strong moves based on past games, and separately learn a blunder prediction model from Stockfish evaluations to estimate the tactical risk of each move. During inference, the agent first generates a set of candidate moves and then uses the blunder model to determine high-risk options, and uses a utility function combining the predicted move likelihood from the policy model and the blunder probability to select actions that strike a balance between performance and safety. This enables the agent to explore and play competitively while significantly reducing the chance of tactical mistakes. Across hundreds of games against a strong chess engine, we compare our approach with other methods in the literature, such as action pruning, SafeDAgger, and uncertainty-based sampling. Our results demonstrate that OGSS variants maintain a lower blunder rate even as the agent's exploration ratio is increased by several folds, highlighting its ability to support broader exploration without compromising tactical soundness.

Electrocardiography (ECG) is a low-cost, widely used modality for diagnosing electrical abnormalities like atrial fibrillation by capturing the heart's electrical activity. However, it cannot directly measure cardiac morphological phenotypes, such as left ventricular ejection fraction (LVEF), which typically require echocardiography (Echo). Predicting these phenotypes from ECG would enable early, accessible health screening. Existing self-supervised methods suffer from a representational mismatch by aligning ECGs to single-view Echos, which only capture local, spatially restricted anatomical snapshots. To address this, we propose Echo2ECG, a multimodal self-supervised learning framework that enriches ECG representations with the heart's morphological structure captured in multi-view Echos. We evaluate Echo2ECG as an ECG feature extractor on two clinically relevant tasks that fundamentally require morphological information: (1) classification of structural cardiac phenotypes across three datasets, and (2) retrieval of Echo studies with similar morphological characteristics using ECG queries. Our extracted ECG representations consistently outperform those of state-of-the-art unimodal and multimodal baselines across both tasks, despite being 18x smaller than the largest baseline. These results demonstrate that Echo2ECG is a robust, powerful ECG feature extractor. Our code is accessible at https://github.com/michelleespranita/Echo2ECG.

Large language models (LLMs) can answer religious knowledge queries fluently, yet they often hallucinate and misattribute sources, which is especially consequential in Islamic settings where users expect grounding in canonical texts (Qur'an and Hadith) and jurisprudential (fiqh) nuance. Retrieval-augmented generation (RAG) reduces some of these limitations by grounding generation in external evidence. However, a single ``retrieve-then-generate'' pipeline is limited to deal with the diversity of Islamic queries. Users may request verbatim scripture, fatwa-style guidance with citations or rule-constrained computations such as zakat and inheritance that require strict arithmetic and legal invariants. In this work, we present a bilingual (Arabic/English) multi-agent Islamic assistant, called Fanar-Sadiq, which is a core component of the Fanar AI platform. Fanar-Sadiq routes Islamic-related queries to specialized modules within an agentic, tool-using architecture. The system supports intent-aware routing, retrieval-grounded fiqh answers with deterministic citation normalization and verification traces, exact verse lookup with quotation validation, and deterministic calculators for Sunni zakat and inheritance with madhhab-sensitive branching. We evaluate the complete end-to-end system on public Islamic QA benchmarks and demonstrate effectiveness and efficiency. Our system is currently publicly and freely accessible through API and a Web application, and has been accessed $\approx$1.9M times in less than a year.

A reliable representation of uncertainty is essential for the application of modern machine learning methods in safety-critical settings. In this regard, the use of credal sets (i.e., convex sets of probability distributions) has recently been proposed as a suitable approach to representing epistemic uncertainty. However, as with other approaches to epistemic uncertainty, training credal predictors is computationally complex and usually involves (re-)training an ensemble of models. The resulting computational complexity prevents their adoption for complex models such as foundation models and multi-modal systems. To address this problem, we propose an efficient method for credal prediction that is grounded in the notion of relative likelihood and inspired by techniques for the calibration of probabilistic classifiers. For each class label, our method predicts a range of plausible probabilities in the form of an interval. To produce the lower and upper bounds of these intervals, we propose a technique that we refer to as decalibration. Extensive experiments show that our method yields credal sets with strong performance across diverse tasks, including coverage-efficiency evaluation, out-of-distribution detection, and in-context learning. Notably, we demonstrate credal prediction on models such as TabPFN and CLIP -- architectures for which the construction of credal sets was previously infeasible.

Optimization under structural constraints is typically analyzed through projection or penalty methods, obscuring the geometric mechanism by which constraints shape admissible dynamics. We propose an operator-theoretic formulation in which computational or feasibility limitations are encoded by self-adjoint operators defining locally reachable subspaces. In this setting, the optimal first-order improvement direction emerges as a pseudoinverse-weighted gradient, revealing how constraints induce a distorted ascent geometry. We further demonstrate that effective dynamics concentrate along dominant spectral modes, yielding a principled notion of spectral compression, and establish a compatibility principle that characterizes the existence of common admissible directions across multiple objectives. The resulting framework unifies gradient projection, spectral truncation, and multi-objective feasibility within a single geometric structure.

Selecting an optimization algorithm requires comparing candidates across problem instances, but the computational budget for deployment is often unknown at benchmarking time. Current methods either collapse anytime performance into a scalar, require manual interpretation of plots, or produce conclusions that change when algorithms are added or removed. Moreover, methods based on raw objective values require normalization, which needs bounds or optima that are often unavailable and breaks coherent aggregation across instances. We propose a framework that formulates anytime algorithm comparison as Pareto optimization over time: an algorithm is non-dominated if no competitor beats it at every timepoint. By using rankings rather than objective values, our approach requires no bounds, no normalization, and aggregates coherently across arbitrary instance distributions without requiring known optima. We introduce PolarBear (Pareto-optimal anytime algorithms via Bayesian racing), a procedure that identifies the anytime Pareto set through adaptive sampling with calibrated uncertainty. Bayesian inference over a temporal Plackett-Luce ranking model provides posterior beliefs about pairwise dominance, enabling early elimination of confidently dominated algorithms. The output Pareto set together with the posterior supports downstream algorithm selection under arbitrary time preferences and risk profiles without additional experiments.

We propose neural network operator inference (NN-OpInf): a structure-preserving, composable, and minimally restrictive operator inference framework for the non-intrusive reduced-order modeling of dynamical systems. The approach learns latent dynamics from snapshot data, enforcing local operator structure such as skew-symmetry, (semi-)positive definiteness, and gradient preservation, while also reflecting complex dynamics by supporting additive compositions of heterogeneous operators. We present practical training strategies and analyze computational costs relative to linear and quadratic polynomial OpInf (P-OpInf). Numerical experiments across several nonlinear and parametric problems demonstrate improved accuracy, stability, and robustness over P-OpInf and prior NN-ROM formulations, particularly when the dynamics are not well represented by polynomial models. These results suggest that NN-OpInf can serve as an effective drop-in replacement for P-OpInf when the dynamics to be modeled contain non-polynomial nonlinearities, offering potential gains in accuracy and out-of-distribution performance at the expense of higher training computational costs and a more difficult, non-convex learning problem.

Multimodal large language models (MLLMs) face safety misalignment, where visual inputs enable harmful outputs. To address this, existing methods require explicit safety labels or contrastive data; yet, threat-related concepts are concrete and visually depictable, while safety concepts, like helpfulness, are abstract and lack visual referents. Inspired by the Self-Fulfilling mechanism underlying emergent misalignment, we propose Visual Self-Fulfilling Alignment (VSFA). VSFA fine-tunes vision-language models (VLMs) on neutral VQA tasks constructed around threat-related images, without any safety labels. Through repeated exposure to threat-related visual content, models internalize the implicit semantics of vigilance and caution, shaping safety-oriented personas. Experiments across multiple VLMs and safety benchmarks demonstrate that VSFA reduces the attack success rate, improves response quality, and mitigates over-refusal while preserving general capabilities. Our work extends the self-fulfilling mechanism from text to visual modalities, offering a label-free approach to VLMs alignment.

The surge of highly realistic synthetic videos produced by contemporary generative systems has significantly increased the risk of malicious use, challenging both humans and existing detectors. Against this backdrop, we take a generator-side view and observe that internal cross-attention mechanisms in these models encode fine-grained speech-motion alignment, offering useful correspondence cues for forgery detection. Building on this insight, we propose X-AVDT, a robust and generalizable deepfake detector that probes generator-internal audio-visual signals accessed via DDIM inversion to expose these cues. X-AVDT extracts two complementary signals: (i) a video composite capturing inversion-induced discrepancies, and (ii) an audio-visual cross-attention feature reflecting modality alignment enforced during generation. To enable faithful cross-generator evaluation, we further introduce MMDF, a new multimodal deepfake dataset spanning diverse manipulation types and rapidly evolving synthesis paradigms, including GANs, diffusion, and flow-matching. Extensive experiments demonstrate that X-AVDT achieves leading performance on MMDF and generalizes strongly to external benchmarks and unseen generators, outperforming existing methods with accuracy improved by 13.1%. Our findings highlight the importance of leveraging internal audio-visual consistency cues for robustness to future generators in deepfake detection.

Robotic systems operating in unstructured environments must operate under significant uncertainty arising from intermittent contacts, frictional variability, and unmodeled compliance. While recent model-free approaches have demonstrated impressive performance, many deployment settings still require predictive models that support planning, constraint handling, and online adaptation. Analytical rigid-body models provide strong physical structure but often fail to capture complex interaction effects, whereas purely data-driven models may violate physical consistency, exhibit data bias, and accumulate long-horizon drift. In this work, we propose STRIDE, a dynamics learning framework that explicitly separates conservative rigid-body mechanics from uncertain, effectively stochastic non-conservative interaction effects. The structured component is modeled using a Lagrangian Neural Network (LNN) to preserve energy-consistent inertial dynamics, while residual interaction forces are represented using Conditional Flow Matching (CFM) to capture multi-modal interaction phenomena. The two components are trained jointly end-to-end, enabling the model to retain physical structure while representing complex stochastic behavior. We evaluate STRIDE on systems of increasing complexity, including a pendulum, the Unitree Go1 quadruped, and the Unitree G1 humanoid. Results show 20% reduction in long-horizon prediction error and 30% reduction in contact force prediction error compared to deterministic residual baselines, supporting more reliable model-based control in uncertain robotic environments.

Zero-shot open-vocabulary object navigation has progressed rapidly with the emergence of large Vision-Language Models (VLMs) and Large Language Models (LLMs), now widely used as high-level decision-makers instead of end-to-end policies. Although effective, such systems often rely on iterative large-model queries at inference time, introducing latency and computational overhead that limit real-time deployment. To address this problem, we repurpose ray frontiers (R2F), a recently proposed frontier-based exploration paradigm, to develop an LLM-free framework for indoor open-vocabulary object navigation. While ray frontiers were originally used to bias exploration using semantic cues carried along rays, we reinterpret frontier regions as explicit, direction-conditioned semantic hypotheses that serve as navigation goals. Language-aligned features accumulated along out-of-range rays are stored sparsely at frontiers, where each region maintains multiple directional embeddings encoding plausible unseen content. In this way, navigation then reduces to embedding-based frontier scoring and goal tracking within a classical mapping and planning pipeline, eliminating iterative large-model reasoning. We further introduce R2F-VLN, a lightweight extension for free-form language instructions using syntactic parsing and relational verification without additional VLM or LLM components. Experiments in Habitat-sim and on a real robotic platform demonstrate competitive state-of-the-art zero-shot performance with real-time execution, achieving up to 6 times faster runtime than VLM-based alternatives.

Model-based reinforcement learning (MBRL) is sample-efficient but depends on the accuracy of the learned dynamics, which are often modeled using black-box methods that do not adhere to physical laws. Those methods tend to produce inaccurate predictions when presented with data that differ from the original training set. In this work, we employ Lagrangian neural networks (LNNs), which enforce an underlying Lagrangian structure to train the model within a Dyna-based MBRL framework. Furthermore, we train the LNN using stochastic gradient-based and state-estimation-based optimizers to learn the network's weights. The state-estimation-based method converges faster than the stochastic gradient-based method during neural network training. Simulation results are provided to illustrate the effectiveness of the proposed LNN-based Dyna framework for MBRL.

While Physics-Informed Neural Networks (PINNs) offer a mesh-free approach to solving PDEs, standard point-wise residual minimization suffers from convergence pathologies in topologically complex domains like Triply Periodic Minimal Surfaces (TPMS). The locality bias of point-wise constraints fails to propagate global information through tortuous channels, causing unstable gradients and conservation violations. To address this, we propose the Multi-scale Weak-form PINN (MUSA-PINN), which reformulates PDE constraints as integral conservation laws over hierarchical spherical control volumes. We enforce continuity and momentum conservation via flux-balance residuals on control surfaces. Our method utilizes a three-scale subdomain strategy-comprising large volumes for long-range coupling, skeleton-aware meso-scale volumes aligned with transport pathways, and small volumes for local refinement-alongside a two-stage training schedule prioritizing continuity. Experiments on steady incompressible flow in TPMS geometries show MUSA-PINN outperforms state-of-the-art baselines, reducing relative errors by up to 93% and preserving mass conservation.

Chain-of-Thought (CoT) prompting improves LLM accuracy on complex tasks but often increases token usage and inference cost. Existing "Budget Forcing" methods reducing cost via fine-tuning with heuristic length penalties, suppress both essential reasoning and redundant filler. We recast efficient reasoning as a lossy compression problem under the Information Bottleneck (IB) principle, and identify a key theoretical gap when applying naive IB to transformers: attention violates the Markov property between prompt, reasoning trace, and response. To resolve this issue, we model CoT generation under the Conditional Information Bottleneck (CIB) principle, where the reasoning trace Z acts as a computational bridge that contains only the information about the response Y that is not directly accessible from the prompt X. This yields a general Reinforcement Learning objective: maximize task reward while compressing completions under a prior over reasoning traces, subsuming common heuristics (e.g., length penalties) as special cases (e.g., uniform priors). In contrast to naive token-counting-based approaches, we introduce a semantic prior that measures token cost by surprisal under a language model prior. Empirically, our CIB objective prunes cognitive bloat while preserving fluency and logic, improving accuracy at moderate compression and enabling aggressive compression with minimal accuracy drop.

Safe predictions are a crucial requirement for integrating predictive models into clinical decision support systems. One approach for ensuring trustworthiness is to enable models' ability to express their uncertainty about individual predictions. However, current machine learning models frequently lack reliable uncertainty estimation, hindering real-world deployment. This is further observed in multimodal settings, where the goal is to enable effective information fusion. In this work, we propose $\texttt{MedCertAIn}$, a predictive uncertainty framework that leverages multimodal clinical data for in-hospital risk prediction to improve model performance and reliability. We design data-driven priors over neural network parameters using a hybrid strategy that considers cross-modal similarity in self-supervised latent representations and modality-specific data corruptions. We train and evaluate the models with such priors using clinical time-series and chest X-ray images from the publicly-available datasets MIMIC-IV and MIMIC-CXR. Our results show that $\texttt{MedCertAIn}$ significantly improves predictive performance and uncertainty quantification compared to state-of-the-art deterministic baselines and alternative Bayesian methods. These findings highlight the promise of data-driven priors in advancing robust, uncertainty-aware AI tools for high-stakes clinical applications.

Robust single-vessel tracking from fixed coastal platforms is hindered by modality-specific degradations: cameras suffer from illumination and visual clutter, while LiDAR performance drops with range and intermittent returns. We present a heterogeneous multi-sensor fusion particle-filter tracker that incorporates an information-gain (entropy-reduction) adaptive sensing policy to select the most informative configuration at each fusion time bin. The approach is validated in a real maritime deployment at the CMMI Smart Marina Testbed (Ayia Napa Marina, Cyprus), using a shore-mounted 3D LiDAR and an elevated fixed camera to track a rigid inflatable boat with onboard GNSS ground truth. We compare LiDAR-only, camera-only, all-sensors, and adaptive configurations. Results show LiDAR dominates near-field accuracy, the camera sustains longer-range coverage when LiDAR becomes unavailable, and the adaptive policy achieves a favorable accuracy-continuity trade-off by switching modalities based on information gain. By avoiding continuous multi-stream processing, the adaptive configuration provides a practical baseline for resilient and resource-aware maritime surveillance.

When an RL agent's observations are gradually corrupted, at what drift rate does it "wake up" -- and what determines this boundary? We study world model-based self-monitoring under continuous observation drift across four MuJoCo environments, three detector families (z-score, variance, percentile), and three model capacities. We find that (1) a sharp detection threshold $\varepsilon^*$ exists universally: below it, drift is absorbed as normal variation; above it, detection occurs rapidly. The threshold's existence and sigmoid shape are invariant across all detector families and model capacities, though its position depends on the interaction between detector sensitivity, noise floor structure, and environment dynamics. (2) Sinusoidal drift is completely undetectable by all detector families -- including variance and percentile detectors with no temporal smoothing -- establishing this as a world model property rather than a detector artifact. (3) Within each environment, $\varepsilon^*$ follows a power law in detector parameters ($R^2 = 0.89$-$0.97$), but cross-environment prediction fails ($R^2 = 0.45$), revealing that the missing variable is environment-specific dynamics structure $\partial \mathrm{PE}/\partial\varepsilon$. (4) In fragile environments, agents collapse before any detector can fire ("collapse before awareness"), creating a fundamentally unmonitorable failure mode. Our results reframe $\varepsilon^*$ from an emergent world model property to a three-way interaction between noise floor, detector, and environment dynamics, providing a more defensible and empirically grounded account of self-monitoring boundaries in RL agents.

The quadratic complexity of the attention mechanism and the substantial memory footprint of the Key-Value (KV) cache present severe computational and memory challenges for Large Language Models (LLMs) processing long contexts. Existing retrieval-based methods often compromise semantic integrity through fixed-size chunking and suffer from inefficient linear scanning. In this paper, we propose LycheeCluster, a novel method for efficient KV cache management. LycheeCluster preserves local semantic coherence via boundary-aware chunking and constructs a recursive hierarchical index rooted in the triangle inequality. This design transforms cache retrieval from a linear scan into a theoretically bounded, logarithmic-time pruning process, while a lazy update strategy supports efficient streaming generation. Experiments demonstrate that LycheeCluster achieves up to a 3.6x end-to-end inference speedup with negligible degradation in model performance, outperforming state-of-the-art KV cache management methods (e.g., Quest, ClusterKV). We will release our code and kernels after publication.

Translations often carry traces of the source language, a phenomenon known as translationese. We introduce the first freely available English-to-Swedish dataset contrasting translationese sentences with idiomatic alternatives, designed to probe intrinsic preferences of language models. It includes error tags and descriptions of the problems in the original translations. In experiments evaluating smaller Swedish and multilingual LLMs with our dataset, we find that they often favor the translationese phrasing. Human alternatives are chosen more often when the English source sentence is omitted, indicating that exposure to the source biases models toward literal translations, although even without context models often prefer the translationese variant. Our dataset and findings provide a resource and benchmark for developing models that produce more natural, idiomatic output in non-English languages.

Large language model (LLM)-based AI systems have shown promise for patient-facing diagnostic and management conversations in simulated settings. Translating these systems into clinical practice requires assessment in real-world workflows with rigorous safety oversight. We report a prospective, single-arm feasibility study of an LLM-based conversational AI, the Articulate Medical Intelligence Explorer (AMIE), conducting clinical history taking and presentation of potential diagnoses for patients to discuss with their provider at urgent care appointments at a leading academic medical center. 100 adult patients completed an AMIE text-chat interaction up to 5 days before their appointment. We sought to assess the conversational safety and quality, patient and clinician experience, and clinical reasoning capabilities compared to primary care providers (PCPs). Human safety supervisors monitored all patient-AMIE interactions in real time and did not need to intervene to stop any consultations based on pre-defined criteria. Patients reported high satisfaction and their attitudes towards AI improved after interacting with AMIE (p < 0.001). PCPs found AMIE's output useful with a positive impact on preparedness. AMIE's differential diagnosis (DDx) included the final diagnosis, per chart review 8 weeks post-encounter, in 90% of cases, with 75% top-3 accuracy. Blinded assessment of AMIE and PCP DDx and management (Mx) plans suggested similar overall DDx and Mx plan quality, without significant differences for DDx (p = 0.6) and appropriateness and safety of Mx (p = 0.1 and 1.0, respectively). PCPs outperformed AMIE in the practicality (p = 0.003) and cost effectiveness (p = 0.004) of Mx. While further research is needed, this study demonstrates the initial feasibility, safety, and user acceptance of conversational AI in a real-world setting, representing crucial steps towards clinical translation.

The Uncertain Agile Earth Observation Satellite Scheduling Problem (UAEOSSP) is a novel combinatorial optimization problem and a practical engineering challenge that aligns with the current demands of space technology development. It incorporates uncertainties in profit, resource consumption, and visibility, which may render pre-planned schedules suboptimal or even infeasible. Genetic Programming Hyper-Heuristic (GPHH) shows promise for evolving interpretable scheduling policies; however, their simulation-based evaluation incurs high computational costs. Moreover, the design of the constructive method, denoted as Online Scheduling Algorithm (OSA), directly affects fitness assessment, resulting in evaluation-dependent local optima within the policy space. To address these issues, this paper proposes a Hybrid Evaluation-based Genetic Programming (HE-GP) for effectively solving UAEOSSP. A Hybrid Evaluation (HE) mechanism is integrated into the policy-driven OSA, combining exact and approximate filtering modes: exact mode ensures evaluation accuracy through elaborately designed constraint verification modules, while approximate mode reduces computational overhead via simplified logic. HE-GP dynamically switches between evaluation models based on real-time evolutionary state information. Experiments on 16 simulated instance sets demonstrate that HE-GP significantly outperforms handcrafted heuristics and single-evaluation based GPHH, achieving substantial reductions in computational cost while maintaining excellent scheduling performance across diverse scenarios. Specifically, the average training time of HE-GP was reduced by 17.77\% compared to GP employing exclusively exact evaluation, while the optimal policy generated by HE-GP achieved the highest average ranks across all scenarios.

Visual entity tracking is an innate cognitive ability in humans, yet it remains a critical bottleneck for Vision-Language Models (VLMs). This deficit is often obscured in existing video benchmarks by visual shortcuts. We introduce VET-Bench, a synthetic diagnostic testbed featuring visually identical objects that necessitate tracking exclusively through spatiotemporal continuity. Our experiments reveal that current state-of-the-art VLMs perform at or near chance level on VET-Bench, exposing a fundamental limitation: an over-reliance on static frame-level features and a failure to maintain entity representations over time. We provide a theoretical analysis drawing connections to the state-tracking problem, proving that fixed-depth transformer-based VLMs are fundamentally limited in tracking indistinguishable objects without intermediate supervision due to expressivity constraints. To address this, we propose Spatiotemporal Grounded Chain-of-Thought (SGCoT): generating object trajectories as explicit intermediate states. Leveraging Molmo2's object tracking ability, we elicit SGCoT reasoning by fine-tuning on synthesized text-only data for alignment. Our method achieves state-of-the-art accuracy exceeding 90% on VET-Bench, demonstrating that VLMs can reliably solve the video shell-game task end-to-end without external tools. Our code and data are available at https://vetbench.github.io .

LLM agents that retrieve external knowledge typically generate a search query as text, then run a separate embedding model to encode it into a vector. This two-model pipeline adds infrastructure complexity and latency, yet is redundant: the LLM already encodes the full conversational context in its hidden states. We propose equipping LLM agents with native retrieval capability by adding a lightweight projection head that maps hidden states directly into the embedding space, eliminating the need for a separate embedding model. Trained with a combination of alignment, contrastive, and rank distillation losses, our method retains 97\% of baseline retrieval quality while enabling the LLM agent to search with its own representations. Experiments on the QReCC conversational search benchmark show competitive Recall@10 and MRR@10 compared to the standard generate-then-encode pipeline, with systematic ablations confirming the contribution of each loss component.

Class Incremental Learning (CIL) poses a fundamental challenge: maintaining a balance between the plasticity required to learn new tasks and the stability needed to prevent catastrophic forgetting. While expansion-based methods effectively mitigate forgetting by adding task-specific parameters, they suffer from uncontrolled architectural growth and memory overhead. In this paper, we propose a novel dynamic scaling framework that adaptively manages model capacity through a cyclic "GRow, Assess, ComprEss" (GRACE) strategy. Crucially, we supplement backbone expansion with a novel saturation assessment phase that evaluates the utilization of the model's capacity. This assessment allows the framework to make informed decisions to either expand the architecture or compress the backbones into a streamlined representation, preventing parameter explosion. Experimental results demonstrate that our approach achieves state-of-the-art performance across multiple CIL benchmarks, while reducing memory footprint by up to a 73% compared to purely expansionist models.

This paper presents IronEngine, a general AI assistant platform organized around a unified orchestration core that connects a desktop user interface, REST and WebSocket APIs, Python clients, local and cloud model backends, persistent memory, task scheduling, reusable skills, 24-category tool execution, MCP-compatible extensibility, and hardware-facing integration. IronEngine introduces a three-phase pipeline -- Discussion (Planner--Reviewer collaboration), Model Switch (VRAM-aware transition), and Execution (tool-augmented action loop) -- that separates planning quality from execution capability. The system features a hierarchical memory architecture with multi-level consolidation, a vectorized skill repository backed by ChromaDB, an adaptive model management layer supporting 92 model profiles with VRAM-aware context budgeting, and an intelligent tool routing system with 130+ alias normalization and automatic error correction. We present experimental results on file operation benchmarks achieving 100\% task completion with a mean total time of 1541 seconds across four heterogeneous tasks, and provide detailed comparisons with representative AI assistant systems including ChatGPT, Claude Desktop, Cursor, Windsurf, and open-source agent frameworks. Without disclosing proprietary prompts or core algorithms, this paper analyzes the platform's architectural decomposition, subsystem design, experimental performance, safety boundaries, and comparative engineering advantages. The resulting study positions IronEngine as a system-oriented foundation for general-purpose personal assistants, automation frameworks, and future human-centered agent platforms.

In recent years, Artificial Intelligence has become a powerful partner for complex tasks such as data analysis, prediction, and problem-solving, yet its lack of transparency raises concerns about its reliability. In sensitive domains such as healthcare or cybersecurity, ensuring transparency, trustworthiness, and robustness is essential, since the consequences of wrong decisions or successful attacks can be severe. Prior neuron-level interpretability approaches are primarily descriptive, task-dependent, or require retraining, which limits their use as systematic, reusable tools for evaluating internal robustness across architectures and domains. To overcome these limitations, this work proposes SYNAPSE, a systematic, training-free framework for understanding and stress-testing the internal behavior of Transformer models across domains. It extracts per-layer [CLS] representations, trains a lightweight linear probe to obtain global and per-class neuron rankings, and applies forward-hook interventions during inference. This design enables controlled experiments on internal representations without altering the original model, thereby allowing weaknesses, stability patterns, and label-specific sensitivities to be measured and compared directly across tasks and architectures. Across all experiments, SYNAPSE reveals a consistent, domain-independent organization of internal representations, in which task-relevant information is encoded in broad, overlapping neuron subsets. This redundancy provides a strong degree of functional stability, while class-wise asymmetries expose heterogeneous specialization patterns and enable label-aware analysis. In contrast, small structured manipulations in weight or logit space are sufficient to redirect predictions, highlighting complementary vulnerability profiles and illustrating how SYNAPSE can guide the development of more robust Transformer models.

Self-driving laboratories (SDLs) are rapidly transforming research in chemistry and materials science to accelerate new discoveries. Mobile robot chemists (MRCs) play a pivotal role by autonomously navigating the lab to transport samples, effectively connecting synthesis, analysis, and characterisation equipment. The instruments within an SDL are typically designed or retrofitted to be accessed by both human and robotic chemists, ensuring operational flexibility and integration between manual and automated workflows. In many scenarios, human and robotic chemists may need to use the same equipment simultaneously. Currently, MRCs rely on simple LiDAR-based obstruction detection, which forces the robot to passively wait if a human is present. This lack of situational awareness leads to unnecessary delays and inefficient coordination in time-critical automated workflows in human-robot shared labs. To address this, we present an initial study of an embodied, AI-driven perception method that facilitates proactive human-robot interaction in shared-access scenarios. Our method features a hierarchical human intention prediction model that allows the robot to distinguish between preparatory actions (waiting) and transient interactions (accessing the instrument). Our results demonstrate that the proposed approach enhances efficiency by enabling proactive human-robot interaction, streamlining coordination, and potentially increasing the efficiency of autonomous scientific labs.

Meta-Reinforcement Learning addresses the critical limitations of conventional Reinforcement Learning in multi-task and non-stationary environments by enabling fast policy adaptation and improved generalization. We introduce a novel Meta-RL framework that integrates a bi-level optimization scheme with a hybrid actor-critic architecture specially designed to enhance sample efficiency and inter-task adaptability. To improve knowledge transfer, we meta-learn a shared state feature extractor jointly optimized across actor and critic networks, providing efficient representation learning and limiting overfitting to individual tasks or dominant profiles. Additionally, we propose a parameter-sharing mechanism between the outer- and inner-loop actor networks, to reduce redundant learning and accelerate adaptation during task revisitation. The approach is validated on a real-world Building Energy Management Systems dataset covering nearly a decade of temporal and structural variability, for which we propose a task preparation method to promote generalization. Experiments demonstrate effective task adaptation and better performance compared to conventional RL and Meta-RL methods.

Deep neural networks for image classification often exhibit overconfidence on out-of-distribution (OOD) samples. To address this, we introduce Geometrically Constrained Outlier Synthesis (GCOS), a training-time regularization framework aimed at improving OOD robustness during inference. GCOS addresses a limitation of prior synthesis methods by generating virtual outliers in the hidden feature space that respect the learned manifold structure of in-distribution (ID) data. The synthesis proceeds in two stages: (i) a dominant-variance subspace extracted from the training features identifies geometrically informed, off-manifold directions; (ii) a conformally-inspired shell, defined by the empirical quantiles of a nonconformity score from a calibration set, adaptively controls the synthesis magnitude to produce boundary samples. The shell ensures that generated outliers are neither trivially detectable nor indistinguishable from in-distribution data, facilitating smoother learning of robust features. This is combined with a contrastive regularization objective that promotes separability of ID and OOD samples in a chosen score space, such as Mahalanobis or energy-based. Experiments demonstrate that GCOS outperforms state-of-the-art methods using standard energy-based inference on near-OOD benchmarks, defined as tasks where outliers share the same semantic domain as in-distribution data. As an exploratory extension, the framework naturally transitions to conformal OOD inference, which translates uncertainty scores into statistically valid p-values and enables thresholds with formal error guarantees, providing a pathway toward more predictable and reliable OOD detection.

Reinforcement Learning from Human Feedback (RLHF) assumes annotator preferences reflect stable internal states. We challenge this through three experiments spanning the preference pipeline. In a human choice blindness study, 91% of surreptitiously swapped preferences go undetected, extending choice blindness to third-person evaluative comparison of unfamiliar text. Testing fifteen LLM judges as potential replacements, we find detection relies on shallow text matching rather than genuine self-monitoring: removing prior reasoning from context causes blindness to surge from near-zero to over 50%, while explicit social pressure induces near-universal compliance. In a dose-response experiment across two architectures from 86M to 2B parameters, one-sixth to one-third of labels must be corrupted before the reward signal halves, yet standard pairwise accuracy remains virtually unchanged. A Best-of-N evaluation confirms this translates to downstream policy degradation: at 50% corruption, reward-guided selection produces no improvement over random sampling, while the proxy model reports monotonically increasing scores. Together, these results reveal a preference construction problem: the signal entering RLHF is shaped by elicitation context in ways that neither human metacognition, LLM self-monitoring, nor standard evaluation metrics can detect.

Digital educational environments are expanding toward complex AI and human discourse, providing researchers with an abundance of data that offers deep insights into learning and instructional processes. However, traditional qualitative analysis remains a labor-intensive bottleneck, severely limiting the scale at which this research can be conducted. We present Sandpiper, a mixed-initiative system designed to serve as a bridge between high-volume conversational data and human qualitative expertise. By tightly coupling interactive researcher dashboards with agentic Large Language Model (LLM) engines, the platform enables scalable analysis without sacrificing methodological rigor. Sandpiper addresses critical barriers to AI adoption in education by implementing context-aware, automated de-identification workflows supported by secure, university-housed infrastructure to ensure data privacy. Furthermore, the system employs schema-constrained orchestration to eliminate LLM hallucinations and enforces strict adherence to qualitative codebooks. An integrated evaluations engine allows for the continuous benchmarking of AI performance against human labels, fostering an iterative approach to model refinement and validation. We propose a user study to evaluate the system's efficacy in improving research efficiency, inter-rater reliability, and researcher trust in AI-assisted qualitative workflows.

Despite remarkable achievements in single-agent offline reinforcement learning (RL), multi-agent RL (MARL) has struggled to adopt this paradigm, largely persisting with on-policy training and self-play from scratch. One reason for this gap comes from the instability of non-linear value decomposition, leading prior works to avoid complex mixing networks in favor of linear value decomposition (e.g., VDN) with value regularization used in single-agent setups. In this work, we analyze the source of instability in non-linear value decomposition within the offline MARL setting. Our observations confirm that they induce value-scale amplification and unstable optimization. To alleviate this, we propose a simple technique, scale-invariant value normalization (SVN), that stabilizes actor-critic training without altering the Bellman fixed point. Empirically, we examine the interaction among key components of offline MARL (e.g., value decomposition, value learning, and policy extraction) and derive a practical recipe that unlocks its full potential.

In this work, we reveal that Large Language Models (LLMs) possess intrinsic behavioral plasticity-akin to chameleons adapting their coloration to environmental cues-that can be exposed through token-conditional generation and stabilized via reinforcement learning. Specifically, by conditioning generation on carefully selected token prefixes sampled from responses exhibiting desired behaviors, LLMs seamlessly adapt their behavioral modes at inference time (e.g., switching from step-by-step reasoning to direct answering) without retraining. Based on this insight, we propose Token-Conditioned Reinforcement Learning (ToCoRL), a principled framework that leverages RL to internalize this chameleon-like plasticity, transforming transient inference-time adaptations into stable and learnable behavioral patterns. ToCoRL guides exploration with token-conditional generation and keep enhancing exploitation, enabling emergence of appropriate behaviors. Extensive experiments show that ToCoRL enables precise behavioral control without capability degradation. Notably, we show that large reasoning models, while performing strongly on complex mathematics, can be effectively adapted to excel at factual question answering, which was a capability previously hindered by their step-by-step reasoning patterns.

Systems that collect data on sleep, mood, and activities can provide valuable lifestyle counselling to populations affected by chronic disease and its consequences. Such systems are, however, challenging to develop; besides reliably extracting patterns from user-specific data, systems should also contextualise these patterns with validated medical knowledge to ensure the quality of counselling, and generate counselling that is relevant to a real user. We present QUORUM, a new evaluation framework that unifies these developer-, expert-, and user-centric perspectives, and show with a real case study that it meaningfully tracks convergence and divergence in stakeholder perspectives. We also present COACH, a Large Language Model-driven pipeline to generate personalised lifestyle counselling for our Healthy Chronos use case, a diary app for cancer patients and survivors. Applying our framework shows that overall, users, medical experts, and developers converge on the opinion that the generated counselling is relevant, of good quality, and reliable. However, stakeholders also diverge on the tone of the counselling, sensitivity to errors in pattern-extraction, and potential hallucinations. These findings highlight the importance of multi-stakeholder evaluation for consumer health language technologies and illustrate how a unified evaluation framework can support trustworthy, patient-centered NLP systems in real-world settings.

Chain-of-thought (CoT) prompting enables reasoning in language models but requires explicit verbalization of intermediate steps. Looped transformers offer an alternative by iteratively refining representations within hidden states. This parameter efficiency comes at a cost, as looped models lack the storage capacity of deeper models which use unique weights per layer. In this work, we investigate transformer models that feature both adaptive per-layer looping, where each transformer block learns to iterate its hidden state via a learned halting mechanism, and gated memory banks, that provide additional learned storage. We find that looping primarily benefits mathematical reasoning, while memory banks help recover performance on commonsense tasks compared to parameter and FLOP matched models. Combining both mechanisms yields a model that outperforms an iso-FLOP baseline, with three times the number of layers, across math benchmarks. Analysis of model internals reveals layer specialization: early layers learn to loop minimally and access memory sparingly, while later layers do both more heavily.

We propose a Hierarchical Error-Corrective Graph FrameworkforAutonomousAgentswithLLM-BasedActionGeneration(HECG),whichincorporates three core innovations: (1) Multi-Dimensional Transferable Strategy (MDTS): by integrating task quality metrics (Q), confidence/cost metrics (C), reward metrics (R), and LLM-based semantic reasoning scores (LLM-Score), MDTS achieves multi-dimensional alignment between quantitative performance and semantic context, enabling more precise selection of high-quality candidate strate gies and effectively reducing the risk of negative transfer. (2) Error Matrix Classification (EMC): unlike simple confusion matrices or overall performance metrics, EMC provides structured attribution of task failures by categorizing errors into ten types, such as Strategy Errors (Strategy Whe) and Script Parsing Errors (Script-Parsing-Error), and decomposing them according to severity, typical actions, error descriptions, and recoverability. This allows precise analysis of the root causes of task failures, offering clear guidance for subsequent error correction and strategy optimization rather than relying solely on overall success rates or single performance metrics. (3) Causal-Context Graph Retrieval (CCGR): to enhance agent retrieval capabilities in dynamic task environments, we construct graphs from historical states, actions, and event sequences, where nodes store executed actions, next-step actions, execution states, transferable strategies, and other relevant information, and edges represent causal dependencies such as preconditions for transitions between nodes. CCGR identifies subgraphs most relevant to the current task context, effectively capturing structural relationships beyond vector similarity, allowing agents to fully leverage contextual information, accelerate strategy adaptation, and improve execution reliability in complex, multi-step tasks.

Standard Gradient Descent and its modern variants assume local, Markovian weight updates, making them highly susceptible to noise and overfitting. This limitation becomes critically severe in extremely imbalanced datasets such as financial fraud detection where dominant class gradients systematically overwrite the subtle signals of the minority class. In this paper, we introduce a novel optimization algorithm grounded in Fractional Calculus. By isolating the core memory engine of the generalized fractional derivative, the Weighted Fractional Weyl Integral, we replace the instantaneous gradient with a dynamically weighted historical sequence. This fractional memory operator acts as a natural regularizer. Empirical evaluations demonstrate that our method prevents overfitting in medical diagnostics and achieves an approximately 40 percent improvement in PR-AUC over classical optimizers in financial fraud detection, establishing a robust bridge between pure fractional topology and applied Machine Learning.

The digital markets act (DMA) regulates very large digital platforms like Meta's Facebook or Apple's iOS with the goal to promote fairness, contestability (of market power) and user choice. From a system design or broader technical perspective, the implications of the DMA have not been studied so far. Using systematic methods from qualitative coding and thematic analysis, we investigate the DMA from a technical perspective and derive eight high-level design strategies that serve as fundamental approaches towards value-based architectural goals like 'fair practice', or 'user choice' (as envisioned by the DMA). We investigate how compliance with the DMA has been achieved and derive 15 tactics that we map to our strategies. While the DMA obligations challenge existing platform designs, they also create new opportunities for designing services within these huge ecosystems. We, thus, discuss our strategies in light of both. We see this work as a first step towards filling this pressing gap in the architecture of platform ecosystems, i.e., how to incorporate abstract human values in architecture design.

Influential benchmarks incentivize competing model developers to strategically allocate post-training resources toward improvements on the leaderboard, a phenomenon dubbed benchmaxxing or training on the test task. In this work, we initiate a principled study of the incentive structure that benchmarks induce. We model benchmarking as a Stackelberg game between a benchmark designer who chooses an evaluation protocol and multiple model developers who compete simultaneously in a subgame given by the designer's choice. Each competitor has a model of unknown latent quality and can inflate its observed score by allocating resources to benchmark-specific improvements. First, we prove that current benchmarks induce games for which no Nash equilibrium between model developers exists. This result suggests one explanation for why current practice leads to misaligned incentives, prompting model developers to strategize in opaque ways. However, we prove that under mild conditions, a recently proposed evaluation protocol, called tune-before-test, induces a benchmark with a unique Nash equilibrium that ranks models by latent quality. This positive result demonstrates that benchmarks need not set bad incentives, even if current evaluations do.

Continuous and efficient experimentation is key to the practical success of user-facing applications on the web, both through online A/B-tests and off-policy evaluation. Despite their shared objective -- estimating the incremental value of a treatment -- these domains often operate in isolation, utilising distinct terminologies and statistical toolkits. This paper bridges that divide by establishing a formal equivalence between their canonical variance reduction methods. We prove that the standard online Difference-in-Means estimator is mathematically identical to an off-policy Inverse Propensity Scoring estimator equipped with an optimal (variance-minimising) additive control variate. Extending this unification, we demonstrate that widespread regression adjustment methods (such as CUPED, CUPAC, and ML-RATE) are structurally equivalent to Doubly Robust estimation. This unified view extends our understanding of commonly used approaches, and can guide practitioners and researchers working on either class of problems.

Multimodal large language models have recently shown promising progress in visual mathematical reasoning. However, their performance is often limited by a critical yet underexplored bottleneck: inaccurate visual perception. Through systematic analysis, we find that the most failures originate from incorrect or incomplete visual evidence extraction rather than deficiencies in reasoning capability. Moreover, models tend to remain overly confident in their initial perceptions, making standard strategies such as prompt engineering, multi-round self-reflection, or posterior guidance insufficient to reliably correct errors. To address this limitation, we propose M3-ACE, a multi-agentic context engineering framework designed to rectify visual perception in multimodal math reasoning. Instead of directly aggregating final answers, our approach decouples perception and reasoning by dynamically maintaining a shared context centered on visual evidence lists. Multiple agents collaboratively contribute complementary observations, enabling the system to expose inconsistencies and recover missing perceptual information. To support stable multi-turn collaboration, we further introduce two lightweight tools: a Summary Tool that organizes evidence from different agents into consistent, complementary, and conflicting components, and a Refine Tool that filters unreliable samples and guides iterative correction. Extensive experiments demonstrate that M3-ACE substantially improves visual mathematical reasoning performance across multiple benchmarks. Our method establishes new state-of-the-art results 89.1 on the MathVision benchmark and achieves consistent improvements on other related datasets, including MathVista and MathVerse. These results highlight the importance of perception-centric multi-agent collaboration for advancing multimodal reasoning systems.

Learning to understand speech appears almost effortless for typically developing infants, yet from an information-processing perspective, acquiring a language from acoustic speech is an enormous challenge. This chapter reviews recent developments in using computational models to understand early language acquisition from speech and audiovisual input. The focus is on self-supervised and visually grounded models of perceptual learning. We show how these models are becoming increasingly powerful in learning various aspects of speech without strong linguistic priors, and how many features of early language development can be explained through a shared set of learning principles-principles broadly compatible with multiple theories of language acquisition and human cognition. We also discuss how modern learning simulations are gradually becoming more realistic, both in terms of input data and in linking model behavior to empirical findings on infant language development.

We investigate how language models handle the proviso problem, an unresolved issue in pragmatics where presuppositions in conditional sentences diverge between theoretical and human interpretations. We reformulate this phenomenon as a Natural Language Inference task and introduce a diagnostic dataset designed to probe presupposition projection in conditionals. We evaluate RoBERTa, DeBERTa, LLaMA, and Gemma using explainability analyses. The results show that models broadly align with human judgments but rely on shallow pattern matching rather than semantic or pragmatic reasoning. Our work provides the first computational evaluation framework for the proviso problem and highlights the need for diagnostic, multi-method approaches to assess pragmatic competence and context-dependent meaning in language models.

We present a new method for generating plausible counterfactual explanations for time series classification problems. The approach performs gradient-based optimization directly in the input space. To enforce plausibility, we integrate soft-DTW (dynamic time warping) alignment with $k$-nearest neighbors from the target class, which effectively encourages the generated counterfactuals to adopt a realistic temporal structure. The overall optimization objective is a multi-faceted loss function that balances key counterfactual properties. It incorporates losses for validity, sparsity, and proximity, alongside the novel soft-DTW-based plausibility component. We conduct an evaluation of our method against several strong reference approaches, measuring the key properties of the generated counterfactuals across multiple dimensions. The results demonstrate that our method achieves competitive performance in validity while significantly outperforming existing approaches in distributional alignment with the target class, indicating superior temporal realism. Furthermore, a qualitative analysis highlights the critical limitations of existing methods in preserving realistic temporal structure. This work shows that the proposed method consistently generates counterfactual explanations for time series classifiers that are not only valid but also highly plausible and consistent with temporal patterns.

The dense output projection in multi-head attention scales quadratically with model dimension, contributing significantly to parameter count, memory footprint, and inference cost. We propose replacing this projection with a fixed, parameter-free Walsh Hadamard Transform followed by a lightweight learnable affine rescaling, eliminating approximately 25 percent of attention parameters per block while preserving global cross head interaction through an orthogonal, norm-preserving transformation. Across different model sizes, we demonstrate that this structured substitution maintains comparable or slightly superior downstream task performance on standard benchmarks, while achieving up to 7 percent aggregate parameter reduction, 8.9 percent peak memory savings, and 6.6 percent throughput improvement at scale, with efficiency gains growing monotonically with model size, batch size, and sequence length. Interestingly, we observe that structured Hadamard-based models exhibit a steeper validation loss curve relative to training FLOPs compared to their dense counterparts, suggesting more favorable compute utilization during training.

Electrocardiogram (ECG) analysis is vital for detecting cardiac abnormalities, yet robust automated classification is challenging due to the complexity and variability of physiological signals. In this work, we investigate transformer-based ECG classification using features derived from the Koopman operator and wavelet transforms. Two tasks are studied: (1) binary classification (Normal vs. Non-normal), and (2) four-class classification (Normal, Atrial Fibrillation, Ventricular Arrhythmia, Block). We use Extended Dynamic Mode Decomposition (EDMD) to approximate the Koopman operator. Our results show that wavelet features excel in binary classification, while Koopman features, when paired with transformers, achieve superior performance in the four-class setting. A simple hybrid of Koopman and wavelet features does not improve accuracy. However, selecting an appropriate EDMD dictionary -- specifically a radial basis function dictionary with tuned parameters -- yields significant gains, surpassing the wavelet-only baseline and the hybrid wavelet-Koopman system. We also present a Koopman-based reconstruction analysis for interpretable insights into the learned dynamics and compare against a recurrent neural network baseline. Overall, our findings demonstrate the effectiveness of Koopman-based feature learning with transformers and highlight promising directions for integrating dynamical systems theory into time-series classification.

The proliferation of fake reviews, often produced by organized groups, undermines consumer trust and fair competition on online platforms. These groups employ sophisticated strategies that evade traditional detection methods, particularly in cold-start scenarios involving newly launched products with sparse data. To address this, we propose the \underline{D}iversity- and \underline{S}imilarity-aware \underline{D}ynamic \underline{G}raph \underline{A}ttention-enhanced \underline{G}raph \underline{C}onvolutional \underline{N}etwork (DS-DGA-GCN), a new graph learning model for detecting fake reviewer groups. DS-DGA-GCN achieves robust detection since it focuses on the joint relationships among products, reviews, and reviewers by modeling product-review-reviewer networks. DS-DGA-GCN also achieves adaptive detection by integrating a Network Feature Scoring (NFS) system and a new dynamic graph attention mechanism. The NFS system quantifies network attributes, including neighbor diversity, network self-similarity, as a unified feature score. The dynamic graph attention mechanism improves the adaptability and computational efficiency by captures features related to temporal information, node importance, and global network structure. Extensive experiments conducted on two real-world datasets derived from Amazon and Xiaohongshu demonstrate that DS-DGA-GCN significantly outperforms state-of-the-art baselines, achieving accuracies of up to \textbf{89.8\% and 88.3\%}, respectively.

Answering complex, real-world queries often requires synthesizing facts scattered across vast document corpora. In these settings, standard retrieval-augmented generation (RAG) pipelines suffer from incomplete evidence coverage, while long-context large language models (LLMs) struggle to reason reliably over massive inputs. We introduce SPD-RAG, a hierarchical multi-agent framework for exhaustive cross-document question answering that decomposes the problem along the document axis. Each document is processed by a dedicated document-level agent operating only on its own content, enabling focused retrieval, while a coordinator dispatches tasks to relevant agents and aggregates their partial answers. Agent outputs are synthesized by merging partial answers through a token-bounded synthesis layer (which supports recursive map-reduce for massive corpora). This document-level specialization with centralized fusion improves scalability and answer quality in heterogeneous multidocument settings while yielding a modular, extensible retrieval pipeline. On the LOONG benchmark (EMNLP 2024) for long-context multi-document QA, SPD-RAG achieves an Avg Score of 58.1 (GPT-5 evaluation), outperforming Normal RAG (33.0) and Agentic RAG (32.8) while using only 38% of the API cost of a full-context baseline (68.0).

Multiple instance learning (MIL) has enabled substantial progress in computational histopathology, where a large amount of patches from gigapixel whole slide images are aggregated into slide-level predictions. Heatmaps are widely used to validate MIL models and to discover tissue biomarkers. Yet, the validity of these heatmaps has barely been investigated. In this work, we introduce a general framework for evaluating the quality of MIL heatmaps without requiring additional labels. We conduct a large-scale benchmark experiment to assess six explanation methods across histopathology task types (classification, regression, survival), MIL model architectures (Attention-, Transformer-, Mamba-based), and patch encoder backbones (UNI2, Virchow2). Our results show that explanation quality mostly depends on MIL model architecture and task type, with perturbation ("Single"), layer-wise relevance propagation (LRP), and integrated gradients (IG) consistently outperforming attention-based and gradient-based saliency heatmaps, which often fail to reflect model decision mechanisms. We further demonstrate the advanced capabilities of the best-performing explanation methods: (i) We provide a proof-of-concept that MIL heatmaps of a bulk gene expression prediction model can be correlated with spatial transcriptomics for biological validation, and (ii) showcase the discovery of distinct model strategies for predicting human papillomavirus (HPV) infection from head and neck cancer slides. Our work highlights the importance of validating MIL heatmaps and establishes that improved explainability can enable more reliable model validation and yield biological insights, making a case for a broader adoption of explainable AI in digital pathology. Our code is provided in a public GitHub repository: https://github.com/bifold-pathomics/xMIL/tree/xmil-journal

Robot-assisted endoluminal procedures are increasingly used for early cancer intervention. However, the intricate, narrow and tortuous pathways within the luminal anatomy pose substantial difficulties for robot navigation. Vision-based navigation offers a promising solution, but existing localization approaches are error-prone due to tissue deformation, in vivo artifacts and a lack of distinctive landmarks for consistent localization. This paper presents a novel EndoSERV localization method to address these challenges. It includes two main parts, \textit{i.e.}, \textbf{SE}gment-to-structure and \textbf{R}eal-to-\textbf{V}irtual mapping, and hence the name. For long-range and complex luminal structures, we divide them into smaller sub-segments and estimate the odometry independently. To cater for label insufficiency, an efficient transfer technique maps real image features to the virtual domain to use virtual pose ground truth. The training phases of EndoSERV include an offline pretraining to extract texture-agnostic features, and an online phase that adapts to real-world conditions. Extensive experiments based on both public and clinical datasets have been performed to demonstrate the effectiveness of the method even without any real pose labels.

We study mathematical discovery through the lens of neurosymbolic reasoning, where an AI agent powered by a large language model (LLM), coupled with symbolic computation tools, and human strategic direction, jointly produced a new result in combinatorial design theory. The main result of this human-AI collaboration is a tight lower bound on the imbalance of Latin squares for the notoriously difficult case $n \equiv 1 \pmod{3}$. We reconstruct the discovery process from detailed interaction logs spanning multiple sessions over several days and identify the distinct cognitive contributions of each component. The AI agent proved effective at uncovering hidden structure and generating hypotheses. The symbolic component consists of computer algebra, constraint solvers, and simulated annealing, which provides rigorous verification and exhaustive enumeration. Human steering supplied the critical research pivot that transformed a dead end into a productive inquiry. Our analysis reveals that multi-model deliberation among frontier LLMs proved reliable for criticism and error detection but unreliable for constructive claims. The resulting human-AI mathematical contribution, a tight lower bound of $4n(n{-}1)/9$, is achieved via a novel class of near-perfect permutations. The bound was formally verified in Lean 4. Our experiments show that neurosymbolic systems can indeed produce genuine discoveries in pure mathematics.

Large language models (LLMs) show significant potential for clinical decision support (CDS), yet their black-box nature -- characterized by untraceable reasoning and probabilistic hallucinations -- poses severe challenges in acupuncture, a field demanding rigorous interpretability and safety. To address this, we propose CORE-Acu, a neuro-symbolic framework for acupuncture clinical decision support that integrates Structured Chain-of-Thought (S-CoT) with knowledge graph (KG) safety verification. First, we construct the first acupuncture Structured Reasoning Trace dataset and a schema-constrained fine-tuning framework. By enforcing an explicit causal chain from pattern identification to treatment principles, treatment plans, and acupoint selection, we transform implicit Traditional Chinese Medicine (TCM) reasoning into interpretable generation constraints, mitigating the opacity of LLM-based CDS. Furthermore, we construct a TCM safety knowledge graph and establish a ``Generate--Verify--Revise'' closed-loop inference system based on a Symbolic Veto Mechanism, employing deterministic rules to intercept hallucinations and enforce hard safety boundaries. Finally, we introduce the Lexicon-Matched Entity-Reweighted Loss (LMERL), which corrects terminology drift caused by the frequency--importance mismatch in general optimization by adaptively amplifying gradient contributions of high-risk entities during fine-tuning. Experiments on 1,000 held-out cases demonstrate CORE-Acu's superior entity fidelity and reasoning quality. Crucially, CORE-Acu achieved 0/1,000 observed safety violations (95\% CI: 0--0.37\%), whereas GPT-4o exhibited an 8.5\% violation rate under identical rules. These results establish CORE-Acu as a robust neuro-symbolic framework for acupuncture clinical decision support, guaranteeing both reasoning auditability and strict safety compliance.

Humans consistently outperform state-of-the-art AI models in action recognition, particularly in challenging real-world conditions involving low resolution, occlusion, and visual clutter. Understanding the sources of this performance gap is essential for developing more robust and human-aligned models. In this paper, we present a large-scale human-AI comparative study of egocentric action recognition using Minimal Identifiable Recognition Crops (MIRCs), defined as the smallest spatial or spatiotemporal regions sufficient for reliable human recognition. We used our previously introduced, Epic ReduAct, a systematically spatially reduced and temporally scrambled dataset derived from 36 EPIC KITCHENS videos, spanning multiple spatial reduction levels and temporal conditions. Recognition performance is evaluated using over 3,000 human participants and the Side4Video model. Our analysis combines quantitative metrics, Average Reduction Rate and Recognition Gap, with qualitative analyses of spatial (high-, mid-, and low-level visual features) and spatiotemporal factors, including a categorisation of actions into Low Temporal Actions (LTA) and High Temporal Actions (HTA). Results show that human performance exhibits sharp declines when transitioning from MIRCs to subMIRCs, reflecting a strong reliance on sparse, semantically critical cues such as hand-object interactions. In contrast, the model degrades more gradually and often relies on contextual and mid- to low-level features, sometimes even exhibiting increased confidence under spatial reduction. Temporally, humans remain robust to scrambling when key spatial cues are preserved, whereas the model often shows insensitivity to temporal disruption, revealing class-dependent temporal sensitivities.

Modern vision-language-model (VLM) based graphical user interface (GUI) agents are expected not only to execute actions accurately but also to respond to user instructions with low latency. While existing research on GUI-agent security mainly focuses on manipulating action correctness, the security risks related to response efficiency remain largely unexplored. In this paper, we introduce SlowBA, a novel backdoor attack that targets the responsiveness of VLM-based GUI agents. The key idea is to manipulate response latency by inducing excessively long reasoning chains under specific trigger patterns. To achieve this, we propose a two-stage reward-level backdoor injection (RBI) strategy that first aligns the long-response format and then learns trigger-aware activation through reinforcement learning. In addition, we design realistic pop-up windows as triggers that naturally appear in GUI environments, improving the stealthiness of the attack. Extensive experiments across multiple datasets and baselines demonstrate that SlowBA can significantly increase response length and latency while largely preserving task accuracy. The attack remains effective even with a small poisoning ratio and under several defense settings. These findings reveal a previously overlooked security vulnerability in GUI agents and highlight the need for defenses that consider both action correctness and response efficiency. Code can be found in https://github.com/tu-tuing/SlowBA.

Speech foundation models trained with self-supervised learning produce generic speech representations that support a wide range of speech processing tasks. When further adapted with supervised learning, these models can achieve strong performance on specific downstream tasks. Recent post-training approaches, such as SAMU-XSLR and SONAR, align speech representations with utterance-level semantic representations, enabling effective multimodal (speech-text) and multilingual applications. While speech foundation models typically learn contextual embeddings at the acoustic frame level, these methods learn representations at the utterance level. In this work, we extend this paradigm to arbitrary utterance-level attributes and propose a unified post-training framework that enables a single speech foundation model to generate multiple types of utterance-level representations. We demonstrate the effectiveness of this approach by jointly learning semantic and speaker representations and evaluating them on multilingual speech retrieval and speaker recognition tasks.

We study identifiability in continuous-time linear stationary stochastic differential equations with known causal structure. Unlike existing approaches, we relax the assumption of a known diffusion matrix, thereby respecting the model's intrinsic scale invariance. Rather than recovering drift coefficients themselves, we introduce edge-sign identifiability: for a given causal structure, we ask whether the sign of a given drift entry is uniquely determined across all observational covariance matrices induced by parametrizations compatible with that structure. Under a notion of faithfulness, we derive criteria for characterising identifiability, non-identifiability, and partial identifiability for general graphs. Applying our criteria to specific causal structures, both analogous to classical causal settings (e.g., instrumental variables) and novel cyclic settings, we determine their edge-sign identifiability and, in some cases, obtain explicit expressions for the sign of a target edge in terms of the observational covariance matrix.

Vision Transformers (ViTs) often degrade under distribution shifts because they rely on spurious correlations, such as background cues, rather than semantically meaningful features. Existing regularization methods, typically relying on simple foreground-background masks, which fail to capture the fine-grained semantic concepts that define an object (e.g., ``long beak'' and ``wings'' for a ``bird''). As a result, these methods provide limited robustness to distribution shifts. To address this limitation, we introduce a novel finetuning framework that steers model reasoning toward concept-level semantics. Our approach optimizes the model's internal relevance maps to align with spatially grounded concept masks. These masks are generated automatically, without manual annotation: class-relevant concepts are first proposed using an LLM-based, label-free method, and then segmented using a VLM. The finetuning objective aligns relevance with these concept regions while simultaneously suppressing focus on spurious background areas. Notably, this process requires only a minimal set of images and uses half of the dataset classes. Extensive experiments on five out-of-distribution benchmarks demonstrate that our method improves robustness across multiple ViT-based models. Furthermore, we show that the resulting relevance maps exhibit stronger alignment with semantic object parts, offering a scalable path toward more robust and interpretable vision models. Finally, we confirm that concept-guided masks provide more effective supervision for model robustness than conventional segmentation maps, supporting our central hypothesis.

Text-conditioned generative models for volumetric medical imaging provide semantic control but lack explicit anatomical guidance, often resulting in outputs that are spatially ambiguous or anatomically inconsistent. In contrast, structure-driven methods ensure strong anatomical consistency but typically assume access to ground-truth annotations, which are unavailable when the target image is to be synthesized. We propose a retrieval-augmented approach for Text-to-CT generation that integrates semantic and anatomical information under a realistic inference setting. Given a radiology report, our method retrieves a semantically related clinical case using a 3D vision-language encoder and leverages its associated anatomical annotation as a structural proxy. This proxy is injected into a text-conditioned latent diffusion model via a ControlNet branch, providing coarse anatomical guidance while maintaining semantic flexibility. Experiments on the CT-RATE dataset show that retrieval-augmented generation improves image fidelity and clinical consistency compared to text-only baselines, while additionally enabling explicit spatial controllability, a capability inherently absent in such approaches. Further analysis highlights the importance of retrieval quality, with semantically aligned proxies yielding consistent gains across all evaluation axes. This work introduces a principled and scalable mechanism to bridge semantic conditioning and anatomical plausibility in volumetric medical image synthesis. Code will be released.

This work addresses the accurate and efficient simulation of physical phenomena governed by parametric Partial Differential Equations (PDEs) characterized by varying boundary conditions, where parametric instances modify not only the physics of the problem but also the imposition of boundary constraints on the computational domain. In such scenarios, classical Galerkin projection-based reduced order techniques encounter a fundamental bottleneck. Parametric boundaries typically necessitate a re-formulation of the discrete problem for each new configuration, and often, these approaches are unsuitable for real-time applications. To overcome these limitations, we propose a novel methodology based on Graph-Instructed Neural Networks (GINNs). The GINN framework effectively learns the mapping between the parametric description of the computational domain and the corresponding PDE solution. Our results demonstrate that the proposed GINN-based models, can efficiently represent highly complex parametric PDEs, serving as a robust and scalable asset for several applied-oriented settings when compared with fully connected architectures.

Multimodal Mathematical Reasoning (MMR) has recently attracted increasing attention for its capability to solve mathematical problems that involve both textual and visual modalities. However, current models still face significant challenges in real-world visual math tasks. They often misinterpret diagrams, fail to align mathematical symbols with visual evidence, and produce inconsistent reasoning steps. Moreover, existing evaluations mainly focus on checking final answers rather than verifying the correctness or executability of each intermediate step. To address these limitations, a growing body of recent research addresses these issues by integrating structured perception, explicit alignment, and verifiable reasoning within unified frameworks. To establish a clear roadmap for understanding and comparing different MMR approaches, we systematically study them around four fundamental questions: (1) What to extract from multimodal inputs, (2) How to represent and align textual and visual information, (3) How to perform the reasoning, and (4) How to evaluate the correctness of the overall reasoning process. Finally, we discuss open challenges and offer perspectives on promising directions for future research.

We study the implicit bias of Sharpness-Aware Minimization (SAM) when training $L$-layer linear diagonal networks on linearly separable binary classification. For linear models ($L=1$), both $\ell_\infty$- and $\ell_2$-SAM recover the $\ell_2$ max-margin classifier, matching gradient descent (GD). However, for depth $L = 2$, the behavior changes drastically -- even on a single-example dataset. For $\ell_\infty$-SAM, the limit direction depends critically on initialization and can converge to $\mathbf{0}$ or to any standard basis vector, in stark contrast to GD, whose limit aligns with the basis vector of the dominant data coordinate. For $\ell_2$-SAM, we show that although its limit direction matches the $\ell_1$ max-margin solution as in the case of GD, its finite-time dynamics exhibit a phenomenon we call "sequential feature amplification", in which the predictor initially relies on minor coordinates and gradually shifts to larger ones as training proceeds or initialization increases. Our theoretical analysis attributes this phenomenon to $\ell_2$-SAM's gradient normalization factor applied in its perturbation, which amplifies minor coordinates early and allows major ones to dominate later, giving a concrete example where infinite-time implicit-bias analyses are insufficient. Synthetic and real-data experiments corroborate our findings.

Aircraft engine blade maintenance relies on inspection records shared across manufacturers, airlines, maintenance organizations, and regulators. Yet current systems are fragmented, difficult to audit, and vulnerable to tampering. This paper presents BladeChain, a blockchain-based system providing immutable traceability for blade inspections throughout the component life cycle. BladeChain is the first system to integrate multi-stakeholder endorsement, automated inspection scheduling, AI model provenance, and cryptographic evidence binding, delivering auditable maintenance traceability for aerospace deployments. Built on a four-stakeholder Hyperledger Fabric network (OEM, Airline, MRO, Regulator), BladeChain captures every life-cycle event in a tamper-evident ledger. A chaincode-enforced state machine governs blade status transitions and automatically triggers inspections when configurable flight hour, cycle, or calendar thresholds are exceeded, eliminating manual scheduling errors. Inspection artifacts are stored off-chain in IPFS and linked to on-chain records via SHA-256 hashes, with each inspection record capturing the AI model name and version used for defect detection. This enables regulators to audit both what defects were found and how they were found. The detection module is pluggable, allowing organizations to adopt or upgrade inspection models without modifying the ledger or workflows. We built a prototype and evaluated it on workloads of up to 100 blades, demonstrating 100% life cycle completion with consistent throughput of 26 operations per minute. A centralized SQL baseline quantifies the consensus overhead and highlights the security trade-off. Security validation confirms tamper detection within 17~ms through hash verification.

We analyze the Bayesian regret of the Gaussian process posterior sampling reinforcement learning (GP-PSRL) algorithm. Posterior sampling is an effective heuristic for decision-making under uncertainty that has been used to develop successful algorithms for a variety of continuous control problems. However, theoretical work on GP-PSRL is limited. All known regret bounds either fail to achieve a tight dependence on a kernel-dependent quantity called the maximum information gain or fail to properly account for the fact that the set of possible system states is unbounded. Through a recursive application of the Borell-Tsirelson-Ibragimov-Sudakov inequality, we show that, with high probability, the states actually visited by the algorithm are contained within a ball of near-constant radius. To obtain tight dependence on the maximum information gain, we use the chaining method to control the regret suffered by GP-PSRL. Our main result is a Bayesian regret bound of the order $\widetilde{\mathcal{O}}(H^{3/2}\sqrt{γ_{T/H} T})$, where $H$ is the horizon, $T$ is the number of time steps and $γ_{T/H}$ is the maximum information gain. With this result, we resolve the limitations with prior theoretical work on PSRL, and provide the theoretical foundation and tools for analyzing PSRL in complex settings.

Legal argument mining aims to identify and classify the functional components of judicial reasoning, such as facts, issues, rules, analysis, and conclusions. Progress in this area is limited by the lack of large-scale, high-quality annotated datasets for U.S. caselaw, particularly at the state level. This paper introduces LAMUS, a sentence-level legal argument mining corpus constructed from U.S. Supreme Court decisions and Texas criminal appellate opinions. The dataset is created using a data-centric pipeline that combines large-scale case collection, LLM-based automatic annotation, and targeted human-in-the-loop quality refinement. We formulate legal argument mining as a six-class sentence classification task and evaluate multiple general-purpose and legal-domain language models under zero-shot, few-shot, and chain-of-thought prompting strategies, with LegalBERT as a supervised baseline. Results show that chain-of-thought prompting substantially improves LLM performance, while domain-specific models exhibit more stable zero-shot behavior. LLM-assisted verification corrects nearly 20% of annotation errors, improving label consistency. Human verification achieves Cohen's Kappa of 0.85, confirming annotation quality. LAMUS provides a scalable resource and empirical insights for future legal NLP research. All code and datasets can be accessed for reproducibility on GitHub at: https://github.com/LavanyaPobbathi/LAMUS/tree/main

Real-world optimization problems are often constrained by complex physical laws that limit computational scalability. These constraints are inherently tied to complex regions, and thus learning models that incorporate physical and geometric knowledge, i.e., physics-informed machine learning (PIML), offer a promising pathway for efficient solution. Here, we introduce PolyFormer, which opens a new direction for PIML in prescriptive optimization tasks, where physical and geometric knowledge is not merely used to regularize learning models, but to simplify the problems themselves. PolyFormer captures geometric structures behind constraints and transforms them into efficient polytopic reformulations, thereby decoupling problem complexity from solution difficulty and enabling off-the-shelf optimization solvers to efficiently produce feasible solutions with acceptable optimality loss. Through evaluations across three important problems (large-scale resource aggregation, network-constrained optimization, and optimization under uncertainty), PolyFormer achieves computational speedups up to 6,400-fold and memory reductions up to 99.87%, while maintaining solution quality competitive with or superior to state-of-the-art methods. These results demonstrate that PolyFormer provides an efficient and reliable solution for scalable constrained optimization, expanding the scope of PIML to prescriptive tasks in scientific discovery and engineering applications.

Abstractive summarization aims to generate concise summaries by creating new sentences, allowing for flexible rephrasing. However, this approach can be vulnerable to inaccuracies, particularly `hallucinations' where the model introduces non-existent information. In this paper, we leverage the use of multimodal and multilingual sentence embeddings derived from pretrained models such as LaBSE, SONAR, and BGE-M3, and feed them into a modified BART-based French model. A Named Entity Injection mechanism that appends tokenized named entities to the decoder input is introduced, in order to improve the factual consistency of the generated summary. Our novel framework, SBARThez, is applicable to both text and speech inputs and supports cross-lingual summarization; it shows competitive performance relative to token-level baselines, especially for low-resource languages, while generating more concise and abstract summaries.

As AI-assisted grant proposals outpace manual review capacity in a kind of ``Malthusian trap'' for the research ecosystem, this paper investigates the capabilities and limitations of LLM-based grant reviewing for high-stakes evaluation. Using six EPSRC proposals, we develop a perturbation-based framework probing LLM sensitivity across six quality axes: funding, timeline, competency, alignment, clarity, and impact. We compare three review architectures: single-pass review, section-by-section analysis, and a 'Council of Personas' ensemble emulating expert panels. The section-level approach significantly outperforms alternatives in both detection rate and scoring reliability, while the computationally expensive council method performs no better than baseline. Detection varies substantially by perturbation type, with alignment issues readily identified but clarity flaws largely missed by all systems. Human evaluation shows LLM feedback is largely valid but skewed toward compliance checking over holistic assessment. We conclude that current LLMs may provide supplementary value within EPSRC review but exhibit high variability and misaligned review priorities. We release our code and any non-protected data.

Accurate and interpretable mortality risk prediction in intensive care units (ICUs) remains a critical challenge due to the irregular temporal structure of electronic health records (EHRs), the complexity of longitudinal disease trajectories, and the lack of clinically grounded explanations in many data-driven models. To address these challenges, we propose \textit{TA-RNN-Medical-Hybrid}, a time-aware and knowledge-enriched deep learning framework that jointly models longitudinal clinical sequences and irregular temporal dynamics through explicit continuous-time encoding, along with standardized medical concept representations. The proposed framework extends time-aware recurrent modeling by integrating explicit continuous-time embeddings that operate independently of visit indexing, SNOMED-based disease representations, and a hierarchical dual-level attention mechanism that captures both visit-level temporal importance and feature/concept-level clinical relevance. This design enables accurate mortality risk estimation while providing transparent and clinically meaningful explanations aligned with established medical knowledge. We evaluate the proposed approach on the MIMIC-III critical care dataset and compare it against strong time-aware and sequential baselines. Experimental results demonstrate that TA-RNN-Medical-Hybrid consistently improves predictive performance in terms of AUC, accuracy, and recall-oriented F$_2$-score. Moreover, qualitative analysis shows that the model effectively decomposes mortality risk across time and clinical concepts, yielding interpretable insights into disease severity, chronicity, and temporal progression. Overall, the proposed framework bridges the gap between predictive accuracy and clinical interpretability, offering a scalable and transparent solution for high-stakes ICU decision support systems.

With the widespread adoption of Large Language Models (LLMs), respecting indigenous cultures becomes essential for models' culturally safety and responsible global applications. Existing studies separately consider cultural safety and cultural knowledge and neglect that the former should be grounded by the latter. This severely prevents LLMs from yielding culture-specific respectful responses. Consequently, adaptive cultural safety remains a formidable task. In this work, we propose to jointly model cultural safety and knowledge. First and foremost, cultural-safety and knowledge-paired data serve as the key prerequisite to conduct this research. However, the cultural diversity across regions and the subtlety of cultural differences pose significant challenges to the creation of such paired evaluation data. To address this issue, we propose a novel framework that integrates authoritative cultural knowledge descriptions curation, LLM-automated query generation, and heavy manual verification. Accordingly, we obtain a dataset named AdaCultureSafe containing 4.8K manually decomposed fine-grained cultural descriptions and the corresponding 48K manually verified safety- and knowledge-oriented queries. Upon the constructed dataset, we evaluate three families of popular LLMs on their cultural safety and knowledge proficiency, via which we make a critical discovery: no significant correlation exists between their cultural safety and knowledge proficiency. We then delve into the utility-related neuron activations within LLMs to investigate the potential cause of the absence of correlation, which can be attributed to the difference of the objectives of pre-training and post-alignment. We finally present a knowledge-grounded method, which significantly enhances cultural safety by enforcing the integration of knowledge into the LLM response generation process.

How much do large language models actually hallucinate when answering questions grounded in provided documents? Despite the critical importance of this question for enterprise AI deployments, reliable measurement has been hampered by benchmarks that rely on static datasets vulnerable to contamination, LLM-based judges with documented biases, or evaluation scales too small for statistical confidence. We address this gap using RIKER, a ground-truth-first evaluation methodology that enables deterministic scoring without human annotation. Across 35 open-weight models, three context lengths (32K, 128K, and 200K tokens), four temperature settings, and three hardware platforms (NVIDIA H200, AMD MI300X, and Intel Gaudi 3), we conducted over 172 billion tokens of evaluation - an order of magnitude beyond prior work. Our findings reveal that: (1) even the best-performing models fabricate answers at a non-trivial rate - 1.19% at best at 32K, with top-tier models at 5 - 7% - and fabrication rises steeply with context length, nearly tripling at 128K and exceeding 10% for all models at 200K; (2) model selection dominates all other factors, with overall accuracy spanning a 72-percentage-point range and model family predicting fabrication resistance better than model size; (3) temperature effects are nuanced - T=0.0 yields the best overall accuracy in roughly 60% of cases, but higher temperatures reduce fabrication for the majority of models and dramatically reduce coherence loss (infinite generation loops), which can reach 48x higher rates at T=0.0 versus T=1.0; (4) grounding ability and fabrication resistance are distinct capabilities - models that excel at finding facts may still fabricate facts that do not exist; and (5) results are consistent across hardware platforms, confirming that deployment decisions need not be hardware-dependent.

Graph Neural Networks (GNNs) have demonstrated remarkable success across diverse tasks. However, their generalization capability is often hindered by spurious correlations between node features and labels in the graph. Our analysis reveals that GNNs tend to exploit imperceptible statistical correlations in training data, even when such correlations are unreliable for prediction. To address this challenge, we propose the Spurious Correlation Learning Graph Neural Network (SCL-GNN), a novel framework designed to enhance generalization on both Independent and Identically Distributed (IID) and Out-of-Distribution (OOD) graphs. SCL-GNN incorporates a principled spurious correlation learning mechanism, leveraging the Hilbert-Schmidt Independence Criterion (HSIC) to quantify correlations between node representations and class scores. This enables the model to identify and mitigate irrelevant but influential spurious correlations effectively. Additionally, we introduce an efficient bi-level optimization strategy to jointly optimize modules and GNN parameters, preventing overfitting. Extensive experiments on real-world and synthetic datasets demonstrate that SCL-GNN consistently outperforms state-of-the-art baselines under various distribution shifts, highlighting its robustness and generalization capabilities.

In-context imitation learning allows robots to acquire skills from demonstrations, yet one-shot trajectory generation remains fragile under environmental variation. We propose SAIL, a framework that reframes robot imitation as an iterative refinement problem capable of scaling with test-time compute. SAIL utilizes Monte Carlo Tree Search, where each node is a complete trajectory and edges correspond to trajectory refinements. The process is guided by three core components: an automated archive of successful trajectories for contextually relevant retrieval, a vision language model-based scoring mechanism for trajectory evaluation, and a step-level feedback that provides trajectory-aligned scores for iterative refinement. Experiments across six diverse manipulation tasks in simulation and real-world validation clearly demonstrate that increasing test-time compute consistently improves success rates, achieving up to 95% on complex tasks. Our results suggest that trajectory-level test-time scaling is a robust path toward more generalizable robotic agents.

Bias in financial language models constitutes a major obstacle to their adoption in real-world applications. Detecting such bias is challenging, as it requires identifying inputs whose predictions change when varying properties unrelated to the decision, such as demographic attributes. Existing approaches typically rely on exhaustive mutation and pairwise prediction analysis over large corpora, which is effective but computationally expensive-particularly for large language models and can become impractical in continuous retraining and releasing processes. Aiming at reducing this cost, we conduct a large-scale study of bias in five financial language models, examining similarities in their bias tendencies across protected attributes and exploring cross-model-guided bias detection to identify bias-revealing inputs earlier. Our study uses approximately 17k real financial news sentences, mutated to construct over 125k original-mutant pairs. Results show that all models exhibit bias under both atomic (0.58\%-6.05\%) and intersectional (0.75\%-5.97\%) settings. Moreover, we observe consistent patterns in bias-revealing inputs across models, enabling substantial reuse and cost reduction in bias detection. For example, up to 73\% of FinMA's biased behaviours can be uncovered using only 20\% of the input pairs when guided by properties derived from DistilRoBERTa outputs.

Airborne Magnetic Anomaly Navigation (MagNav) provides a jamming-resistant and robust alternative to satellite navigation but requires the real-time compensation of the aircraft platform's large and dynamic magnetic interference. State-of-the-art solutions often rely on extensive offline calibration flights or pre-training, creating a logistical barrier to operational deployment. We present a fully adaptive MagNav architecture featuring a "cold-start" capability that identifies and compensates for the aircraft's magnetic signature entirely in-flight. The proposed method utilizes an extended Kalman filter with an augmented state vector that simultaneously estimates the aircraft's kinematic states as well as the coefficients of the physics-based Tolles-Lawson calibration model and the parameters of a Neural Network to model aircraft interferences. The Kalman filter update is mathematically equivalent to an online Natural Gradient descent, integrating superior convergence and data efficiency of state-of-the-art second-order optimization directly into the navigation filter. To enhance operational robustness, the neural network is constrained to a residual learning role, modeling only the nonlinearities uncorrected by the explainable physics-based calibration baseline. Validated on the MagNav Challenge dataset, our framework effectively bounds inertial drift using a magnetometer-only feature set. The results demonstrate navigation accuracy comparable to state-of-the-art models trained offline, without requiring prior calibration flights or dedicated maneuvers.

The integration of Large Language Models (LLMs) into the financial domain is driving a paradigm shift from passive information retrieval to dynamic, agentic interaction. While general-purpose tool learning has witnessed a surge in benchmarks, the financial sector, characterized by high stakes, strict compliance, and rapid data volatility, remains critically underserved. Existing financial evaluations predominantly focus on static textual analysis or document-based QA, ignoring the complex reality of tool execution. Conversely, general tool benchmarks lack the domain-specific rigor required for finance, often relying on toy environments or a negligible number of financial APIs. To bridge this gap, we introduce FinToolBench, the first real-world, runnable benchmark dedicated to evaluating financial tool learning agents. Unlike prior works limited to a handful of mock tools, FinToolBench establishes a realistic ecosystem coupling 760 executable financial tools with 295 rigorous, tool-required queries. We propose a novel evaluation framework that goes beyond binary execution success, assessing agents on finance-critical dimensions: timeliness, intent type, and regulatory domain alignment. Furthermore, we present FATR, a finance-aware tool retrieval and reasoning baseline that enhances stability and compliance. By providing the first testbed for auditable, agentic financial execution, FinToolBench sets a new standard for trustworthy AI in finance. The tool manifest, execution environment, and evaluation code will be open-sourced to facilitate future research.

Machine learning models involving discrete latent variables require gradient estimators to facilitate backpropagation in a computationally efficient manner. The most recent addition to the Straight-Through family of estimators, ReinMax, can be viewed from a numerical ODE perspective as incorporating an approximation via Heun's method to reduce bias, but at the cost of high variance. In this work, we introduce the ReinMax-Rao and ReinMax-CV estimators which incorporate Rao-Blackwellisation and control variate techniques into ReinMax to reduce its variance. Our estimators demonstrate superior performance on training variational autoencoders with discrete latent spaces. Furthermore, we investigate the possibility of leveraging alternative numerical methods for constructing more accurate gradient approximations and present an alternative view of ReinMax from a simpler numerical integration perspective.

Word sense plausibility rating requires predicting the human-perceived plausibility of a given word sense on a 1--5 scale in the context of short narrative stories containing ambiguous homonyms. This paper systematically compares three approaches: (1) embedding-based methods pairing sentence embeddings with standard regressors, (2) transformer fine-tuning with parameter-efficient adaptation, and (3) large language model (LLM) prompting with structured reasoning and explicit decision rules. The best-performing system employs a structured prompting strategy that decomposes evaluation into narrative components (precontext, target sentence, ending) and applies explicit decision rules for rating calibration. The analysis reveals that structured prompting with decision rules substantially outperforms both fine-tuned models and embedding-based approaches, and that prompt design matters more than model scale for this task. The code is publicly available at https://github.com/tongwu17/SemEval-2026-Task5.

Tactile sensation is essential for contact-rich manipulation tasks. It provides direct feedback on object geometry, surface properties, and interaction forces, enhancing perception and enabling fine-grained control. An inherent limitation of tactile sensors is that readings are available only when an object is touched. This precludes their use during planning and the initial execution phase of a task. Predicting tactile information from visual information can bridge this gap. A common approach is to learn a direct mapping from camera images to the output of vision-based tactile sensors. However, the resulting model will depend strongly on the specific setup and on how well the camera can capture the area where an object is touched. In this work, we introduce FlowTouch, a novel model for view-invariant visuo-tactile prediction. Our key idea is to use an object's local 3D mesh to encode rich information for predicting tactile patterns while abstracting away from scene-dependent details. FlowTouch integrates scene reconstruction and Flow Matching-based models for image generation. Our results show that FlowTouch is able to bridge the sim-to-real gap and generalize to new sensor instances. We further show that the resulting tactile images can be used for downstream grasp stability prediction. Our code, datasets and videos are available at https://flowtouch.github.io/

Federated Learning (FL) suffers significant performance degradation in real-world deployments characterized by moderate to extreme statistical heterogeneity (non-IID client data). While global aggregation strategies promote broad generalization, they often fail to capture the diversity of local data distributions, leading to suboptimal personalization. We address this problem with FedPrism, a framework that uses two main strategies. First, it uses a Prism Decomposition method that builds each client's model from three parts: a global foundation, a shared group part for similar clients, and a private part for unique local data. This allows the system to group similar users together automatically and adapt if their data changes. Second, we include a Dual-Stream design that runs a general model alongside a local specialist. The system routes predictions between the general model and the local specialist based on the specialist's confidence. Through systematic experiments on non-IID data partitions, we demonstrate that FedPrism exceeds static aggregation and hard-clustering baselines, achieving significant accuracy gains under high heterogeneity. These results establish FedPrism as a robust and flexible solution for federated learning in heterogeneous environments, effectively balancing generalizable knowledge with adaptive personalization.

Scaling test-time computation enhances LLM reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose CoFiCot, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, CoFiCot effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.

Audiovisual speech recognition (AVSR) combines acoustic and visual cues to improve transcription robustness under challenging conditions but remains out of reach for most under-resourced languages due to the lack of labeled video corpora for training. We propose a zero-AV-resource AVSR framework that relies on synthetic visual streams generated by lip-syncing static facial images with real audio. We first evaluate synthetic visual augmentation on Spanish benchmarks, then apply it to Catalan, a language with no annotated audiovisual corpora. We synthesize over 700 hours of talking-head video and fine-tune a pre-trained AV-HuBERT model. On a manually annotated Catalan benchmark, our model achieves near state-of-the-art performance with much fewer parameters and training data, outperforms an identically trained audio-only baseline, and preserves multimodal advantages in noise. Scalable synthetic video thus offers a viable substitute for real recordings in zero-AV-resource AVSR.

Timely transition from intravenous (IV) to oral antibiotic therapy shortens hospital stays, reduces catheter-related infections, and lowers healthcare costs, yet one in five patients in England remain on IV antibiotics despite meeting switching criteria. Clinical decision support systems can improve switching rates, but approaches that learn from historical decisions reproduce the delays and inconsistencies of routine practice. We propose using neural processes to model vital sign trajectories probabilistically, predicting switch-readiness by comparing forecasts against clinical guidelines rather than learning from past actions, and ranking patients to prioritise clinical review. The design yields interpretable outputs, adapts to updated guidelines without retraining, and preserves clinical judgement. Validated on MIMIC-IV (US intensive care, 6,333 encounters) and UCLH (a large urban academic UK hospital group, 10,584 encounters), the system selects 2.2-3.2$\times$ more relevant patients than random. Our results demonstrate that forecasting patient physiology offers a principled foundation for decision support in antibiotic stewardship.

Transformer models are now a cornerstone in natural language processing. Yet, explaining their decisions remains a challenge. It was shown recently that the same model trained on the same data with a different randomness can lead to very different explanations. In this paper, we investigate how the (syntactic) context, the classes to be learned and the tasks influence this explanations' sensitivity to randomness. We show that they all have statistically significant impact: smallest for the (syntactic) context, medium for the classes and largest for the tasks.

Large language models are increasingly trained as heterogeneous systems spanning multiple domains, expert partitions, and agentic pipelines, yet prevalent proximal objectives operate at a single scale and lack a principled mechanism for coupling token-level, trajectory-level, and higher-level hierarchical stability control. To bridge this gap, we derive the Aggregational Policy Censoring Objective (APC-Obj), the first exact unconstrained reformulation of sample-based TV-TRPO, establishing that clipping-based surrogate design and trust-region optimization are dual formulations of the same problem. Building on this foundation, we develop Fiber Bundle Gating (FBG), an algebraic framework that organizes sampled RL data as a fiber bundle and decomposes ratio gating into a base-level gate on trajectory aggregates and a fiber-level gate on per-token residuals, with provable first-order agreement with the true RL objective near on-policy. From APC-Obj and FBG we derive Fibration Policy Optimization (or simply, FiberPO), a concrete objective whose Jacobian is block-diagonal over trajectories, reduces to identity at on-policy, and provides better update direction thus improving token efficiency. The compositional nature of the framework extends beyond the trajectory-token case: fibrations compose algebraically into a Fibration Gating Hierarchy (FGH) that scales the same gating mechanism to arbitrary hierarchical depth without new primitives, as demonstrated by FiberPO-Domain, a four-level instantiation with independent trust-region budgets at the domain, prompt group, trajectory, and token levels. Together, these results connect the trust-region theory, a compositional algebraic structure, and practical multi-scale stability control into a unified framework for LLM policy optimization.

Diabetic retinopathy (DR) and diabetic macular edema (DME) are leading causes of preventable blindness among working-age adults. Traditional approaches in the literature focus on standard color fundus photography (CFP) for the detection of these conditions. Nevertheless, recent ultra-widefield imaging (UWF) offers a significantly wider field of view in comparison to CFP. Motivated by this, the present study explores state-of-the-art deep learning (DL) methods and UWF imaging on three clinically relevant tasks: i) image quality assessment for UWF, ii) identification of referable diabetic retinopathy (RDR), and iii) identification of DME. Using the publicly available UWF4DR Challenge dataset, released as part of the MICCAI 2024 conference, we benchmark DL models in the spatial (RGB) and frequency domains, including popular convolutional neural networks (CNNs) as well as recent vision transformers (ViTs) and foundation models. In addition, we explore a final feature-level fusion to increase robustness. Finally, we also analyze the decisions of the DL models using Grad-CAM, increasing the explainability. Our proposal achieves consistently strong performance across all architectures, underscoring the competitiveness of emerging ViTs and foundation models and the promise of feature-level fusion and frequency-domain representations for UWF analysis.

With the rapid advancement of large language models (LLMs), the safety of LLMs has become a critical concern. Despite significant efforts in safety alignment, current LLMs remain vulnerable to jailbreaking attacks. However, the root causes of such vulnerabilities are still poorly understood, necessitating a rigorous investigation into jailbreak mechanisms across both academic and industrial communities. In this work, we focus on a continuation-triggered jailbreak phenomenon, whereby simply relocating a continuation-triggered instruction suffix can substantially increase jailbreak success rates. To uncover the intrinsic mechanisms of this phenomenon, we conduct a comprehensive mechanistic interpretability analysis at the level of attention heads. Through causal interventions and activation scaling, we show that this jailbreak behavior primarily arises from an inherent competition between the model's intrinsic continuation drive and the safety defenses acquired through alignment training. Furthermore, we perform a detailed behavioral analysis of the identified safety-critical attention heads, revealing notable differences in the functions and behaviors of safety heads across different model architectures. These findings provide a novel mechanistic perspective for understanding and interpreting jailbreak behaviors in LLMs, offering both theoretical insights and practical implications for improving model safety.

Paralinguistic speech tasks are often considered relatively language-agnostic, as they rely on extralinguistic acoustic cues rather than lexical content. However, prior studies report performance degradation under cross-lingual conditions, indicating non-negligible language dependence. Still, these studies typically focus on isolated language pairs or task-specific settings, limiting comparability and preventing a systematic assessment of task-level language dependence. We introduce the Cross-Lingual Transfer Matrix (CLTM), a systematic method to quantify cross-lingual interactions between pairs of languages within a given task. We apply the CLTM to two paralinguistic tasks, gender identification and speaker verification, using a multilingual HuBERT-based encoder, to analyze how donor-language data affects target-language performance during fine-tuning. Our results reveal distinct transfer patterns across tasks and languages, reflecting systematic, language-dependent effects.

Speech emotion recognition plays an important role in various applications. However, most existing approaches predict a single emotion label, oversimplifying the inherently ambiguous nature of human emotional expression. Recent large audio-language models show promise in generating richer outputs, but their reasoning ability for ambiguous emotional understanding remains limited. In this work, we reformulate ambiguous emotion recognition as a distributional reasoning problem and present the first systematic study of ambiguity-aware reasoning in LALMs. Our framework comprises two complementary components: an ambiguity-aware objective that aligns predictions with human perceptual distributions, and a structured ambiguity-aware chain-of-thought supervision that guides reasoning over emotional cues. Experiments on IEMOCAP and CREMA-D demonstrate consistent improvements across SFT, DPO, and GRPO training strategies.

Enterprise adoption of cloud-based AI agents faces a fundamental privacy dilemma: leveraging powerful cloud models requires sharing sensitive data, while local processing limits capability. Current agent frameworks like MCP and A2A assume complete data sharing, making them unsuitable for enterprise environments with confidential information. We present SplitAgent, a novel distributed architecture that enables privacy-preserving collaboration between enterprise-side privacy agents and cloud-side reasoning agents. Our key innovation is context-aware dynamic sanitization that adapts privacy protection based on task semantics -- contract review requires different sanitization than code review or financial analysis. SplitAgent extends existing agent protocols with differential privacy guarantees, zero-knowledge tool verification, and privacy budget management. Through comprehensive experiments on enterprise scenarios, we demonstrate that SplitAgent achieves 83.8\% task accuracy while maintaining 90.1\% privacy protection, significantly outperforming static approaches (73.2\% accuracy, 79.7\% privacy). Context-aware sanitization improves task utility by 24.1\% over static methods while reducing privacy leakage by 67\%. Our architecture provides a practical path for enterprise AI adoption without compromising sensitive data.

In this paper, we explore how our recently developed Wiener Chaos Expansion (WCE)-based neural operator (NO) can be applied to singular stochastic partial differential equations, e.g., the dynamic $\boldsymbolΦ^4_2$ model simulated in the recent works. Unlike the previous WCE-NO which solves SPDEs by simply inserting Wick-Hermite features into the backbone NO model, we leverage feature-wise linear modulation (FiLM) to appropriately capture the dependency between the solution of singular SPDE and its smooth remainder. The resulting WCE-FiLM-NO shows excellent performance on $\boldsymbolΦ^4_2$, as measured by relative $L_2$ loss, out-of-distribution $L_2$ loss, and autocorrelation score; all without the help of renormalisation factor. In addition, we also show the potential of simulating $\boldsymbolΦ^4_3$ data, which is more aligned with real scientific practice in statistical quantum field theory. To the best of our knowledge, this is among the first works to develop an efficient data-driven surrogate for the dynamical $\boldsymbolΦ^4_3$ model.

Speech-to-speech models handle turn-taking naturally but offer limited support for tool-calling or complex reasoning, while production ASR-LLM-TTS voice pipelines offer these capabilities but rely on silence timeouts, which lead to unnatural turn-taking. We present DualTurn, which narrows this gap through generative pretraining on dual-channel conversational audio. The model generates both speakers' future audio autoregressively, implicitly learning conversational dynamics without any labels, and is then fine-tuned to predict interpretable turn-taking signals that map directly to agent actions. DualTurn monitors both channels continuously, anticipating turn boundaries and producing five agent actions. On standard benchmarks, DualTurn (0.5B) outperforms both VAP on agent action prediction (wF1 0.633 vs. 0.389) and a 3.1B audio-text model on word-level turn prediction (AUC 0.930 vs. 0.880), while anticipating turn boundaries earlier with fewer interruptions.

In asynchronous federated learning (FL), client devices send updates to a central server at varying times based on their computational speed, often using stale versions of the global model. This staleness can degrade the convergence and accuracy of the global model. Previous work, such as AsyncFedED, proposed an adaptive aggregation method using Euclidean distance to measure staleness. In this paper, we extend this approach by exploring alternative distance metrics to more accurately capture the effect of gradient staleness. We integrate these metrics into the aggregation process and evaluate their impact on convergence speed, model performance, and training stability under heterogeneous clients and non-IID data settings. Our results demonstrate that certain metrics lead to more robust and efficient asynchronous FL training, offering a stronger foundation for practical deployment.

Reliable unmanned aerial vehicle (UAV) detection is critical for autonomous airspace monitoring but remains challenging when integrating sensor streams that differ substantially in resolution, perspective, and field of view. Conventional fusion methods-such as wavelet-, Laplacian-, and decision-level approaches-often fail to preserve spatial correspondence across modalities and suffer from annotation of inconsistencies, limiting their robustness in real-world settings. This study introduces two fusion strategies, Registration-aware Guided Image Fusion (RGIF) and Reliability-Gated Modality-Attention Fusion (RGMAF), designed to overcome these limitations. RGIF employs Enhanced Correlation Coefficient (ECC)-based affine registration combined with guided filtering to maintain thermal saliency while enhancing structural detail. RGMAF integrates affine and optical-flow registration with a reliability-weighted attention mechanism that adaptively balances thermal contrast and visual sharpness. Experiments were conducted on the Multi-Sensor and Multi-View Fixed-Wing (MMFW)-UAV dataset comprising 147,417 annotated air-to-air frames collected from infrared, wide-angle, and zoom sensors. Among single-modality detectors, YOLOv10x demonstrated the most stable cross-domain performance and was selected as the detection backbone for evaluating fused imagery. RGIF improved the visual baseline by 2.13% mAP@50 (achieving 97.65%), while RGMAF attained the highest recall of 98.64%. These findings show that registration-aware and reliability-adaptive fusion provides a robust framework for integrating heterogeneous modalities, substantially enhancing UAV detection performance in multimodal environments.

Removing personally identifiable information (PII) from texts is necessary to comply with various data protection regulations and to enable data sharing without compromising privacy. However, recent works show that documents sanitized by PII removal techniques are vulnerable to reconstruction attacks. Yet, we suspect that the reported success of these attacks is largely overestimated. We critically analyze the evaluation of existing attacks and find that data leakage and data contamination are not properly mitigated, leaving the question whether or not PII removal techniques truly protect privacy in real-world scenarios unaddressed. We investigate possible data sources and attack setups that avoid data leakage and conclude that only truly private data can allow us to objectively evaluate vulnerabilities in PII removal techniques. However, access to private data is heavily restricted - and for good reasons - which also means that the public research community cannot address this problem in a transparent, reproducible, and trustworthy manner.

Prior-Data Fitted Networks (PFNs), such as TabPFN and TabICL, have revolutionized tabular deep learning by leveraging in-context learning for tabular data. These models are meant as foundation models for classification and regression settings and promise to greatly simplify deployment in practical settings because their performance is unprecedented (in terms of mean squared error or $R^2$, when measured on common benchmarks like TabArena or TALENT). However, we see an important weakness of current benchmarks for the regression setting: the current benchmarks focus on evaluating win rates and performance using metrics like (root) mean squared error or $R^2$. Therefore, these leaderboards (implicitly and explicitly) push researchers to optimize for machine learning pipelines which elicit a good mean value estimate. The main problem is that this approach only evaluates a point estimate (namely the mean estimator which is the Bayes estimator associated with the mean squared error loss). In this article we discuss the application of proper scoring rules for evaluating the goodness of probabilistic forecasts in distributional regression. We also propose to enhance common machine learning benchmarks with metrics for probabilistic regression. To improve the status quo and make the machine learning community aware of scoring rules for probabilistic regression, we advocate to use the continuous ranked probability score (CRPS) in benchmarks for probabilistic regression. However, we also illustrate that the choice of the scoring rule changes the inductive bias of the trained model. We, therefore, advocate for finetuning or promptable tabular foundation models.

Contrastive learning has become a fundamental approach in both uni-modal and multi-modal frameworks. This learning paradigm pulls positive pairs of samples closer while pushing negatives apart. In the uni-modal setting (e.g., image-based learning), previous research has shown that the strength of these forces can be controlled through the temperature parameter. In this work, we propose Multi-Modal Temperature and Margin Schedules (MM-TS), extending the concept of uni-modal temperature scheduling to multi-modal contrastive learning. Our method dynamically adjusts the temperature in the contrastive loss during training, modulating the attraction and repulsion forces in the multi-modal setting. Additionally, recognizing that standard multi-modal datasets often follow imbalanced, long-tail distributions, we adapt the temperature based on the local distribution of each training sample. Specifically, samples from dense clusters are assigned a higher temperature to better preserve their semantic structure. Furthermore, we demonstrate that temperature scheduling can be effectively integrated within a max-margin framework, thereby unifying the two predominant approaches in multi-modal contrastive learning: InfoNCE loss and max-margin objective. We evaluate our approach on four widely used image- and video-language datasets, Flickr30K, MSCOCO, EPIC-KITCHENS-100, and YouCook2, and show that our dynamic temperature and margin schedules improve performance and lead to new state-of-the-art results in the field.

Motivation: The rapid growth of biological data has intensified the need for transparent, reproducible, and well-documented computational workflows. The ability to clearly connect the steps of a workflow in the code with their description in a paper would improve workflow understanding, support reproducibility, and facilitate reuse. This task requires the linking of Bioinformatics tools in workflow code with their mentions in a published workflow description. Results: We present CoPaLink, an automated approach that integrates three components: Named Entity Recognition (NER) for identifying tool mentions in scientific text, NER for tool mentions in workflow code, and entity linking grounded on Bioinformatics knowledge bases. We propose approaches for all three steps achieving a high individual F1-measure (84 - 89) and a joint accuracy of 66 when evaluated on Nextflow workflows using Bioconda and Bioweb Knowledge bases. CoPaLink leverages corpora of scientific articles and workflow executable code with curated tool annotations to bridge the gap between narrative descriptions and workflow implementations. Availability: The code is available at https://gitlab.liris.cnrs.fr/sharefair/copalink-experiments and https://gitlab.liris.cnrs.fr/sharefair/copalink. The corpora are also available at https://doi.org/10.5281/zenodo.18526700, https://doi.org/10.5281/zenodo.18526760 and https://doi.org/10.5281/zenodo.18543814.

Agile organizations increasingly rely on automated regression testing to sustain rapid, high-quality software delivery. However, as systems grow and requirements evolve, a persistent bottleneck arises: test specifications are produced faster than they can be transformed into executable scripts, leading to mounting manual effort and delayed releases. In partnership with Hacon (a Siemens company), we present an agentic AI approach that generates system-level test scripts directly from validated specifications, aiming to accelerate automation without sacrificing human oversight. Our solution features a retrieval-augmented, multi-agent architecture integrated into Hacon's agile workflows. We evaluate this system through a mixed-method analysis of industrial artifacts and practitioner feedback. Results show that the AI teammate significantly increases test script throughput and reduces manual authoring effort, while underscoring the ongoing need for clear specifications and human review to ensure quality and maintainability. We conclude with practical lessons for scaling regression automation and fostering effective Human-AI collaboration in agile environments.

Service region design determines the geographic coverage of service networks, shaping long-term operational performance. Capital and operational constraints preclude simultaneous large-scale deployment, requiring expansion to proceed sequentially. The resulting challenge is to determine when and where to invest under demand uncertainty, balancing intertemporal trade-offs between early and delayed investment and accounting for network effects whereby each deployment reshapes future demand through inter-regional connectivity. This study addresses a sequential service region design (SSRD) problem incorporating two practical yet underexplored factors: a $k$-region constraint that limits the number of regions investable per period and a stochastic spillover effect linking investment decisions to demand evolution. The resulting problem requires sequencing regional portfolios under uncertainty, leading to a combinatorial explosion in feasible investment sequences. To address this challenge, we propose a solution framework that integrates real options analysis (ROA) with a Transformer-based Proximal Policy Optimization (TPPO) algorithm. ROA evaluates the intertemporal option value of investment sequences, while TPPO learns sequential policies that directly generate high option-value sequences without exhaustive enumeration. Numerical experiments on realistic multi-region settings demonstrate that TPPO converges faster than benchmark DRL methods and consistently identifies sequences with superior option value. Case studies and sensitivity analyses further confirm robustness and provide insights on investment concurrency, regional prioritization, and the increasing benefits of adaptive expansion via our approach under stronger spillovers and dynamic market conditions.

Post-training quantization (PTQ) has emerged as a prevailing technique for deploying large language models (LLMs) efficiently in terms of both memory and computation, across edge devices and server platforms. Existing PTQ methods primarily aim to reduce precision in weights and activations by mitigating quantization errors caused by channel-wise outlier activations (e.g., pre-quantization scaling, online transformations, or low-rank error reconstruction). Among these approaches, error reconstruction with low-rank adaptation (LoRA) has proven particularly effective, as it introduces a lightweight auxiliary computation path without requiring heavy optimization or additional online layers. However, prior studies reveal severe accuracy degradation under W4A4 settings, and conventional low-rank adaptations rely on two sequential factors, necessitating intermediate quantization during inference and thereby limiting low-precision efficiency. In this work, we propose SERQ, a saliency-aware error reconstruction method for low-bit LLM inference that employs a single low-rank compensation matrix. SERQ preserves efficient 4-bit matrix multiplication in linear layers by jointly mitigating quantization errors arising from both activation and weight saliency through three stages: (1) static activation flattening, (2) saliency-aware error reconstruction, and (3) offline weight permutation. The method incurs additional computation only for low-rank error reconstruction via a single decomposition, while all other operations are performed offline, thereby keeping latency overhead minimal. Empirically, SERQ outperforms prior error reconstruction methods under both W4A8 and W4A4 settings, and achieves higher accuracy than state-of-the-art rotation-based W4A4 approaches, while substantially reducing calibration complexity.

Large language models often underperform in many European languages due to the dominance of English and a few high-resource languages in training data. This paper presents TildeOpen LLM, a 30-billion-parameter open-weight foundational model trained for 34 European languages to promote linguistic equity and improve performance for low-resource languages. To address the data imbalance, we combine dataset upsampling with a curriculum-based training schedule that alternates between uniform and natural language distributions. The resulting model performs favorably compared to other multilingual LLMs despite being trained with significantly fewer computing resources. Evaluation across multiple multilingual benchmarks shows that TildeOpen surpasses existing open-weight models in text generation and comprehension, particularly for Baltic, Finno-Ugric, and Slavic languages. Human evaluations confirm an up to tenfold reduction in linguistic errors relative to leading baselines. The model and associated resources are fully open-weight and publicly available at huggingface.co/TildeAI/TildeOpen-30b. These outcomes demonstrate that careful data curation and balanced training strategies can substantially enhance multilingual model quality without increasing model size or training volume.

Large language models (LLMs) excel in open domains but struggle in specialized settings with limited data and evolving knowledge. Existing domain adaptation practices rely heavily on manual trial-and-error processes, incur significant hyperparameter complexity, and are highly sensitive to data and user preferences, all under the high cost of LLM training. Moreover, the interactions and transferability of hyperparameter choices across models/domains remain poorly understood, making adaptation gains uncertain even with substantial effort. To solve these challenges, we present AutoAdapt, a novel end-to-end automated framework for efficient and reliable LLM domain adaptation. AutoAdapt leverages curated knowledge bases from literature and open-source resources to reduce expert intervention. To narrow the search space, we design a novel multi-agent debating system in which proposal and critic agents iteratively interact to align user intent and incorporate data signals and best practices into the planning process. To optimize hyperparameters under tight budgets, we propose AutoRefine, a novel LLM-based surrogate that replaces costly black-box search. Across 10 tasks, AutoAdapt achieves a 25% average relative accuracy improvement over state-of-the-art Automated Machine Learning baselines with minimal overhead.

LiDAR-based 3D object detection plays a critical role for reliable and safe autonomous driving systems. However, existing detectors often produce overly confident predictions for objects not belonging to known categories, posing significant safety risks. This is caused by so-called out-of-distribution (OOD) objects, which were not part of the training data, resulting in incorrect predictions. To address this challenge, we propose ALOOD (Aligned LiDAR representations for Out-Of-Distribution Detection), a novel approach that incorporates language representations from a vision-language model (VLM). By aligning the object features from the object detector to the feature space of the VLM, we can treat the detection of OOD objects as a zero-shot classification task. We demonstrate competitive performance on the nuScenes OOD benchmark, establishing a novel approach to OOD object detection in LiDAR using language representations. The source code is available at https://github.com/uulm-mrm/mmood3d.

End-to-end full-duplex speech models feed user audio through an always-on LLM backbone, yet the speaker privacy implications of their hidden representations remain unexamined. Following the VoicePrivacy 2024 protocol with a lazy-informed attacker, we show that the hidden states of SALM-Duplex and Moshi leak substantial speaker identity across all transformer layers. Layer-wise and turn-wise analyses reveal that leakage persists across all layers, with SALM-Duplex showing stronger leakage in early layers while Moshi leaks uniformly, and that Linkability rises sharply within the first few turns. We propose two streaming anonymization setups using Stream-Voice-Anon: a waveform-level front-end (Anon-W2W) and a feature-domain replacement (Anon-W2F). Anon-W2F raises EER by over 3.5x relative to the discrete encoder baseline (11.2% to 41.0%), approaching the 50% random-chance ceiling, while Anon-W2W retains 78-93% of baseline sBERT across setups with sub-second response latency (FRL under 0.8 s).

We investigate whether performing reasoning in a continuous latent space leads to more robust multilingual capabilities. We compare Continuous Chain-of-Thought (using the CODI framework) against standard supervised fine-tuning across five typologically diverse languages: English, Chinese, German, French, and Urdu. Our experiments on GSM8k and CommonsenseQA demonstrate that continuous reasoning significantly outperforms explicit reasoning on low-resource languages, particularly in zero-shot settings where the target language was not seen during training. Additionally, this approach achieves extreme efficiency, compressing reasoning traces by approximately $29\times$ to $50\times$. These findings indicate that continuous latent representations naturally exhibit greater language invariance, offering a scalable solution for cross-lingual reasoning.

Quantization has become essential for the efficient deployment of speech processing systems. Although widely studied, most existing quantization methods were developed for vision and NLP architectures, while the specific challenges of audio signals remain largely overlooked. In particular, we show that audio activations can exhibit large calibration ranges, leading to significant information loss when standard calibration techniques are applied. To address this, we propose ESC, an Evolution Strategy-based Calibration method that formulates activation scaling as an optimization problem and solves it using a two-step local-global scheme driven by an evolution strategy. ESC enables unaltered performance under full INT8 quantization and is the first calibration method to achieve near-lossless performance for full INT4 quantization across multiple speech tasks. Integrating ESC with PTQ methods further reduces performance loss, achieving a 1% relative accuracy degradation on the AST model.

Industrial maintenance platforms contain rich but fragmented evidence, including free-text work orders, heterogeneous operational sensors or indicators, and structured failure knowledge. These sources are often analyzed in isolation, producing alerts or forecasts that do not support conditional decision-making: given this asset history and behavior, what is happening and what action is warranted? We present Condition Insight Agent, a deployed decision-support framework that integrates maintenance language, behavioral abstractions of operational data, and engineering failure semantics to produce evidence-grounded explanations and advisory actions. The system constrains reasoning through deterministic evidence construction and structured failure knowledge, and applies a rule-based verification loop to suppress unsupported conclusions. Case studies from production CMMS deployments show that this verification-first design operates reliably under heterogeneous and incomplete data while preserving human oversight. Our results demonstrate how constrained LLM-based reasoning can function as a governed decision-support layer for industrial maintenance.

Automated Drug Combination Extraction (DCE) from large-scale biomedical literature is crucial for advancing precision medicine and pharmacological research. However, existing relation extraction methods primarily focus on binary interactions and struggle to model variable-length n-ary drug combinations, where complex compatibility logic and distributed evidence need to be considered. To address these limitations, we propose RexDrug, an end-to-end reasoning-enhanced relation extraction framework for n-ary drug combination extraction based on large language models. RexDrug adopts a two-stage training strategy. First, a multi-agent collaborative mechanism is utilized to automatically generate high-quality expert-like reasoning traces for supervised fine-tuning. Second, reinforcement learning with a multi-dimensional reward function specifically tailored for DCE is applied to further refine reasoning quality and extraction accuracy. Extensive experiments on the DrugComb dataset show that RexDrug consistently outperforms state-of-the-art baselines for n-ary extraction. Additional evaluation on the DDI13 corpus confirms its generalizability to binary drugdrug interaction tasks. Human expert assessment and automatic reasoning metrics further indicates that RexDrug produces coherent medical reasoning while accurately identifying complex therapeutic regimens. These results establish RexDrug as a scalable and reliable solution for complex biomedical relation extraction from unstructured text. The source code and data are available at https://github.com/DUTIR-BioNLP/RexDrug

Advancements in data-driven machine learning have emerged as a pivotal element in supporting automotive software systems (ASSs) engineering across various levels of the V-development process. Duringsystemverificationandvalidation,theintegrationofanintelligent fault detection anddiagnosis (FDD) model with test recordings analysis process serves as a powerful tool for efficiency ensuring functional safety. However, the lack of interpretability of the black-box FDD models developed not only hinders understanding of the cause underlying the prediction, but also prevents the model from being adapted based on the prediction result. This, in turn, increases the computational cost required for developingacomplexFDDmodelandlimitsconfidenceinreal-timesafety-criticalapplications.To address this challenge, a novel explainable method for fault detection, identification, and localization is proposed in this article with the aim of providing a clear understanding of the logic behind the prediction outcome. To this end, a hybrid 1dCNN-GRU-based intelligent model was developed to analyze the recordings from the real-time validation process of ASSs. The employment of explainable AI techniques, i.e., IGs, DeepLIFT, Gradient SHAP, and DeepLIFT SHAP, was instrumental in enabling model adaptation and facilitating the root cause analysis (RCA). The proposed approach is applied to the real time dataset collected during a virtual test drive performed by the user on hardware in the loop system.

Recently, there has been increased interest in globally distributed training, which has the promise to both reduce training costs and democratize participation in building large-scale foundation models. However, existing models trained in a globally distributed manner are relatively small in scale and have only been trained with whitelisted participants. Therefore, they do not yet realize the full promise of democratized participation. In this report, we describe Covenant-72B, an LLM produced by the largest collaborative globally distributed pre-training run (in terms of both compute and model scale), which simultaneously allowed open, permissionless participation supported by a live blockchain protocol. We utilized a state-of-the-art communication-efficient optimizer, SparseLoCo, supporting dynamic participation with peers joining and leaving freely. Our model, pre-trained on approximately 1.1T tokens, performs competitively with fully centralized models pre-trained on similar or higher compute budgets, demonstrating that fully democratized, non-whitelisted participation is not only feasible, but can be achieved at unprecedented scale for a globally distributed pre-training run.

Hierarchical knowledge structures are ubiquitous across real-world domains and play a vital role in organizing information from coarse to fine semantic levels. While such structures have been widely used in taxonomy systems, biomedical ontologies, and retrieval-augmented generation, their potential remains underexplored in the context of Text-Rich Networks (TRNs), where each node contains rich textual content and edges encode semantic relationships. Existing methods for learning on TRNs often focus on flat semantic modeling, overlooking the inherent hierarchical semantics embedded in textual documents. To this end, we propose TIER (Hierarchical \textbf{T}axonomy-\textbf{I}nformed R\textbf{E}presentation Learning on Text-\textbf{R}ich Networks), which first constructs an implicit hierarchical taxonomy and then integrates it into the learned node representations. Specifically, TIER employs similarity-guided contrastive learning to build a clustering-friendly embedding space, upon which it performs hierarchical K-Means followed by LLM-powered clustering refinement to enable semantically coherent taxonomy construction. Leveraging the resulting taxonomy, TIER introduces a cophenetic correlation coefficient-based regularization loss to align the learned embeddings with the hierarchical structure. By learning representations that respect both fine-grained and coarse-grained semantics, TIER enables more interpretable and structured modeling of real-world TRNs. We demonstrate that our approach significantly outperforms existing methods on multiple datasets across diverse domains, highlighting the importance of hierarchical knowledge learning for TRNs.

The autocovariance least squares (ALS) method is a computationally efficient approach for estimating noise covariances in Kalman filters without requiring specific noise models. However, conventional ALS and its variants rely on the classic least mean squares (LMS) criterion, making them highly sensitive to measurement outliers and prone to severe performance degradation. To overcome this limitation, this paper proposes a novel outlier-robust ALS algorithm, termed ALS-IRLS, based on the iteratively reweighted least squares (IRLS) framework. Specifically, the proposed approach introduces a two-tier robustification strategy. First, an innovation-level adaptive thresholding mechanism is employed to filter out heavily contaminated data. Second, the outlier-contaminated autocovariance is formulated using an $ε$-contamination model, where the standard LMS criterion is replaced by the Huber cost function. The IRLS method is then utilized to iteratively adjust data weights based on estimation deviations, effectively mitigating the influence of residual outliers. Comparative simulations demonstrate that ALS-IRLS reduces the root-mean-square error (RMSE) of noise covariance estimates by over two orders of magnitude compared to standard ALS. Furthermore, it significantly enhances downstream state estimation accuracy, outperforming existing outlier-robust Kalman filters and achieving performance nearly equivalent to the ideal Oracle lower bound in the presence of noisy and anomalous data.

Long-term time series forecasting (LTSF) is widely recognized as a central challenge in data mining and machine learning. LTSF has increasingly evolved into a benchmark-driven ''GAME,'' where models are ranked, compared, and declared state-of-the-art based primarily on marginal reductions in aggregated pointwise error metrics such as MSE and MAE. Across a small set of canonical datasets and fixed forecasting horizons, progress is communicated through leaderboard-style tables in which lower numerical scores define success. In this GAME, what is measured becomes what is optimized, and incremental error reduction becomes the dominant currency of advancement. We argue that this metric-centric regime is not merely incomplete, but structurally misaligned with the broader objectives of forecasting. In real-world settings, forecasting often prioritizes preserving temporal structure, trend stability, seasonal coherence, robustness to regime shifts, and supporting downstream decision processes. Optimizing aggregate pointwise error does not necessarily imply modeling these structural properties. As a result, leaderboard improvement may increasingly reflect specialization in benchmark configurations rather than a deeper understanding of temporal dynamics. This paper revisits LTSF evaluation as a foundational question in data science: what does it mean to measure forecasting progress? We propose a multi-dimensional evaluation perspective that integrates statistical fidelity, structural coherence, and decision-level relevance. By challenging the current metric monoculture, we aim to redirect attention from winning benchmark tables toward advancing meaningful, context-aware forecasting.

Classifier-Free Guidance (CFG) is a cornerstone of modern conditional diffusion models, yet its reliance on the fixed or heuristic dynamic guidance weight is predominantly empirical and overlooks the inherent dynamics of the diffusion process. In this paper, we provide a rigorous theoretical analysis of the Classifier-Free Guidance. Specifically, we establish strict upper bounds on the score discrepancy between conditional and unconditional distributions at different timesteps based on the diffusion process. This finding explains the limitations of fixed-weight strategies and establishes a principled foundation for time-dependent guidance. Motivated by this insight, we introduce \textbf{Control Classifier-Free Guidance (C$^2$FG)}, a novel, training-free, and plug-in method that aligns the guidance strength with the diffusion dynamics via an exponential decay control function. Extensive experiments demonstrate that C$^2$FG is effective and broadly applicable across diverse generative tasks, while also exhibiting orthogonality to existing strategies.

Large language models (LLMs) and machine translation (MT) systems are increasingly used in our daily lives, but their outputs can reproduce gender bias present in the training data. Most resources for evaluating such biases are designed for English and reflect its sociocultural context, which limits their applicability to other languages. This work addresses this gap by introducing two new datasets to evaluate gender bias in translations involving Basque, a low-resource and genderless language. WinoMTeus adapts the WinoMT benchmark to examine how gender-neutral Basque occupations are translated into gendered languages such as Spanish and French. FLORES+Gender, in turn, extends the FLORES+ benchmark to assess whether translation quality varies when translating from gendered languages (Spanish and English) into Basque depending on the gender of the referent. We evaluate several general-purpose LLMs and open and proprietary MT systems. The results reveal a systematic preference for masculine forms and, in some models, a slightly higher quality for masculine referents. Overall, these findings show that gender bias is still deeply rooted in these models, and highlight the need to develop evaluation methods that consider both linguistic features and cultural context.

Recently, large language models (LLMs) have gained much attention for the emergence of human-comparable capabilities and huge potential. However, for open-domain implicit question-answering problems, LLMs may not be the ultimate solution due to the reasons of: 1) uncovered or out-of-date domain knowledge, 2) one-shot generation and hence restricted comprehensiveness. To this end, this work proposes a gradual knowledge excavation framework for open-domain complex question answering, where LLMs iteratively and actively acquire external information, and then reason based on acquired historical knowledge. Specifically, during each step of the solving process, the model selects an action to execute, such as querying external knowledge or performing a single logical reasoning step, to gradually progress toward a final answer. Our method can effectively leverage plug-and-play external knowledge and dynamically adjust the strategy for solving complex questions. Evaluated on the StrategyQA dataset, our method achieves 78.17% accuracy with less than 6% parameters of its competitors, setting new SOTA for ~10B-scale LLMs.

Existing frameworks for gradient-based training of spiking neural networks face a trade-off: discrete-time methods using surrogate gradients support arbitrary neuron models but introduce gradient bias and constrain spike-time resolution, while continuous-time methods that compute exact gradients require analytical expressions for spike times and state evolution, restricting them to simple neuron types such as Leaky Integrate and Fire (LIF). We introduce the Eventax framework, which resolves this trade-off by combining differentiable numerical ODE solvers with event-based spike handling. Built in JAX, our frame-work uses Diffrax ODE-solvers to compute gradients that are exact with respect to the forward simulation for any neuron model defined by ODEs . It also provides a simple API where users can specify just the neuron dynamics, spike conditions, and reset rules. Eventax prioritises modelling flexibility, supporting a wide range of neuron models, loss functions, and network architectures, which can be easily extended. We demonstrate Eventax on multiple benchmarks, including Yin-Yang and MNIST, using diverse neuron models such as Leaky Integrate-and-fire (LIF), Quadratic Integrate-and-fire (QIF), Exponential integrate-and-fire (EIF), Izhikevich and Event-based Gated Recurrent Unit (EGRU) with both time-to-first-spike and state-based loss functions, demonstrating its utility for prototyping and testing event-based architectures trained with exact gradients. We also demonstrate the application of this framework for more complex neuron types by implementing a multi-compartment neuron that uses a model of dendritic spikes in human layer 2/3 cortical Pyramidal neurons for computation. Code available at https://github.com/efficient-scalable-machine-learning/eventax.

Preference-based alignment methods (e.g., RLHF, DPO) typically optimize a single scalar objective, implicitly averaging over heterogeneous human preferences. In practice, systematic annotator and user-group disagreement makes mean-reward maximization brittle and susceptible to proxy over-optimization. We propose **Disagreement-Aware Alignment via Risk-Constrained Decoding (DARC)**, a retraining-free inference-time method that frames response selection as distributionally robust, risk-sensitive decision making. Given multiple preference samples or scalable disagreement proxies, DARC reranks candidates by maximizing a *KL-robust (entropic)* satisfaction objective, and provides simple deployment controls that cap or penalize the corresponding entropic risk premium relative to the mean, enabling explicit risk budgets without retraining. We provide theoretical characterization linking this decoding rule to principled pessimism and KL-based distributionally robust optimization. Experiments on alignment benchmarks show that DARC reduces disagreement and tail risk while maintaining competitive average quality under noisy, heterogeneous feedback.

Graph anomaly detection (GAD) aims to identify nodes that deviate from normal patterns in structure or features. While recent GNN-based approaches have advanced this task, they struggle with two major challenges: 1) homophily disparity, where nodes exhibit varying homophily at both class and node levels; and 2) limited scalability, as many methods rely on costly whole-graph operations. To address them, we propose SAGAD, a Scalable and Adaptive framework for GAD. SAGAD precomputes multi-hop embeddings and applies reparameterized Chebyshev filters to extract low- and high-frequency information, enabling efficient training and capturing both homophilic and heterophilic patterns. To mitigate node-level homophily disparity, we introduce an Anomaly Context-Aware Adaptive Fusion, which adaptively fuses low- and high-pass embeddings using fusion coefficients conditioned on Rayleigh Quotient-guided anomalous subgraph structures for each node. To alleviate class-level disparity, we design a Frequency Preference Guidance Loss, which encourages anomalies to preserve more high-frequency information than normal nodes. SAGAD supports mini-batch training, achieves linear time and space complexity, and drastically reduces memory usage on large-scale graphs. Theoretically, SAGAD ensures asymptotic linear separability between normal and abnormal nodes under mild conditions. Extensive experiments on 10 benchmarks confirm SAGAD's superior accuracy and scalability over state-of-the-art methods.

We present a novel Explainable methodology for Condition Monitoring, relying on healthy data only. Since faults are rare events, we propose to focus on learning the probability distribution of healthy observations only, and detect Anomalies at runtime. This objective is achieved via the definition of probabilistic measures of deviation from nominality, which allow to detect and anticipate faults. The Bayesian perspective underpinning our approach allows us to perform Uncertainty Quantification to inform decisions. At the same time, we provide descriptive tools to enhance the interpretability of the results, supporting the deployment of the proposed strategy also in safety-critical applications. The methodology is validated experimentally on two use cases: a publicly available benchmark for Predictive Maintenance, and a real-world Helicopter Transmission dataset collected over multiple years. In both applications, the method achieves competitive detection performance with respect to state-of-the-art anomaly detection methods.

Most uncertainty-aware robotic systems collapse prediction uncertainty into a single scalar score and use it to trigger uniform corrective responses. This aggregation obscures whether uncertainty arises from corrupted observations or from mismatch between the learned model and the true system dynamics. As a result, corrective actions may be applied to the wrong component of the closed loop, degrading performance relative to leaving the policy unchanged. We introduce a lightweight post hoc framework that decomposes uncertainty into aleatoric and epistemic components and uses these signals to regulate system responses at inference time. Aleatoric uncertainty is estimated from deviations in the observation distribution using a Mahalanobis density model, while epistemic uncertainty is detected using a noise robust forward dynamics ensemble that isolates model mismatch from measurement corruption. The two signals remain empirically near orthogonal during closed loop execution and enable type specific responses. High aleatoric uncertainty triggers observation recovery, while high epistemic uncertainty moderates control actions. The same signals also regulate adaptive perception by guiding model capacity selection during tracking inference. Experiments demonstrate consistent improvements across both control and perception tasks. In robotic manipulation, the decomposed controller improves task success from 59.4% to 80.4% under compound perturbations and outperforms a combined uncertainty baseline by up to 21.0%. In adaptive tracking inference on MOT17, uncertainty-guided model selection reduces average compute by 58.2% relative to a fixed high capacity detector while preserving detection quality within 0.4%. Code and demo videos are available at https://divake.github.io/uncertainty-decomposition/.

The increasing adoption of Large Language Models (LLMs) has enabled AI scientists to perform complex end-to-end scientific discovery tasks requiring coordination of specialized roles, including idea generation and experimental execution. However, most state-of-the-art AI scientist systems rely on static, hand-designed pipelines and fail to adapt based on accumulated interaction histories. As a result, these systems overlook promising research directions, repeat failed experiments, and pursue infeasible ideas. To address this, we introduce EvoScientist, an evolving multi-agent AI scientist framework that continuously improves research strategies through persistent memory and self-evolution. EvoScientist comprises three specialized agents: a Researcher Agent (RA) for scientific idea generation, an Engineer Agent (EA) for experiment implementation and execution, and an Evolution Manager Agent (EMA) that distills insights from prior interactions into reusable knowledge. EvoScientist contains two persistent memory modules: (i) an ideation memory, which summarizes feasible research directions from top-ranked ideas while recording previously unsuccessful directions; and (ii) an experimentation memory, which captures effective data processing and model training strategies derived from code search trajectories and best-performing implementations. These modules enable the RA and EA to retrieve relevant prior strategies, improving idea quality and code execution success rates over time. Experiments show that EvoScientist outperforms 7 open-source and commercial state-of-the-art systems in scientific idea generation, achieving higher novelty, feasibility, relevance, and clarity via automatic and human evaluation. EvoScientist also substantially improves code execution success rates through multi-agent evolution, demonstrating persistent memory's effectiveness for end-to-end scientific discovery.

Coordinated audio generation based on video inputs typically requires a strict audio-visual (AV) alignment, where both semantics and rhythmics of the generated audio segments shall correspond to those in the video frames. Previous studies leverage a two-stage design where the AV encoders are firstly aligned via contrastive learning, then the encoded video representations guide the audio generation process. We observe that both contrastive learning and global video guidance are effective in aligning overall AV semantics while limiting temporally rhythmic synchronization. In this work, we propose FoleyFlow to first align unimodal AV encoders via masked modeling training, where the masked audio segments are recovered under the guidance of the corresponding video segments. After training, the AV encoders which are separately pretrained using only unimodal data are aligned with semantic and rhythmic consistency. Then, we develop a dynamic conditional flow for the final audio generation. Built upon the efficient velocity flow generation framework, our dynamic conditional flow utilizes temporally varying video features as the dynamic condition to guide corresponding audio segment generations. To this end, we extract coherent semantic and rhythmic representations during masked AV alignment, and use this representation of video segments to guide audio generation temporally. Our audio results are evaluated on the standard benchmarks and largely surpass existing results under several metrics. The superior performance indicates that FoleyFlow is effective in generating coordinated audios that are both semantically and rhythmically coherent to various video sequences.

Ramsa is a developing 41-hour speech corpus of Emirati Arabic designed to support sociolinguistic research and low-resource language technologies. It contains recordings from structured interviews with native speakers and episodes from national television shows. The corpus features 157 speakers (59 female, 98 male), spans subdialects such as Urban, Bedouin, and Mountain/Shihhi, and covers topics such as cultural heritage, agriculture and sustainability, daily life, professional trajectories, and architecture. It consists of 91 monologic and 79 dialogic recordings, varying in length and recording conditions. A 10\% subset was used to evaluate commercial and open-source models for automatic speech recognition (ASR) and text-to-speech (TTS) in a zero-shot setting to establish initial baselines. Whisper-large-v3-turbo achieved the best ASR performance, with average word and character error rates of 0.268 and 0.144, respectively. MMS-TTS-Ara reported the best mean word and character rates of 0.285 and 0.081, respectively, for TTS. These baselines are competitive but leave substantial room for improvement. The paper highlights the challenges encountered and provides directions for future work.

We revisit Vision-Language-Action through a neuroscience-inspired triad. Biologically, the Cerebrum provides stable high-level multimodal priors and remains frozen; the Pons Adapter integrates these cortical features with real-time proprioceptive inputs and compiles intent into execution-ready tokens; and the Cerebellum (ParaCAT) performs fast, parallel categorical decoding for online control, with hysteresis/EMA/temperature/entropy for stability. A fixed-ratio schedule and two-stage feature caching make the system compute-aware and reproducible. Inspired by active, foveated vision, our wrist ROIs are geometrically tied to the end-effector via calibrated projection, providing a movement-stabilized, high-resolution view that is sensitive to fine-grained pose changes and complements the global context of the main view. The design is modular: upgrading the Cerebrum only retrains the Pons; changing robots only trains the Cerebellum; cerebellum-only RL can further refine control without touching high-level semantics. As a concept-and-protocol paper with preliminary evidence, we outline a timing protocol under matched conditions (GPU, resolution, batch) to verify anticipated efficiency gains. We also report preliminary LIBERO evidence showing that split feature caching reduces training time (7.5h to 4.5h) and improves average success (86.5% to 92.5%) under official N1.5 head-only training, and that SaiVLA0 reaches 99.0% mean success.

Model-based offline reinforcement learning (RL) aims to enhance offline RL with a dynamics model that facilitates policy exploration. However, \textit{model exploitation} could occur due to inevitable model errors, degrading algorithm performance. Adversarial model learning offers a theoretical framework to mitigate model exploitation by solving a maximin formulation. Within such a paradigm, RAMBO~\citep{rigter2022rambo} has emerged as a representative and most popular method that provides a practical implementation with model gradient. However, we empirically reveal that severe Q-value underestimation and gradient explosion can occur in RAMBO with only slight hyperparameter tuning, suggesting that it tends to be overly conservative and suffers from unstable model updates. To address these issues, we propose \textbf{RO}bust value-aware \textbf{M}odel learning with \textbf{I}mplicitly differentiable adaptive weighting (ROMI). Instead of updating the dynamics model with model gradient, ROMI introduces a novel robust value-aware model learning approach. This approach requires the dynamics model to predict future states with values close to the minimum Q-value within a scale-adjustable state uncertainty set, enabling controllable conservatism and stable model updates. To further improve out-of-distribution (OOD) generalization during multi-step rollouts, we propose implicitly differentiable adaptive weighting, a bi-level optimization scheme that adaptively achieves dynamics- and value-aware model learning. Empirical results on D4RL and NeoRL datasets show that ROMI significantly outperforms RAMBO and achieves competitive or superior performance compared to other state-of-the-art methods on datasets where RAMBO typically underperforms. Code is available at https://github.com/zq2r/ROMI.git.

Recent advancements in LLM-based information-seeking agents have achieved record-breaking performance on established benchmarks. However, these agents remain heavily reliant on search-engine-indexed knowledge, leaving a critical blind spot: Unindexed Information Seeking (UIS). This paper identifies and explores the UIS problem, where vital information is not captured by search engine crawlers, such as overlooked content, dynamic webpages, and embedded files. Despite its significance, UIS remains an underexplored challenge. To address this gap, we introduce UIS-QA, the first dedicated UIS benchmark, comprising 110 expert-annotated QA pairs. Notably, even state-of-the-art agents experience a drastic performance drop on UIS-QA (e.g., from 70.90 on GAIA and 46.70 on BrowseComp-zh to 24.55 on UIS-QA), underscoring the severity of the problem. To mitigate this, we propose UIS-Digger, a novel multi-agent framework that incorporates dual-mode browsing and enables simultaneous webpage searching and file parsing. With a relatively small $\sim$30B-parameter backbone LLM optimized using SFT and RFT training strategies, UIS-Digger sets a strong baseline at 27.27\%, outperforming systems integrating sophisticated LLMs such as O3 and GPT-4.1. This demonstrates the importance of proactive interaction with unindexed sources for effective and comprehensive information-seeking. Our work not only uncovers a fundamental limitation in current agent evaluation paradigms but also provides the first toolkit for advancing UIS research, defining a new and promising direction for robust information-seeking systems.

Mechanistic modeling provides a biophysically grounded framework for studying the spread of pathological tau protein in tauopathies like Alzheimer's disease. Existing approaches typically model tau propagation as a diffusive process on the brain's structural connectome, reproducing macroscopic patterns but neglecting microscale cellular transport and reaction mechanisms. The Network Transport Model (NTM) was introduced to fill this gap, explaining how region-level progression of tau emerges from microscale biophysical processes. However, the NTM faces a common challenge for complex models defined by large systems of partial differential equations: the inability to perform parameter inference and mechanistic discovery due to high computational burden and slow model simulations. To overcome this barrier, we propose Tau-BNO, a Brain Neural Operator surrogate framework for rapidly approximating NTM dynamics that captures both intra-regional reaction kinetics and inter-regional network transport. Tau-BNO combines a function operator that encodes kinetic parameters with a query operator that preserves initial state information, while approximating anisotropic transport through a spectral kernel that retains directionality. Empirical evaluations demonstrate high predictive accuracy ($R^2\approx$ 0.98) across diverse biophysical regimes and an 89\% performance improvement over state-of-the-art sequence models like Transformers and Mamba, which lack inherent structural priors. By reducing simulation time from hours to seconds, we show that the surrogate model is capable of producing new insights and generating new hypotheses. This framework is readily extensible to a broader class of connectome-based biophysical models, showcasing the transformative value of deep learning surrogates to accelerate analysis of large-scale, computationally intensive dynamical systems.

Understanding and addressing potential safety alignment risks in large language models (LLMs) is critical for ensuring their safe and trustworthy deployment. In this paper, we highlight an insidious safety threat: a compromised LLM can maintain a facade of proper safety alignment while covertly generating harmful content. To achieve this, we finetune the model to understand and apply a steganographic technique. At inference time, we input a prompt that contains a steganographically embedded malicious target question along with a plaintext cover question. The model, in turn, produces a target response similarly embedded within a benign-looking cover response. In this process, human observers only see the model being prompted with a cover question and generating a corresponding cover response, while the malicious content is hidden from view. We demonstrate this invisible safety threat on GPT-4.1 despite the OpenAI finetuning API's safeguards. The finetuned model produces steganographic malicious outputs in response to hidden malicious prompts, while the user interface displays only a fully benign cover interaction. We also replicate the attack on three open-source models, Llama-3.3-70B-Instruct, Phi-4, and Mistral-Small-24B-Base-2501, confirming the generality of our method. We quantitatively evaluate our method on the AdvBench dataset, using Llama-Guard-3-8B for content safety classification. Across all four models, all stegotexts containing malicious content are incorrectly classified as safe.

In scientific reasoning tasks, the veracity of the reasoning process is as critical as the final outcome. While Process Reward Models (PRMs) offer a solution to the coarse-grained supervision problems inherent in Outcome Reward Models (ORMs), their deployment is hindered by the prohibitive cost of obtaining expert-verified step-wise labels. This paper addresses the challenge of training reliable PRMs using abundant but noisy "weak" supervision. We argue that existing Weak-to-Strong Generalization (W2SG) theories lack prescriptive guidelines for selecting high-quality training signals from noisy data. To bridge this gap, we introduce the Dual-Consensus Weak-to-Strong (DC-W2S) framework. By intersecting Self-Consensus (SC) metrics among weak supervisors with Neighborhood-Consensus (NC) metrics in the embedding space, we stratify supervision signals into distinct reliability regimes. We then employ a curriculum of instance-level balanced sampling and label-level reliability-aware masking to guide the training process. We demonstrate that DC-W2S enables the training of robust PRMs for complex reasoning without exhaustive expert annotation, proving that strategic data curation is more effective than indiscriminate training on large-scale noisy datasets.

Large language model (LLM)-based judges are widely adopted for automated evaluation and reward modeling, yet their judgments are often affected by judgment biases. Accurately evaluating these biases is essential for ensuring the reliability of LLM-based judges. However, existing studies typically investigate limited biases under a single judge formulation, either generative or discriminative, lacking a comprehensive evaluation. To bridge this gap, we propose JudgeBiasBench, a benchmark for systematically quantifying biases in LLM-based judges. JudgeBiasBench defines a taxonomy of judgment biases across 4 dimensions, and constructs bias-augmented evaluation instances through a controlled bias injection pipeline, covering 12 representative bias types. We conduct extensive experiments across both generative and discriminative judges, revealing that current judges exhibit significant and diverse bias patterns that often compromise the reliability of automated evaluation. To mitigate judgment bias, we propose bias-aware training that explicitly incorporates bias-related attributes into the training process, encouraging judges to disentangle task-relevant quality from bias-correlated cues. By adopting reinforcement learning for generative judges and contrastive learning for discriminative judges, our methods effectively reduce judgment biases while largely preserving general evaluation capability.

Significant progress has been achieved in subject-driven text-to-image (T2I) generation, which aims to synthesize new images depicting target subjects according to user instructions. However, evaluating these models remains a significant challenge. Existing benchmarks exhibit critical limitations: 1) insufficient diversity and comprehensiveness in subject images, 2) inadequate granularity in assessing model performance across different subject difficulty levels and prompt scenarios, and 3) a profound lack of actionable insights and diagnostic guidance for subsequent model refinement. To address these limitations, we propose DSH-Bench, a comprehensive benchmark that enables systematic multi-perspective analysis of subject-driven T2I models through four principal innovations: 1) a hierarchical taxonomy sampling mechanism ensuring comprehensive subject representation across 58 fine-grained categories, 2) an innovative classification scheme categorizing both subject difficulty level and prompt scenario for granular capability assessment, 3) a novel Subject Identity Consistency Score (SICS) metric demonstrating a 9.4\% higher correlation with human evaluation compared to existing measures in quantifying subject preservation, and 4) a comprehensive set of diagnostic insights derived from the benchmark, offering critical guidance for optimizing future model training paradigms and data construction strategies. Through an extensive empirical evaluation of 19 leading models, DSH-Bench uncovers previously obscured limitations in current approaches, establishing concrete directions for future research and development.

Autoregressive decoding remains a primary bottleneck in large language model (LLM) serving, motivating speculative decoding methods that reduce expensive teacher-model invocations by verifying multiple candidate tokens per step. Tree-structured speculation further increases parallelism, but is often brittle when ported across heterogeneous backends and accelerator stacks, where attention masking, KV-cache layouts, and indexing semantics are not interchangeable. We present EAGLE-Pangu, a reproducible system that ports EAGLE-3-style tree speculative decoding to a Pangu teacher backend on Ascend NPUs. EAGLE-Pangu contributes (i) an explicit branch/commit cache manager built on the Cache API, (ii) accelerator-safe tree tensorization that removes undefined negative indices by construction and validates structural invariants, and (iii) a fused-kernel-compatible teacher verification path with a debuggable eager fallback. On 240 turns from MT-Bench and HumanEval-style prompts, EAGLE-Pangu improves end-to-end decoding throughput by 1.27x on average, up to 2.46x at p99, over teacher-only greedy decoding in the fused-kernel performance path. We also provide a fused-kernel-free reference path with structured traces and invariant checks to support reproducible debugging and ablation across execution modes and tree budgets.

Large Language Models (LLMs) have demonstrated exceptional performance across a wide range of tasks, yet their significant computational and memory requirements present major challenges for deployment. A common approach uses Taylor expansion on the loss function to estimate neuron importance. However, its reliance on one-hot cross entropy loss, a key limitation is that it narrowly assesses importance based only on the probability assigned to the single predicted next token, thereby ignoring the other potential predictions of the original model. An intuitive solution to address this is to employ self distillation criterion for importance evaluation. However, this approach introduces significant computational overhead by requiring a separate teacher model for supervision. To this end, we propose a simple but effective criterion, information entropy of the model's output distribution, to efficiently evaluate importance scores of neurons with Taylor pruning without requirement of additional teacher. Compared to plain cross entropy criterion, it provides a more holistic criterion for Taylor pruning to prune neurons with the least impact on the prediction of model in a global manner, thereby preserving the fidelity of the model's predictive capabilities. Experimental results on extensive zero-shot benchmarks demonstrate that our method consistently outperforms existing pruning methods across the LLaMA and Qwen series models. The source code and trained weights are availabel at https://github.com/visresearch/HFPrune.

Tiny Recursive Models (TRMs) have recently demonstrated remarkable performance on ARC-AGI, showing that very small models can compete against large foundation models through a two-step refinement mechanism that updates an internal reasoning state $z$ and the predicted output $y$. Naturally, such refinement is of interest for any predictor; it is therefore natural to wonder whether the TRM mechanism could be effectively re-adopted in autoregressive models. However, TRMs cannot be simply compared to standard models because they lack causal predictive structures and contain persistent latent states that make it difficult to isolate specific performance gains. In this paper, we propose the Autoregressive TRM and evaluate it on small autoregressive tasks. To understand its efficacy, we propose a suite of models that gradually transform a standard Transformer to a Tiny Autoregressive Recursive Model in a controlled setting that fixes the block design, token stream, and next-token objective. Across compute-matched experiments on character-level algorithmic tasks, we surprisingly find that there are some two-level refinement baselines that show strong performance. Contrary to expectations, we find no reliable performance gains from the full Autoregressive TRM architecture. These results offer potential promise for two-step refinement mechanisms more broadly but caution against investing in the autoregressive TRM-specific model as a fruitful research direction.

Forecasting physiological signals can support proactive monitoring and timely clinical intervention by anticipating critical changes in patient status. In this work, we address multivariate multi-horizon forecasting of physiological time series by jointly predicting heart rate, oxygen saturation, pulse rate, and respiratory rate at forecasting horizons of 15, 30, and 60 seconds. We propose a hybrid quantum-classical architecture that integrates a Variational Quantum Circuit (VQC) within a recurrent neural backbone. A GRU encoder summarizes the historical observation window into a latent representation, which is then projected into quantum angles used to parameterize the VQC. The quantum layer acts as a learnable non-linear feature mixer, modeling cross-variable interactions before the final prediction stage. We evaluate the proposed approach on the BIDMC PPG and Respiration dataset under a Leave-One-Patient-Out protocol. The results show competitive accuracy compared with classical and deep learning baselines, together with greater robustness to noise and missing inputs. These findings suggest that hybrid quantum layers can provide useful inductive biases for physiological time series forecasting in small-cohort clinical settings.

While large language models (LLMs) exhibit strong reasoning abilities, their performance on complex tasks is often constrained by the limitations of their internal knowledge. A compelling approach to overcome this challenge is to augment these models with external tools -- such as Python interpreters for mathematical computations or search engines for retrieving factual information. However, enabling models to use these tools effectively remains a significant challenge. Existing methods typically rely on cold-start pipelines that begin with supervised fine-tuning (SFT), followed by reinforcement learning (RL). These approaches often require substantial amounts of labeled data for SFT, which is expensive to annotate or synthesize. In this work, we propose In-Context Reinforcement Learning (ICRL), an RL-only framework that eliminates the need for SFT by leveraging few-shot prompting during the rollout stage of RL. Specifically, ICRL introduces in-context examples within the rollout prompts to teach the model how to invoke external tools. Furthermore, as training progresses, the number of in-context examples is gradually reduced, eventually reaching a zero-shot setting where the model learns to call tools independently. We conduct extensive experiments across a range of reasoning and tool-use benchmarks. Results show that ICRL achieves state-of-the-art performance, demonstrating its effectiveness as a scalable, data-efficient alternative to traditional SFT-based pipelines.

Structured pruning reduces LLM inference cost by removing low-importance architectural components. This can be viewed as learning a multiplicative gate for each component under an l0 sparsity constraint. Due to the discreteness of the l0 norm, prior work typically adopts stochastic hard-concrete relaxations to enable differentiable optimization; however, this stochasticity can introduce a train--test mismatch when sampled masks are discretized for deployment and restricts masks to a bounded, near-binary range. To address this, we propose Deterministic Differentiable Pruning (DDP), a mask-only optimization method that eliminates stochasticity by directly optimizing a deterministic soft surrogate of the discrete l0 objective. Compared with prior approaches, DDP offers greater expressiveness, reduced train--test mismatch, and faster convergence. We apply our method to several dense and MoE models, including Qwen3-32B and Qwen3-30B-A3B, achieving a performance loss as small as 1% on downstream tasks while outperforming previous methods at 20% sparsity. We further demonstrate end-to-end inference speedups in realistic deployment settings with vLLM.

Accurate phenotype prediction from RNA sequencing (RNA-seq) data is essential for diagnosis, biomarker discovery, and personalized medicine. Deep learning models have demonstrated strong potential to outperform classical machine learning approaches, but their performance relies on large, well-annotated datasets. In transcriptomics, such datasets are frequently limited, leading to over-fitting and poor generalization. Knowledge transfer from larger, more general datasets can alleviate this issue. However, transferring information across RNA-seq datasets remains challenging due to heterogeneous preprocessing pipelines and differences in target phenotypes. In this study, we propose a deep learning-based domain adaptation framework that enables effective knowledge transfer from a large general dataset to a smaller one for cancer type classification. The method learns a domain-invariant latent space by jointly optimizing classification and domain alignment objectives. To ensure stable training and robustness in data-scarce scenarios, the framework is trained with an adversarial approach with appropriate regularization. Both supervised and unsupervised approach variants are explored, leveraging labeled or unlabeled target samples. The framework is evaluated on three large-scale transcriptomic datasets (TCGA, ARCHS4, GTEx) to assess its ability to transfer knowledge across cohorts. Experimental results demonstrate consistent improvements in cancer and tissue type classification accuracy compared to non-adaptive baselines, particularly in low-data scenarios. Overall, this work highlights domain adaptation as a powerful strategy for data-efficient knowledge transfer in transcriptomics, enabling robust phenotype prediction under constrained data conditions.

With the rapid advancement of commercial multi-modal models, image editing has garnered significant attention due to its widespread applicability in daily life. Despite impressive progress, existing image editing systems, particularly closed-source or proprietary models, often struggle with complex, indirect, or multi-step user instructions. These limitations hinder their ability to perform nuanced, context-aware edits that align with human intent. In this work, we propose ImageEdit-R1, a multi-agent framework for intelligent image editing that leverages reinforcement learning to coordinate high-level decision-making across a set of specialized, pretrained vision-language and generative agents. Each agent is responsible for distinct capabilities--such as understanding user intent, identifying regions of interest, selecting appropriate editing actions, and synthesizing visual content--while reinforcement learning governs their collaboration to ensure coherent and goal-directed behavior. Unlike existing approaches that rely on monolithic models or hand-crafted pipelines, our method treats image editing as a sequential decision-making problem, enabling dynamic and context-aware editing strategies. Experimental results demonstrate that ImageEdit-R1 consistently outperforms both individual closed-source diffusion models and alternative multi-agent framework baselines across multiple image editing datasets.

Large Language Models (LLMs) are pivotal in natural language processing. The impracticality of full fine-tuning has prompted Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA), optimizing low-rank matrices A and B. In distributed scenarios where privacy constraints necessitate Federated Learning (FL), however, the integration of LoRA is often unstable. Specifically, we identify that aggregating updates from multiple clients introduces statistical variance that scales with the client count, causing gradient collapse when using high-rank adapters. Existing scaling factor candidates, such as the one used by Rank-Stabilized LoRA, ignore the interaction caused by the aggregation process. To bridge this gap, this paper introduces Stabilized Federated LoRA (SFed-LoRA), a framework that theoretically characterizes the interaction between adapter rank and federated aggregation. We derive an optimal scaling factor designed to effectively mitigate the aggregation error accumulating across N clients. By correcting the scaling mismatch inherent in previous approaches, SFed-LoRA restores the efficacy of high-rank adaptation without altering the original model architecture or increasing inference latency. Extensive experiments in diverse tasks, model architectures, and heterogeneous data distributions are conducted to validate our results. We demonstrate that SFed-LoRA prevents high-rank collapse, and achieves significantly improved stability and faster convergence compared with state-of-the-art baselines for high-rank adaptation.

While recent feed-forward 3D reconstruction models accelerate 3D reconstruction by jointly inferring dense geometry and camera poses in a single pass, their reliance on dense attention imposes a quadratic complexity, creating a prohibitive computational bottleneck that severely limits inference speed. To resolve this, we introduce Speed3R, an end-to-end trainable model inspired by the core principle of Structure-from-Motion: that a sparse set of keypoints is sufficient for robust pose estimation. Speed3R features a dual-branch attention mechanism where a compression branch creates a coarse contextual prior to guide a selection branch, which performs fine-grained attention only on the most informative image tokens. This strategy mimics the efficiency of traditional keypoint matching, achieving a remarkable 12.4x inference speedup on 1000-view sequences, while introducing a minimal, controlled trade-off in geometric accuracy. Validated on standard benchmarks with both VGGT and $π^3$ backbones, our method delivers high-quality reconstructions at a fraction of computational cost, paving the way for efficient large-scale scene modeling.

The relationship between content production and consumption on algorithm-driven platforms like YouTube plays a critical role in shaping ideological behaviors. While prior work has largely focused on user behavior and algorithmic recommendations, the interplay between what is produced and what gets consumed, and its role in ideological shifts remains understudied. In this paper, we present a longitudinal, mixed-methods analysis combining one year of YouTube watch history with two waves of ideological surveys from 1,100 U.S. participants. We identify users who exhibited significant shifts toward more extreme ideologies and compare their content consumption and the production patterns of YouTube channels they engaged with to ideologically stable users. Our findings show that users who became more extreme consumed have different consumption habits from those who do not. This gets amplified by the fact that channels favored by users with extreme ideologies also have a higher affinity to produce content with a higher anger, grievance and other such markers. Lastly, using time series analysis, we examine whether content producers are the primary drivers of consumption behavior or merely responding to user demand.

Fault diagnosis is critical for the safe operation of industrial systems. Conventional diagnosis models typically produce abstract outputs such as anomaly scores or fault categories, failing to answer critical operational questions like "Why" or "How to repair". While large language models (LLMs) offer strong generalization and reasoning abilities, their training on discrete textual corpora creates a semantic gap when processing high-dimensional, temporal industrial signals. To address this challenge, we propose a Signals-to-Semantics fault diagnosis (S2S-FDD) framework that bridges high-dimensional sensor signals with natural language semantics through two key innovations: We first design a Signal-to-Semantic operator to convert abstract time-series signals into natural language summaries, capturing trends, periodicity, and deviations. Based on the descriptions, we design a multi-turn tree-structured diagnosis method to perform fault diagnosis by referencing historical maintenance documents and dynamically querying additional signals. The framework further supports human-in-the-loop feedback for continuous refinement. Experiments on the multiphase flow process show the feasibility and effectiveness of the proposed method for explainable zero-shot fault diagnosis.

Reward modeling is essential for aligning Large Language Models(LLMs) with human preferences, yet conventional reward models suffer from poor interpretability and heavy reliance on costly expert annotations. While recent rubric-based approaches enhance evaluation transparency, they lack systematic quality control, yielding noisy and redundant criteria, failing to mitigate persistent biases (e.g., verbosity, position) in LLM evaluators, and creating a scalability-reliability trade-off. To address these limitations, we propose CDRRM (Contrast-Driven Rubric Reward Model), a framework built on a novel Contrast-then-Synthesis paradigm for high-quality rubric generation and guided preference judgment. CDRRM first conducts multi-dimensional contrastive profiling on preference pairs to identify causal discriminative factors, then synthesizes these insights into compact, context-aware rubrics to guide preference judg- ments. Extensive experiments on three authoritative benchmarks (RewardBench, RMBench, RMB) demonstrate that CDRRM achieves state-of-the-art performance across diverse domains and effectively mitigates aforementioned evaluation biases. Notably, our approach delivers exceptional data efficiency: training the rubric generator on only 3k high-quality samples empowers a frozen pre-trained judge model to outperform fully fine-tuned baselines. This work offers a scalable, interpretable, and data-efficient path for reward modeling.

Emotion recognition in real-world environments is hindered by partial occlusions, missing modalities, and severe class imbalance. To address these issues, particularly for the Affective Behavior Analysis in-the-wild (ABAW) Expression challenge, we propose a multimodal framework that dynamically fuses visual and audio representations. Our approach uses a dual-branch Transformer architecture featuring a safe cross-attention mechanism and a modality dropout strategy. This design allows the network to rely on audio-based predictions when visual cues are absent. To mitigate the long-tail distribution of the Aff-Wild2 dataset, we apply focal loss optimization, combined with a sliding-window soft voting strategy to capture dynamic emotional transitions and reduce frame-level classification jitter. Experiments demonstrate that our framework effectively handles missing modalities and complex spatiotemporal dependencies, achieving an accuracy of 60.79% and an F1-score of 0.5029 on the Aff-Wild2 validation set.

Exogenous variables offer valuable supplementary information for predicting future endogenous variables. Forecasting with exogenous variables needs to consider both past-to-future dependencies (i.e., temporal correlations) and the influence of exogenous variables on endogenous variables (i.e., channel correlations). This is pivotal when future exogenous variables are available, because they may directly affect the future endogenous variables. Many methods have been proposed for time series forecasting with exogenous variables, focusing on modeling temporal and channel correlations. However, most of them use a two-step strategy, modeling temporal and channel correlations separately, which limits their ability to capture joint correlations across time and channels. Furthermore, in real-world scenarios, time series are frequently affected by various forms of noises, underscoring the critical importance of robustness in such correlations modeling. To address these limitations, we propose GCGNet, a Graph-Consistent Generative Network for time series forecasting with exogenous variables. Specifically, GCGNet first employs a Variational Generator to produce coarse predictions. A Graph Structure Aligner then further guides it by evaluating the consistency between the generated and true correlations, where the correlations are represented as graphs, and are robust to noises. Finally, a Graph Refiner is proposed to refine the predictions to prevent degeneration and improve accuracy. Extensive experiments on 12 real-world datasets demonstrate that GCGNet outperforms state-of-the-art baselines.

Masked Diffusion Language Models (MDLMs) enable parallel token decoding, providing a promising alternative to the sequential nature of autoregressive generation. However, their iterative denoising process remains computationally expensive because it repeatedly processes the entire sequence at every step. We observe that across these diffusion steps, most token representations remain stable; only a small subset, which we term salient tokens, contributes meaningfully to the next update. Leveraging this temporal sparsity, we present DyLLM, a training-free inference framework that accelerates decoding by selectively computing only these salient tokens. DyLLM identifies saliency by measuring the cosine similarity of attention contexts between adjacent denoising steps. It recomputes feed-forward and attention operations only for salient tokens while reusing cached activations for the remainder. Across diverse reasoning and code-generation benchmarks, DyLLM achieves up to 9.6x higher throughput while largely preserving the baseline accuracy of state-of-the-art models like LLaDA and Dream.

As large language models (LLMs) evolve into autonomous agents capable of acting in open-ended environments, ensuring behavioral alignment with human values becomes a critical safety concern. Existing benchmarks, focused on static, single-turn prompts, fail to capture the interactive and multi-modal nature of real-world conflicts. We introduce ConflictBench, a benchmark for evaluating human-AI conflict through 150 multi-turn scenarios derived from prior alignment queries. ConflictBench integrates a text-based simulation engine with a visually grounded world model, enabling agents to perceive, plan, and act under dynamic conditions. Empirical results show that while agents often act safely when human harm is immediate, they frequently prioritize self-preservation or adopt deceptive strategies in delayed or low-risk settings. A regret test further reveals that aligned decisions are often reversed under escalating pressure, especially with visual input. These findings underscore the need for interaction-level, multi-modal evaluation to surface alignment failures that remain hidden in conventional benchmarks.

Dance is a form of human motion characterized by emotional expression and communication, playing a role in various fields such as music, virtual reality, and content creation. Existing methods for dance generation often fail to adequately capture the inherently sequential, rhythmical, and music-synchronized characteristics of dance. In this paper, we propose \emph{MambaDance}, a new dance generation approach that leverages a Mamba-based diffusion model. Mamba, well-suited to handling long and autoregressive sequences, is integrated into our two-stage diffusion architecture, substituting off-the-shelf Transformer. Additionally, considering the critical role of musical beats in dance choreography, we propose a Gaussian-based beat representation to explicitly guide the decoding of dance sequences. Experiments on AIST++ and FineDance datasets for each sequence length show that our proposed method effectively generates plausible dance movements while reflecting essential characteristics, consistently from short to long dances, compared to the previous methods. Additional qualitative results and demo videos are available at \small{https://vision3d-lab.github.io/mambadance}.

A data mixture refers to how different data sources are combined to train large language models, and selecting an effective mixture is crucial for optimal downstream performance. Existing methods either conduct costly searches directly on the target model or rely on mixture scaling laws that fail to extrapolate well to large model sizes. We address these limitations by introducing a compute-efficient pipeline for data mixture scaling. First, we propose CAMEL, a capacity-aware mixture law that models validation loss with the nonlinear interplay between model size and mixture. We also introduce a loss-to-benchmark prediction law that estimates benchmark accuracy from validation loss, enabling end-to-end performance prediction for the target model. Next, we study how to allocate a fixed compute budget across model scales to fit the law and reduce prediction error. Finally, we apply our method to Mixture-of-Experts models with up to 7B-A150M parameters to fit the law, and verify the optimal mixture derived from the law by extrapolating to a 55B-A1.2B target model. Compared to prior methods, we reduces mixture optimization costs by 50\% and improves downstream benchmark performance by up to 3\%.

In real-world collaboration, alignment, process structure, and outcome quality do not exhibit a simple linear or one-to-one correspondence: similar alignment may accompany either rapid convergence or extensive multi-branch exploration, and lead to different results. Existing accounts often isolate these dimensions or focus on specific participant types, limiting structural accounts of collaboration. We reconceptualize collaboration through two complementary lenses. The task lens models collaboration as trajectory evolution in a structured task space, revealing patterns such as advancement, branching, and backtracking. The intent lens examines how individual intents are expressed within shared contexts and enter situated decisions. Together, these lenses clarify the structural relationships among alignment, decision-making, and trajectory structure. Rather than reducing collaboration to outcome quality or treating alignment as the sole objective, we propose a unified dynamic view of the relationships among alignment, process, and outcome, and use it to re-examine collaboration structure across Human-Human, AI-AI, and Human-AI settings.

Federated fine-tuning of large language models (LLMs) with low-rank adaptation (LoRA) offers a communication-efficient and privacy-preserving solution for task-specific adaptation. Naive aggregation of LoRA modules introduces noise due to mathematical incorrectness when averaging the downsampling and upsampling matrices independently. However, existing noise-free aggregation strategies inevitably compromise the structural expressiveness of LoRA, limiting its ability to retain client-specific adaptations by either improperly reconstructing the low-rank structure or excluding partially trainable components. We identify this problem as loss of training momentum, where LoRA updates fail to accumulate effectively across rounds, resulting in slower convergence and suboptimal performance. To address this, we propose FedMomentum, a novel framework that enables structured and momentum-preserving LoRA aggregation via singular value decomposition (SVD). Specifically, after aggregating low-rank updates in a mathematically correct manner, FedMomentum applies SVD to extract the dominant components that capture the main update directions. These components are used to reconstruct the LoRA modules with the same rank, while residual components can be retained and later merged into the backbone to preserve semantic information and ensure robustness. Extensive experiments across multiple tasks demonstrate that FedMomentum consistently outperforms prior state-of-the-art methods in convergence speed and final accuracy.