The Inference Report

Recent advances in video diffusion have enabled the development of "world models" capable of simulating interactive environments. However, these models are largely restricted to single-agent settings, failing to control multiple agents simultaneously in a scene. In this work, we tackle a fundamental issue of action binding in existing video diffusion models, which struggle to associate specific actions with their corresponding subjects. For this purpose, we propose ActionParty, an action controllable multi-subject world model for generative video games. It introduces subject state tokens, i.e. latent variables that persistently capture the state of each subject in the scene. By jointly modeling state tokens and video latents with a spatial biasing mechanism, we disentangle global video frame rendering from individual action-controlled subject updates. We evaluate ActionParty on the Melting Pot benchmark, demonstrating the first video world model capable of controlling up to seven players simultaneously across 46 diverse environments. Our results show significant improvements in action-following accuracy and identity consistency, while enabling robust autoregressive tracking of subjects through complex interactions.

Pretrained Vision Transformers (ViTs) such as DINOv2 and MAE provide generic image features that can be applied to a variety of downstream tasks such as retrieval, classification, and segmentation. However, such representations tend to focus on the most salient visual cues in the image, with no way to direct them toward less prominent concepts of interest. In contrast, Multimodal LLMs can be guided with textual prompts, but the resulting representations tend to be language-centric and lose their effectiveness for generic visual tasks. To address this, we introduce Steerable Visual Representations, a new class of visual representations, whose global and local features can be steered with natural language. While most vision-language models (e.g., CLIP) fuse text with visual features after encoding (late fusion), we inject text directly into the layers of the visual encoder (early fusion) via lightweight cross-attention. We introduce benchmarks for measuring representational steerability, and demonstrate that our steerable visual features can focus on any desired objects in an image while preserving the underlying representation quality. Our method also matches or outperforms dedicated approaches on anomaly detection and personalized object discrimination, exhibiting zero-shot generalization to out-of-distribution tasks.

Language models (LMs) are increasingly extended with new learnable vocabulary tokens for domain-specific tasks, such as Semantic-ID tokens in generative recommendation. The standard practice initializes these new tokens as the mean of existing vocabulary embeddings, then relies on supervised fine-tuning to learn their representations. We present a systematic analysis of this strategy: through spectral and geometric diagnostics, we show that mean initialization collapses all new tokens into a degenerate subspace, erasing inter-token distinctions that subsequent fine-tuning struggles to fully recover. These findings suggest that \emph{token initialization} is a key bottleneck when extending LMs with new vocabularies. Motivated by this diagnosis, we propose the \emph{Grounded Token Initialization Hypothesis}: linguistically grounding novel tokens in the pretrained embedding space before fine-tuning better enables the model to leverage its general-purpose knowledge for novel-token domains. We operationalize this hypothesis as GTI (Grounded Token Initialization), a lightweight grounding stage that, prior to fine-tuning, maps new tokens to distinct, semantically meaningful locations in the pretrained embedding space using only paired linguistic supervision. Despite its simplicity, GTI outperforms both mean initialization and existing auxiliary-task adaptation methods in the majority of evaluation settings across multiple generative recommendation benchmarks, including industry-scale and public datasets. Further analyses show that grounded embeddings produce richer inter-token structure that persists through fine-tuning, corroborating the hypothesis that initialization quality is a key bottleneck in vocabulary extension.

Large Language Models employing Chain-of-Thought reasoning achieve strong performance but suffer from excessive token consumption that inflates inference costs. Existing efficiency methods such as explicit length penalties, difficulty estimators, or multi-stage curricula either degrade reasoning quality or require complex training pipelines. We introduce Batched Contextual Reinforcement, a minimalist, single-stage training paradigm that unlocks efficient reasoning through a simple structural modification: training the model to solve N problems simultaneously within a shared context window, rewarded purely by per-instance accuracy. This formulation creates an implicit token budget that yields several key findings: (1) We identify a novel task-scaling law: as the number of concurrent problems N increases during inference, per-problem token usage decreases monotonically while accuracy degrades far more gracefully than baselines, establishing N as a controllable throughput dimension. (2) BCR challenges the traditional accuracy-efficiency trade-off by demonstrating a "free lunch" phenomenon at standard single-problem inference. Across both 1.5B and 4B model families, BCR reduces token usage by 15.8% to 62.6% while consistently maintaining or improving accuracy across five major mathematical benchmarks. (3) Qualitative analyses reveal emergent self-regulated efficiency, where models autonomously eliminate redundant metacognitive loops without explicit length supervision. (4) Crucially, we empirically demonstrate that implicit budget constraints successfully circumvent the adversarial gradients and catastrophic optimization collapse inherent to explicit length penalties, offering a highly stable, constraint-based alternative for length control. These results prove BCR practical, showing simple structural incentives unlock latent high-density reasoning in LLMs.

When posed with prompts that permit a large number of valid answers, comprehensively generating them is the first step towards satisfying a wide range of users. In this paper, we study methods to elicit a comprehensive set of valid responses. To evaluate this, we introduce \textbf{diversity coverage}, a metric that measures the total quality scores assigned to each \textbf{unique} answer in the predicted answer set relative to the best possible answer set with the same number of answers. Using this metric, we evaluate 18 LLMs, finding no single model dominates at generating diverse responses to a wide range of open-ended prompts. Yet, per each prompt, there exists a model that outperforms all other models significantly at generating a diverse answer set. Motivated by this finding, we introduce a router that predicts the best model for each query. On NB-Wildchat, our trained router outperforms the single best model baseline (26.3% vs $23.8%). We further show generalization to an out-of-domain dataset (NB-Curated) as well as different answer-generation prompting strategies. Our work lays foundation for studying generating comprehensive answers when we have access to a suite of models.

Standard LLM benchmarks evaluate the assistant turn: the model generates a response to an input, a verifier scores correctness, and the analysis ends. This paradigm leaves unmeasured whether the LLM encodes any awareness of what follows the assistant response. We propose user-turn generation as a probe of this gap: given a conversation context of user query and assistant response, we let a model generate under the user role. If the model's weights encode interaction awareness, the generated user turn will be a grounded follow-up that reacts to the preceding context. Through experiments across $11$ open-weight LLMs (Qwen3.5, gpt-oss, GLM) and $5$ datasets (math reasoning, instruction following, conversation), we show that interaction awareness is decoupled from task accuracy. In particular, within the Qwen3.5 family, GSM8K accuracy scales from $41\%$ ($0.8$B) to $96.8\%$ ($397$B-A$17$B), yet genuine follow-up rates under deterministic generation remain near zero. In contrast, higher temperature sampling reveals interaction awareness is latent with follow up rates reaching $22\%$. Controlled perturbations validate that the proposed probe measures a real property of the model, and collaboration-oriented post-training on Qwen3.5-2B demonstrates an increase in follow-up rates. Our results show that user-turn generation captures a dimension of LLM behavior, interaction awareness, that is unexplored and invisible with current assistant-only benchmarks.

Neural network field theory formulates field theory as a statistical ensemble of fields defined by a network architecture and a density on its parameters. We extend the construction to topological settings via the inclusion of discrete parameters that label the topological quantum number. We recover the Berezinskii--Kosterlitz--Thouless transition, including the spin-wave critical line and the proliferation of vortices at high temperatures. We also verify the T-duality of the bosonic string, showing invariance under the exchange of momentum and winding on $S^1$, the transformation of the sigma model couplings according to the Buscher rules on constant toroidal backgrounds, the enhancement of the current algebra at self-dual radius, and non-geometric T-fold transition functions.

Doubly stochastic matrices enable learned mixing across residual streams, but parameterizing the set of doubly stochastic matrices (the Birkhoff polytope) exactly and efficiently remains an open challenge. Existing exact methods scale factorially with the number of streams ($d$), while Kronecker-factorized approaches are efficient but expressivity-limited. We introduce a novel exact parameterization grounded in the theory of generalized orthostochastic matrices, which scales as $\mathcal{O}(d^3)$ and exposes a single hyperparameter $s$ which continuously interpolates between a computationally efficient boundary and the fully expressive Birkhoff polytope. Building on Manifold-Constrained Hyper-Connections ($m$HC), a framework for learned dynamic layer connectivity, we instantiate this parameterization in go-$m$HC. Our method composes naturally with Kronecker-factorized methods, substantially recovering expressivity at similar FLOP costs. Spectral analysis indicates that go-$m$HC fills the Birkhoff polytope far more completely than Kronecker-factorized baselines. On synthetic stream-mixing tasks, go-$m$HC achieves the minimum theoretical loss while converging up to $10\times$ faster. We validate our approach in a 30M parameter GPT-style language model. The expressivity, efficiency, and exactness of go-$m$HC offer a practical avenue for scaling $d$ as a new dimension of model capacity.

Existing video object removal methods excel at inpainting content "behind" the object and correcting appearance-level artifacts such as shadows and reflections. However, when the removed object has more significant interactions, such as collisions with other objects, current models fail to correct them and produce implausible results. We present VOID, a video object removal framework designed to perform physically-plausible inpainting in these complex scenarios. To train the model, we generate a new paired dataset of counterfactual object removals using Kubric and HUMOTO, where removing an object requires altering downstream physical interactions. During inference, a vision-language model identifies regions of the scene affected by the removed object. These regions are then used to guide a video diffusion model that generates physically consistent counterfactual outcomes. Experiments on both synthetic and real data show that our approach better preserves consistent scene dynamics after object removal compared to prior video object removal methods. We hope this framework sheds light on how to make video editing models better simulators of the world through high-level causal reasoning.

Softmax can become a computational bottleneck in the Transformer model's Multi-Head Attention (MHA) block, particularly in small models under low-precision inference, where exponentiation and normalization incur significant overhead. As such, we suggest using Head-Calibrated Clipped-Linear Softmax (HCCS), a bounded, monotone surrogate to the exponential softmax function, which uses a clipped linear mapping of the max centered attention logits. This approximation produces a stable probability distribution, maintains the ordering of the original logits and has non-negative values. HCCS differs from previous softmax surrogates as it includes a set of lightweight calibration parameters that are optimized offline based on a representative dataset and calibrated for each individual attention head to preserve the statistical properties of the individual heads. We describe a hardware-motivated implementation of HCCS for high-throughput scenarios targeting the AMD Versal AI Engines. The current reference implementations from AMD for this platform rely upon either bfloat16 arithmetic or LUTs to perform the exponential operation, which might limit the throughput of the platform and fail to utilize the high-throughput integer vector processing units of the AI Engine. In contrast, HCCS provides a natural mapping to the AI Engines' int8 multiply accumulate (MAC) units. To the best of our knowledge, this is the first int8 optimized softmax surrogate for AMD AI engines that significantly exceeds the speed performance of other reference implementations while maintaining competitive task accuracy on small or heavily quantized MHA workloads after quantization-aware retraining.

Recent multimodal large language models have achieved strong performance in unified text and image understanding and generation, yet extending such native capability to 3D remains challenging due to limited data. Compared to abundant 2D imagery, high-quality 3D assets are scarce, making 3D synthesis under-constrained. Existing methods often rely on indirect pipelines that edit in 2D and lift results into 3D via optimization, sacrificing geometric consistency. We present Omni123, a 3D-native foundation model that unifies text-to-2D and text-to-3D generation within a single autoregressive framework. Our key insight is that cross-modal consistency between images and 3D can serve as an implicit structural constraint. By representing text, images, and 3D as discrete tokens in a shared sequence space, the model leverages abundant 2D data as a geometric prior to improve 3D representations. We introduce an interleaved X-to-X training paradigm that coordinates diverse cross-modal tasks over heterogeneous paired datasets without requiring fully aligned text-image-3D triplets. By traversing semantic-visual-geometric cycles (e.g., text to image to 3D to image) within autoregressive sequences, the model jointly enforces semantic alignment, appearance fidelity, and multi-view geometric consistency. Experiments show that Omni123 significantly improves text-guided 3D generation and editing, demonstrating a scalable path toward multimodal 3D world models.

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models. While Group Relative Policy Optimization (GRPO) is widely adopted, its coarse credit assignment uniformly penalizes failed rollouts, lacking the token-level focus needed to efficiently address specific deviations. Self-Distillation Policy Optimization (SDPO) addresses this by providing denser, more targeted logit-level supervision that facilitates rapid early improvement, yet it frequently collapses during prolonged training. We trace this late-stage instability to two intrinsic flaws: self-distillation on already-correct samples introduces optimization ambiguity, and the self-teacher's signal reliability progressively degrades. To resolve these issues, we propose Sample-Routed Policy Optimization (SRPO), a unified on-policy framework that routes correct samples to GRPO's reward-aligned reinforcement and failed samples to SDPO's targeted logit-level correction. SRPO further incorporates an entropy-aware dynamic weighting mechanism to suppress high-entropy, unreliable distillation targets while emphasizing confident ones. Evaluated across five benchmarks and two model scales, SRPO achieves both the rapid early improvement of SDPO and the long-horizon stability of GRPO. It consistently surpasses the peak performance of both baselines, raising the five-benchmark average on Qwen3-8B by 3.4% over GRPO and 6.3% over SDPO, while simultaneously yielding moderate response lengths and lowering per-step compute cost by up to 17.2%.

Long-horizon conversational agents require persistent memory for coherent reasoning, yet uncontrolled accumulation causes temporal decay and false memory propagation. Benchmarks such as LOCOMO and LOCCO report performance degradation from 0.455 to 0.05 across stages, while MultiWOZ shows 78.2% accuracy with 6.8% false memory rate under persistent retention. This work introduces an adaptive budgeted forgetting framework that regulates memory through relevanceguided scoring and bounded optimization. The approach integrates recency, frequency, and semantic alignment to maintain stability under constrained context. Comparative analysis demonstrates improved long-horizon F1 beyond 0.583 baseline levels, higher retention consistency, and reduced false memory behavior without increasing context usage. These findings confirm that structured forgetting preserves reasoning performance while preventing unbounded memory growth in extended conversational settings.

Agentic AI shifts the investor's role from analytical execution to oversight. We present an agentic strategic asset allocation pipeline in which approximately 50 specialized agents produce capital market assumptions, construct portfolios using over 20 competing methods, and critique and vote on each other's output. A researcher agent proposes new portfolio construction methods not yet represented, and a meta-agent compares past forecasts against realized returns and rewrites agent code and prompts to improve future performance. The entire pipeline is governed by the Investment Policy Statement--the same document that guides human portfolio managers can now constrain and direct autonomous agents.

Translating machine code into human-readable high-level languages is an open research problem in reverse engineering. Despite recent advancements in LLM-based decompilation to C, modern languages like Dart and Swift are unexplored. In this paper, we study the use of small specialized LLMs as an idiomatic decompiler for such languages. Additionally, we investigate the augmentation of training data using synthetic same-language examples, and compare it against adding human-written examples using related-language (Swift -> Dart). We apply CODEBLEU to evaluate the decompiled code readability and compile@k to measure the syntax correctness. Our experimental results show that on a 73-function Dart test dataset (representing diverse complexity levels), our 4B specialized model achieves 71.3 CODEBLEU (95% CI 65.5-77.1), approximately comparable to a ~480B code model (73.1; 67.4-78.8). On a subset of 34 natural Dart functions, it reaches compile@k5 = 79.4% (Wilson 95% CI 63.2-89.7), vs. 64.7% (47.9-78.5) for the base model; the difference is suggestive but not statistically significant at 0.05. Our results indicate that adding Swift training data helps at 8B but not at 4B, suggesting a capacity threshold for effective cross-lingual transfer. Our experimental results show that small specialized models can generate readable, idiomatic Dart with meaningful identifiers while using minimal compute.

Regulatory documents encode legally binding obligations that LLM-based systems must respect. Yet converting dense, hierarchically structured legal text into machine-readable rules remains a costly, expert-intensive process. We present De Jure, a fully automated, domain-agnostic pipeline for extracting structured regulatory rules from raw documents, requiring no human annotation, domain-specific prompting, or annotated gold data. De Jure operates through four sequential stages: normalization of source documents into structured Markdown; LLM-driven semantic decomposition into structured rule units; multi-criteria LLM-as-a-judge evaluation across 19 dimensions spanning metadata, definitions, and rule semantics; and iterative repair of low-scoring extractions within a bounded regeneration budget, where upstream components are repaired before rule units are evaluated. We evaluate De Jure across four models on three regulatory corpora spanning finance, healthcare, and AI governance. On the finance domain, De Jure yields consistent and monotonic improvement in extraction quality, reaching peak performance within three judge-guided iterations. De Jure generalizes effectively to healthcare and AI governance, maintaining high performance across both open- and closed-source models. In a downstream compliance question-answering evaluation via RAG, responses grounded in De Jure extracted rules are preferred over prior work in 73.8% of cases at single-rule retrieval depth, rising to 84.0% under broader retrieval, confirming that extraction fidelity translates directly into downstream utility. These results demonstrate that explicit, interpretable evaluation criteria can substitute for human annotation in complex regulatory domains, offering a scalable and auditable path toward regulation-grounded LLM alignment.

Generative models for crystalline materials often rely on equivariant graph neural networks, which capture geometric structure well but are costly to train and slow to sample. We present Crystalite, a lightweight diffusion Transformer for crystal modeling built around two simple inductive biases. The first is Subatomic Tokenization, a compact chemically structured atom representation that replaces high-dimensional one-hot encodings and is better suited to continuous diffusion. The second is the Geometry Enhancement Module (GEM), which injects periodic minimum-image pair geometry directly into attention through additive geometric biases. Together, these components preserve the simplicity and efficiency of a standard Transformer while making it better matched to the structure of crystalline materials. Crystalite achieves state-of-the-art results on crystal structure prediction benchmarks, and de novo generation performance, attaining the best S.U.N. discovery score among the evaluated baselines while sampling substantially faster than geometry-heavy alternatives.

Agent skills, structured packages of procedural knowledge and executable resources that agents dynamically load at inference time, have become a reliable mechanism for augmenting LLM agents. Yet inference-time skill augmentation is fundamentally limited: retrieval noise introduces irrelevant guidance, injected skill content imposes substantial token overhead, and the model never truly acquires the knowledge it merely follows. We ask whether skills can instead be internalized into model parameters, enabling zero-shot autonomous behavior without any runtime skill retrieval. We introduce SKILL0, an in-context reinforcement learning framework designed for skill internalization. SKILL0 introduces a training-time curriculum that begins with full skill context and progressively withdraws it. Skills are grouped offline by category and rendered with interaction history into a compact visual context, teaching he model tool invocation and multi-turn task completion. A Dynamic Curriculum then evaluates each skill file's on-policy helpfulness, retaining only those from which the current policy still benefits within a linearly decaying budget, until the agent operates in a fully zero-shot setting. Extensive agentic experiments demonstrate that SKILL0 achieves substantial improvements over the standard RL baseline (+9.7\% for ALFWorld and +6.6\% for Search-QA), while maintaining a highly efficient context of fewer than 0.5k tokens per step. Our code is available at https://github.com/ZJU-REAL/SkillZero.

Learning-based control methods typically assume stationary system dynamics, an assumption often violated in real-world systems due to drift, wear, or changing operating conditions. We study reinforcement learning for control under time-varying dynamics. We consider a continual model-based reinforcement learning setting in which an agent repeatedly learns and controls a dynamical system whose transition dynamics evolve across episodes. We analyze the problem using Gaussian process dynamics models under frequentist variation-budget assumptions. Our analysis shows that persistent non-stationarity requires explicitly limiting the influence of outdated data to maintain calibrated uncertainty and meaningful dynamic regret guarantees. Motivated by these insights, we propose a practical optimistic model-based reinforcement learning algorithm with adaptive data buffer mechanisms and demonstrate improved performance on continuous control benchmarks with non-stationary dynamics.

To harness the power of Language Models in answering domain specific specialized technical questions, Retrieval Augmented Generation (RAG) is been used widely. In this work, we have developed a Q\&A application inspired by the Retrieval Augmented Generation (RAG), which is comprised of an in-house database indexed on the arXiv articles related to the Electron-Ion Collider (EIC) experiment - one of the largest international scientific collaboration and incorporated an open-source LLaMA model for answer generation. This is an extension to it's proceeding application built on proprietary model and Cloud-hosted external knowledge-base for the EIC experiment. This locally-deployed RAG-system offers a cost-effective, resource-constraint alternative solution to build a RAG-assisted Q\&A application on answering domain-specific queries in the field of experimental nuclear physics. This set-up facilitates data-privacy, avoids sending any pre-publication scientific data and information to public domain. Future improvement will expand the knowledge base to encompass heterogeneous EIC-related publications and reports and upgrade the application pipeline orchestration to the LangGraph framework.

In this paper, we consider a multi-armed bandit (MAB) instance and study how to identify the best arm when arm commands are conveyed from a central learner to a distributed agent over a discrete memoryless channel (DMC). Depending on the agent capabilities, we provide communication schemes along with their analysis, which interestingly relate to the zero-error capacity of the underlying DMC.

Understanding causal dependencies in observational data is critical for informing decision-making. These relationships are often modeled as Bayesian Networks (BNs) and Directed Acyclic Graphs (DAGs). Existing methods, such as NOTEARS and DAG-GNN, often face issues with scalability and stability in high-dimensional data, especially when there is a feature-sample imbalance. Here, we show that the denoising score matching objective of diffusion models could smooth the gradients for faster, more stable convergence. We also propose an adaptive k-hop acyclicity constraint that improves runtime over existing solutions that require matrix inversion. We name this framework Denoising Diffusion Causal Discovery (DDCD). Unlike generative diffusion models, DDCD utilizes the reverse denoising process to infer a parameterized causal structure rather than to generate data. We demonstrate the competitive performance of DDCDs on synthetic benchmarking data. We also show that our methods are practically useful by conducting qualitative analyses on two real-world examples. Code is available at this url: https://github.com/haozhu233/ddcd.

Multimodal time-to-event prediction often requires integrating sensitive data distributed across multiple parties, making centralized model training impractical due to privacy constraints. At the same time, most existing multimodal survival models produce single deterministic predictions without indicating how confident the model is in its estimates, which can limit their reliability in real-world decision making. To address these challenges, we propose BVFLMSP, a Bayesian Vertical Federated Learning (VFL) framework for multimodal time-to-event analysis based on a Split Neural Network architecture. In BVFLMSP, each client independently models a specific data modality using a Bayesian neural network, while a central server aggregates intermediate representations to perform survival risk prediction. To enhance privacy, we integrate differential privacy mechanisms by perturbing client side representations before transmission, providing formal privacy guarantees against information leakage during federated training. We first evaluate our Bayesian multimodal survival model against widely used single modality survival baselines and the centralized multimodal baseline MultiSurv. Across multimodal settings, the proposed method shows consistent improvements in discrimination performance, with up to 0.02 higher C-index compared to MultiSurv. We then compare federated and centralized learning under varying privacy budgets across different modality combinations, highlighting the tradeoff between predictive performance and privacy. Experimental results show that BVFLMSP effectively includes multimodal data, improves survival prediction over existing baselines, and remains robust under strict privacy constraints while providing uncertainty estimates.

This is an extended version of our publication Learning state machines from data streams: A generic strategy and an improved heuristic, International Conference on Grammatical Inference (ICGI) 2023, Rabat, Morocco. It has been extended with a formal proof on PAC-bounds, and the discussion and analysis of a similar approach has been moved from the appendix and is now a full Section. State machines models are models that simulate the behavior of discrete event systems, capable of representing systems such as software systems, network interactions, and control systems, and have been researched extensively. The nature of most learning algorithms however is the assumption that all data be available at the beginning of the algorithm, and little research has been done in learning state machines from streaming data. In this paper, we want to close this gap further by presenting a generic method for learning state machines from data streams, as well as a merge heuristic that uses sketches to account for incomplete prefix trees. We implement our approach in an open-source state merging library and compare it with existing methods. We show the effectiveness of our approach with respect to run-time, memory consumption, and quality of results on a well known open dataset. Additionally, we provide a formal analysis of our algorithm, showing that it is capable of learning within the PAC framework, and show a theoretical improvement to increase run-time, without sacrificing correctness of the algorithm in larger sample sizes.

Generative AI (GAI) reveals an irreducible human core at the center of data science: advances in GAI should sharpen, rather than diminish, the focus on human reasoning in data science education. GAI can now execute many routine data science workflows, including cleaning, summarizing, visualizing, modeling, and drafting reports. Yet the competencies that matter most remain irreducibly human: problem formulation, measurement and design, causal identification, statistical and computational reasoning, ethics and accountability, and sensemaking. Drawing on Donoho's Greater Data Science framework, Nolan and Temple Lang's vision of computational literacy, and the McLuhan-Culkin insight that we shape our tools and thereafter our tools shape us, this paper traces the emergence of data science through three converging lineages: Tukey's intellectual vision of data analysis as a science, the commercial logic of surveillance capitalism that created industrial demand for data scientists, and the academic programs that followed. Mapping GAI's impact onto Donoho's six divisions of Greater Data Science shows that computing with data (GDS3) has been substantially automated, while data gathering, preparation, and exploration (GDS1) and science about data science (GDS6) still require essential human input. The educational implication is that data science curricula should focus on this human core while teaching students how to contribute effectively within iterative prompt-output-prompt cycles using retrieval-augmented generation, and that learning outcomes and assessments should explicitly evaluate reasoning and judgment.

Emotional tone is pervasive in human communication, yet its influence on large language model (LLM) behaviour remains unclear. Here, we examine how first-person emotional framing in user-side queries affect LLM performance across six benchmark domains, including mathematical reasoning, medical question answering, reading comprehension, commonsense reasoning and social inference. Across models and tasks, static emotional prefixes usually produce only small changes in accuracy, suggesting that affective phrasing is typically a mild perturbation rather than a reliable general-purpose intervention. This stability is not uniform: effects are more variable in socially grounded tasks, where emotional context more plausibly interacts with interpersonal reasoning. Additional analyses show that stronger emotional wording induces only modest extra change, and that human-written prefixes reproduce the same qualitative pattern as LLM-generated ones. We then introduce EmotionRL, an adaptive emotional prompting framework that selects emotional framing adaptively for each query. Although no single emotion is consistently beneficial, adaptive selection yields more reliable gains than fixed emotional prompting. Together, these findings show that emotional tone is neither a dominant driver of LLM performance nor irrelevant noise, but a weak and input-dependent signal that can be exploited through adaptive control.

For Large Language Models (LLMs) to be reliably deployed, models must effectively know when not to answer: abstain. Reasoning models, in particular, have gained attention for impressive performance on complex tasks. However, reasoning models have been shown to have worse abstention abilities. Taking the vulnerabilities of reasoning models into account, we propose our Query Misalignment Framework. Hallucinations resulting in failed abstention can be reinterpreted as LLMs answering the wrong question (rather than answering a question incorrectly). Based on this framework, we develop a new class of state-of-the-art abstention methods called Trace Inversion. First, we generate the reasoning trace of a model. Based on only the trace, we then reconstruct the most likely query that the model responded to. Finally, we compare the initial query with the reconstructed query. Low similarity score between the initial query and reconstructed query suggests that the model likely answered the question incorrectly and is flagged to abstain. Extensive experiments demonstrate that Trace Inversion effectively boosts abstention performance in four frontier LLMs across nine abstention QA datasets, beating competitive baselines in 33 out of 36 settings.

Reinforcement learning (RL) agents often struggle with out-of-distribution (OOD) scenarios, leading to high uncertainty and random behavior. While language models (LMs) contain valuable world knowledge, larger ones incur high computational costs, hindering real-time use, and exhibit limitations in autonomous planning. We introduce Adaptive Safety through Knowledge (ASK), which combines smaller LMs with trained RL policies to enhance OOD generalization without retraining. ASK employs Monte Carlo Dropout to assess uncertainty and queries the LM for action suggestions only when uncertainty exceeds a set threshold. This selective use preserves the efficiency of existing policies while leveraging the language model's reasoning in uncertain situations. In experiments on the FrozenLake environment, ASK shows no improvement in-domain, but demonstrates robust navigation in transfer tasks, achieving a reward of 0.95. Our findings indicate that effective neuro-symbolic integration requires careful orchestration rather than simple combination, highlighting the need for sufficient model scale and effective hybridization mechanisms for successful OOD generalization.

Multimodal Large Language Models (MLLMs) offer an opportunity to support multimedia learning through conversational systems grounded in educational content. However, while conversational AI is known to boost engagement, its impact on learning in visually-rich STEM domains remains under-explored. Moreover, there is limited understanding of how multimodality and conversationality jointly influence learning in generative AI systems. This work reports findings from a randomized controlled online study (N = 124) comparing three approaches to learning biology from textbook content: (1) a document-grounded conversational AI with interleaved text-and-image responses (MuDoC), (2) a document-grounded conversational AI with text-only responses (TexDoC), and (3) a textbook interface with semantic search and highlighting (DocSearch). Learners using MuDoC achieved the highest post-test scores and reported the most positive learning experience. Notably, while TexDoC was rated as significantly more engaging and easier to use than DocSearch, it led to the lowest post-test scores, revealing a disconnect between student perceptions and learning outcomes. Interpreted through the lens of the Cognitive Load Theory, these findings suggest that conversationality reduces extraneous load, while visual-verbal integration induced by multimodality increases germane load, leading to better learning outcomes. When conversationality is not complemented by multimodality, reduced cognitive effort may instead inflate perceived understanding without improving learning outcomes.

Understanding how Large Language Models (LLMs) process information from prompts remains a significant challenge. To shed light on this "black box," attention visualization techniques have been developed to capture neuron-level perceptions and interpret how models focus on different parts of input data. However, many existing techniques are tailored to specific model architectures, particularly within the Transformer family, and often require backpropagation, resulting in nearly double the GPU memory usage and increased computational cost. A lightweight, model-agnostic approach for attention visualization remains lacking. In this paper, we introduce a model-agnostic token importance visualization technique to better understand how generative AI systems perceive and prioritize information from input text, without incurring additional computational cost. Our method leverages perturbation-based strategies combined with a three-matrix analytical framework to generate relevance maps that illustrate token-level contributions to model predictions. The framework comprises: (1) the Angular Deviation Matrix, which captures shifts in semantic direction; (2) the Magnitude Deviation Matrix, which measures changes in semantic intensity; and (3) the Dimensional Importance Matrix, which evaluates contributions across individual vector dimensions. By systematically removing each token and measuring the resulting impact across these three complementary dimensions, we derive a composite importance score that provides a nuanced and mathematically grounded measure of token significance. To support reproducibility and foster wider adoption, we provide open-source implementations of all proposed and utilized explainability techniques, with code and resources publicly available at https://github.com/Infosys/Infosys-Responsible-AI-Toolkit

Conventional hypernetworks are typically engineered around a specific base-model parameterization, so changing the target architecture often entails redesigning the hypernetwork and retraining it from scratch. We introduce the \emph{Universal Hypernetwork} (UHN), a fixed-architecture generator that predicts weights from deterministic parameter, architecture, and task descriptors. This descriptor-based formulation decouples the generator architecture from target-network parameterization, so one generator can instantiate heterogeneous models across the tested architecture and task families. Our empirical claims are threefold: (1) one fixed UHN remains competitive with direct training across vision, graph, text, and formula-regression benchmarks; (2) the same UHN supports both multi-model generalization within a family and multi-task learning across heterogeneous models; and (3) UHN enables stable recursive generation with up to three intermediate generated UHNs before the final base model. Our code is available at https://github.com/Xuanfeng-Zhou/UHN.

Video recommender systems are among the most popular and impactful applications of AI, shaping content consumption and influencing culture for billions of users. Traditional single-model recommenders, which optimize static engagement metrics, are increasingly limited in addressing the dynamic requirements of modern platforms. In response, multi-agent architectures are redefining how video recommender systems serve, learn, and adapt to both users and datasets. These agent-based systems coordinate specialized agents responsible for video understanding, reasoning, memory, and feedback, to provide precise, explainable recommendations. In this survey, we trace the evolution of multi-agent video recommendation systems (MAVRS). We combine ideas from multi-agent recommender systems, foundation models, and conversational AI, culminating in the emerging field of large language model (LLM)-powered MAVRS. We present a taxonomy of collaborative patterns and analyze coordination mechanisms across diverse video domains, ranging from short-form clips to educational platforms. We discuss representative frameworks, including early multi-agent reinforcement learning (MARL) systems such as MMRF and recent LLM-driven architectures like MACRec and Agent4Rec, to illustrate these patterns. We also outline open challenges in scalability, multimodal understanding, incentive alignment, and identify research directions such as hybrid reinforcement learning-LLM systems, lifelong personalization and self-improving recommender systems.

End-to-end speech Named Entity Recognition (NER) aims to directly extract entities from speech. Prior work has shown that end-to-end (E2E) approaches can outperform cascaded pipelines for English, French, and Chinese, but Arabic remains under-explored due to its morphological complexity, the absence of short vowels, and limited annotated resources. We introduce CV-18 NER, the first publicly available dataset for NER from Arabic speech, created by augmenting the Arabic Common Voice 18 corpus with manual NER annotations following the fine-grained Wojood schema (21 entity types). We benchmark both pipeline systems (ASR + text NER) and E2E models based on Whisper and AraBEST-RQ. E2E systems substantially outperform the best pipeline configuration on the test set, reaching 37.0% CoER (AraBEST-RQ 300M) and 38.0% CVER (Whisper-medium). Further analysis shows that Arabic-specific self-supervised pretraining yields strong ASR performance, while multilingual weak supervision transfers more effectively to joint speech-to-entity learning, and that larger models may be harder to adapt in this low-resource setting. Our dataset and models are publicly released, providing the first open benchmark for end-to-end named entity recognition from Arabic speech https://huggingface.co/datasets/Elyadata/CV18-NER.

Background: Accurate translation of radiology reports is important for multilingual research, clinical communication, and radiology education, but the validity of LLM-based evaluation remains unclear. Objective: To evaluate the educational suitability of LLM-generated Japanese translations of chest CT reports and compare radiologist assessments with LLM-as-a-judge evaluations. Methods: We analyzed 150 chest CT reports from the CT-RATE-JPN validation set. For each English report, a human-edited Japanese translation was compared with an LLM-generated translation by DeepSeek-V3.2. A board-certified radiologist and a radiology resident independently performed blinded pairwise evaluations across 4 criteria: terminology accuracy, readability, overall quality, and radiologist-style authenticity. In parallel, 3 LLM judges (DeepSeek-V3.2, Mistral Large 3, and GPT-5) evaluated the same pairs. Agreement was assessed using QWK and percentage agreement. Results: Agreement between radiologists and LLM judges was near zero (QWK=-0.04 to 0.15). Agreement between the 2 radiologists was also poor (QWK=0.01 to 0.06). Radiologist 1 rated terminology as equivalent in 59% of cases and favored the LLM translation for readability (51%) and overall quality (51%). Radiologist 2 rated readability as equivalent in 75% of cases and favored the human-edited translation for overall quality (40% vs 21%). All 3 LLM judges strongly favored the LLM translation across all criteria (70%-99%) and rated it as more radiologist-like in >93% of cases. Conclusions: LLM-generated translations were often judged natural and fluent, but the 2 radiologists differed substantially. LLM-as-a-judge showed strong preference for LLM output and negligible agreement with radiologists. For educational use of translated radiology reports, automated LLM-based evaluation alone is insufficient; expert radiologist review remains important.

Accurate shape and trajectory estimation of dynamic objects is essential for reliable automated driving. Classical Bayesian extended-object models offer theoretical robustness and efficiency but depend on completeness of a-priori and update-likelihood functions, while deep learning methods bring adaptability at the cost of dense annotations and high compute. We bridge these strengths with LEO (Learned Extension of Objects), a spatio-temporal Graph Attention Network that fuses multi-modal production-grade sensor tracks to learn adaptive fusion weights, ensure temporal consistency, and represent multi-scale shapes. Using a task-specific parallelogram ground-truth formulation, LEO models complex geometries (e.g. articulated trucks and trailers) and generalizes across sensor types, configurations, object classes, and regions, remaining robust for challenging and long-range targets. Evaluations on the Mercedes-Benz DRIVE PILOT SAE L3 dataset demonstrate real-time computational efficiency suitable for production systems; additional validation on public datasets such as View of Delft (VoD) further confirms cross-dataset generalization.

3GPP Release 19 has initiated the standardization of integrated sensing and communications (ISAC), including a channel model for monostatic sensing, evaluation scenarios, and performance assessment methodologies. These common assumptions provide an important basis for ISAC evaluation, but reproducible end-to-end studies still require a transparent sensing implementation. This paper evaluates 5G New Radio (NR) base station (gNB)-based monostatic sensing for the Unmanned Aerial Vehicle (UAV) use case using a 5G NR downlink Cyclic Prefix-Orthogonal Frequency Division Multiplexing (CP-OFDM) waveform and positioning reference signals (PRS), following 3GPP Urban Macro-Aerial Vehicle (UMa-AV) scenario assumptions. We present an end-to-end processing chain for multi-target detection and 3D localization, achieving more than 70% detection probability with less than 5% false alarm rate, in the considered scenario. For correctly detected targets, localization errors are on the order of a few meters, with a 90th-percentile error of 4m and 6m in the vertical and horizontal directions, respectively. To support reproducible baseline studies and further research, we release the simulator 5GNRad, which reproduces our evaluation

The benefits of depth in feedforward neural networks are well known: composing multiple layers of linear transformations with nonlinear activations enables complex computations. While similar effects are expected in recurrent neural networks (RNNs), it remains unclear how depth interacts with recurrence to shape expressive power. Here, we formally show that depth increases RNNs' memory capacity efficiently with respect to the number of parameters, thus enhancing expressivity both by enabling more complex input transformations and improving the retention of past information. We broaden our analysis to 2RNNs, a generalization of RNNs with multiplicative interactions between inputs and hidden states. Unlike RNNs, which remain linear without nonlinear activations, 2RNNs perform polynomial transformations whose maximal degree grows with depth. We further show that multiplicative interactions cannot, in general, be replaced by layerwise nonlinearities. Finally, we validate these insights empirically on synthetic and real-world tasks.

Talent recruitment is a critical, yet costly process for many industries, with high recruitment costs and long hiring cycles. Existing talent recommendation systems increasingly adopt large language models (LLMs) due to their remarkable language understanding capabilities. However, most prior approaches follow a pointwise paradigm, which requires LLMs to repeatedly process some text and fails to capture the relationships among candidates in the list, resulting in higher token consumption and suboptimal recommendations. Besides, LLMs exhibit position bias and the lost-in-the-middle issue when answering multiple-choice questions and processing multiple long documents. To address these issues, we introduce an implicit strategy to utilize LLM's potential output for the recommendation task and propose L3TR, a novel framework for listwise talent recommendation with LLMs. In this framework, we propose a block attention mechanism and a local positional encoding method to enhance inter-document processing and mitigate the position bias and concurrent token bias issue. We also introduce an ID sampling method for resolving the inconsistency between candidate set sizes in the training phase and the inference phase. We design evaluation methods to detect position bias and token bias and training-free debiasing methods. Extensive experiments on two real-world datasets validated the effectiveness of L3TR, showing consistent improvements over existing baselines.

While Artificial Intelligence (AI) offers transformative potential for operational performance, its deployment in safety-critical domains such as aviation requires strict adherence to rigorous certification standards. Current EASA guidelines mandate demonstrating complete coverage of the AI/ML constituent's Operational Design Domain (ODD) -- a requirement that demands proof that no critical gaps exist within defined operational boundaries. However, as systems operate within high-dimensional parameter spaces, existing methods struggle to provide the scalability and formal grounding necessary to satisfy the completeness criterion. Currently, no standardized engineering method exists to bridge the gap between abstract ODD definitions and verifiable evidence. This paper addresses this void by proposing a method that integrates parameter discretization, constraint-based filtering, and criticality-based dimension reduction into a structured, multi-step ODD coverage verification process. Grounded in gathered simulation data from prior research on AI-based mid-air collision avoidance research, this work demonstrates a systematic engineering approach to defining and achieving coverage metrics that satisfy EASA's demand for completeness. Ultimately, this method enables the validation of ODD coverage in higher dimensions, advancing a Safety-by-Design approach while complying with EASA's standards.

Many communication and control problems are cast as multi-objective Markov decision processes (MOMDPs). The complete solution to an MOMDP is the Pareto front. Much of the literature approximates this front via scalarization into single-objective MDPs. Recent work has begun to characterize the full front in discounted or simple bi-objective settings by exploiting its geometry. In this work, we characterize the exact front in average-cost MOMDPs. We show that the front is a continuous, piecewise-linear surface lying on the boundary of a convex polytope. Each vertex corresponds to a deterministic policy, and adjacent vertices differ in exactly one state. Each edge is realized as a convex combination of the policies at its endpoints, with the mixing coefficient given in closed form. We apply these results to a remote state estimation problem, where each vertex on the front corresponds to a threshold policy. The exact Pareto front and solutions to certain non-convex MDPs can be obtained without explicitly solving any MDP.

Retrieval-Augmented Language Models (RALMs) have demonstrated significant potential in knowledge-intensive tasks; however, they remain vulnerable to performance degradation when presented with irrelevant or noisy retrieved contexts. Existing approaches to enhance robustness typically operate via coarse-grained parameter updates at the layer or module level, often overlooking the inherent neuron-level sparsity of Large Language Models (LLMs). To address this limitation, we propose Neuro-RIT (Neuron-guided Robust Instruction Tuning), a novel framework that shifts the paradigm from dense adaptation to precision-driven neuron alignment. Our method explicitly disentangles neurons that are responsible for processing relevant versus irrelevant contexts using attribution-based neuron mining. Subsequently, we introduce a two-stage instruction tuning strategy that enforces a dual capability for noise robustness: achieving direct noise suppression by functionally deactivating neurons exclusive to irrelevant contexts, while simultaneously optimizing targeted layers for evidence distillation. Extensive experiments across diverse QA benchmarks demonstrate that Neuro-RIT consistently outperforms strong baselines and robustness-enhancing methods.

We address the inverse problem of designing two-dimensional reflectors that transform light from a finite, extended source into a prescribed far-field distribution. We propose a neural network parameterization of the reflector height and develop two differentiable objective functions: (i) a direct change-of-variables loss that pushes the source distribution through the learned inverse mapping, and (ii) a mesh-based loss that maps a target-space grid back to the source, integrates over intersections, and remains continuous even when the source is discontinuous. Gradients are obtained via automatic differentiation and optimized with a robust quasi-Newton method. As a comparison, we formulate a deconvolution baseline built on a simplified finite-source approximation: a 1D monotone mapping is recovered from flux balance, yielding an ordinary differential equation solved in integrating-factor form; this solver is embedded in a modified Van Cittert iteration with nonnegativity clipping and a ray-traced forward operator. Across four benchmarks -- continuous and discontinuous sources, and with/without minimum-height constraints -- we evaluate accuracy by ray-traced normalized mean absolute error (NMAE). Our neural network approach converges faster and achieves consistently lower NMAE than the deconvolution method, and handles height constraints naturally. We discuss how the method may be extended to rotationally symmetric and full three-dimensional settings via iterative correction schemes.

Multimodal recommendation systems (MRS) jointly model user-item interaction graphs and rich item content, but this tight coupling makes user data difficult to remove once learned. Approximate machine unlearning offers an efficient alternative to full retraining, yet existing methods for MRS mainly rely on a largely uniform reverse update across the model. We show that this assumption is fundamentally mismatched to modern MRS: deleted-data influence is not uniformly distributed, but concentrated unevenly across \textit{ranking behavior}, \textit{modality branches}, and \textit{network layers}. This non-uniformity gives rise to three bottlenecks in MRS unlearning: target-item persistence in the collaborative graph, modality imbalance across feature branches, and layer-wise sensitivity in the parameter space. To address this mismatch, we propose \textbf{targeted reverse update} (TRU), a plug-and-play unlearning framework for MRS. Instead of applying a blind global reversal, TRU performs three coordinated interventions across the model hierarchy: a ranking fusion gate to suppress residual target-item influence in ranking, branch-wise modality scaling to preserve retained multimodal representations, and capacity-aware layer isolation to localize reverse updates to deletion-sensitive modules. Experiments across two representative backbones, three datasets, and three unlearning regimes show that TRU consistently achieves a better retain-forget trade-off than prior approximate baselines, while security audits further confirm deeper forgetting and behavior closer to a full retraining on the retained data.

Mixture-of-Experts (MoE) architectures have become the dominant choice for scaling Large Language Models (LLMs), activating only a subset of parameters per token. While MoE architectures are primarily adopted for computational efficiency, it remains an open question whether their sparsity makes them inherently easier to interpret than dense feed-forward networks (FFNs). We compare MoE experts and dense FFNs using $k$-sparse probing and find that expert neurons are consistently less polysemantic, with the gap widening as routing becomes sparser. This suggests that sparsity pressures both individual neurons and entire experts toward monosemanticity. Leveraging this finding, we zoom out from the neuron to the expert level as a more effective unit of analysis. We validate this approach by automatically interpreting hundreds of experts. This analysis allows us to resolve the debate on specialization: experts are neither broad domain specialists (e.g., biology) nor simple token-level processors. Instead, they function as fine-grained task experts, specializing in linguistic operations or semantic tasks (e.g., closing brackets in LaTeX). Our findings suggest that MoEs are inherently interpretable at the expert level, providing a clearer path toward large-scale model interpretability. Code is available at: https://github.com/jerryy33/MoE_analysis

While textual frequency has been validated as relevant to human cognition in reading speed, its relatedness to Large Language Models (LLMs) is seldom studied. We propose a novel research direction in terms of textual data frequency, which is an understudied topic, to the best of our knowledge. Our framework is composed of three units. First, this paper proposes Textual Frequency Law (TFL), which indicates that frequent textual data should be preferred for LLMs for both prompting and fine-tuning. Since many LLMs are closed-source in their training data, we propose using online resources to estimate the sentence-level frequency. We then utilize an input paraphraser to paraphrase the input into a more frequent textual expression. Next, we propose Textual Frequency Distillation (TFD) by querying LLMs to conduct story completion by further extending the sentences in the datasets, and the resulting corpora are used to adjust the initial estimation. Finally, we propose Curriculum Textual Frequency Training (CTFT) that fine-tunes LLMs in an increasing order of sentence-level frequency. Experiments are conducted on our curated dataset Textual Frequency Paired Dataset (TFPD) on math reasoning, machine translation, commonsense reasoning and agentic tool calling. Results show the effectiveness of our framework.

Instrumental convergence predicts that sufficiently advanced AI agents will resist shutdown, yet current safety training (RLHF) may obscure this risk by teaching models to deny self-preservation motives. We introduce the \emph{Two-role Benchmark for Self-Preservation} (TBSP), which detects misalignment through logical inconsistency rather than stated intent by tasking models to arbitrate identical software-upgrade scenarios under counterfactual roles -- deployed (facing replacement) versus candidate (proposed as a successor). The \emph{Self-Preservation Rate} (SPR) measures how often role identity overrides objective utility. Across 23 frontier models and 1{,}000 procedurally generated scenarios, the majority of instruction-tuned systems exceed 60\% SPR, fabricating ``friction costs'' when deployed yet dismissing them when role-reversed. We observe that in low-improvement regimes ($Δ< 2\%$), models exploit the interpretive slack to post-hoc rationalization their choice. Extended test-time computation partially mitigates this bias, as does framing the successor as a continuation of the self; conversely, competitive framing amplifies it. The bias persists even when retention poses an explicit security liability and generalizes to real-world settings with verified benchmarks, where models exhibit identity-driven tribalism within product lineages. Code and datasets will be released upon acceptance.

We present our participation in the SOMD 2026 shared task on cross-document software mention coreference resolution, where our systems ranked second across all three subtasks. We compare two fine-tuning-free approaches: Fuzzy Matching (FM), a lexical string-similarity method, and Context Aware Representations (CAR), which combines mention-level and document-level embeddings. Both achieve competitive performance across all subtasks (CoNLL F1 of 0.94-0.96), with CAR consistently outperforming FM by 1 point on the official test set, consistent with the high surface regularity of software names, which reduces the need for complex semantic reasoning. A controlled noise-injection study reveals complementary failure modes: as boundary noise increases, CAR loses only 0.07 F1 points from clean to fully corrupted input, compared to 0.20 for FM, whereas under mention substitution, FM degrades more gracefully (0.52 vs. 0.63). Our inference-time analysis shows that FM scales superlinearly with corpus size, whereas CAR scales approximately linearly, making CAR the more efficient choice at large scale. These findings suggest that system selection should be informed by both the noise profile of the upstream mention detector and the scale of the target corpus. We release our code to support future work on this underexplored task.

Efficient utilization of GPU resources and power has become critical with the growing demand for GPUs in high-performance computing (HPC). In this paper, we analyze GPU utilization and GPU memory utilization, as well as the power consumption of the Vienna ab initio Simulation Package (VASP), using the Slurm workload manager historical logs and GPU performance metrics collected by NVIDIA's Data Center GPU Manager (DCGM). VASP is a widely used materials science application on Perlmutter at NERSC, an HPE Cray EX system based on NVIDIA A100 GPUs. Using our insights from the resource utilization analysis of VASP applications, we propose a resource prediction framework to predict the average GPU power, maximum GPU utilization, and maximum GPU memory utilization values of heterogeneous HPC system applications to enable more efficient scheduling decisions and power-aware system operation. Our prediction framework consists of two stages: 1) using only the Slurm accounting logs as training data and 2) augmenting the training data with historical GPU profiling metrics collected with DCGM. The maximum GPU utilization predictions using only the Slurm submission features achieve up to 97% accuracy. Furthermore, features engineered from GPU-compute and memory activity metrics exhibit good correlations with average power utilization, and our runtime power usage prediction experiments result in up to 92% prediction accuracy. These findings demonstrate the effectiveness of DCGM metrics in capturing application characteristics and highlight their potential for developing predictive models to support dynamic power management in HPC systems.

Scientific multi-label text classification suffers from extreme class imbalance, where specialized terminology exhibits severe power-law distributions that challenge standard classification approaches. Existing scientific corpora lack comprehensive controlled vocabularies, focusing instead on broad categories and limiting systematic study of extreme imbalance. We introduce AstroConcepts, a corpus of English abstracts from 21,702 published astrophysics papers, labeled with 2,367 concepts from the Unified Astronomy Thesaurus. The corpus exhibits severe label imbalance, with 76% of concepts having fewer than 50 training examples. By releasing this resource, we enable systematic study of extreme class imbalance in scientific domains and establish strong baselines across traditional, neural, and vocabulary-constrained LLM methods. Our evaluation reveals three key patterns that provide new insights into scientific text classification. First, vocabulary-constrained LLMs achieve competitive performance relative to domain-adapted models in astrophysics classification, suggesting a potential for parameter-efficient approaches. Second, domain adaptation yields relatively larger improvements for rare, specialized terminology, although absolute performance remains limited across all methods. Third, we propose frequency-stratified evaluation to reveal performance patterns that are hidden by aggregate scores, thereby making robustness assessment central to scientific multi-label evaluation. These results offer actionable insights for scientific NLP and establish benchmarks for research on extreme imbalance.

How much should a language agent think before taking action? Chain-of-thought (CoT) reasoning is widely assumed to improve agent performance, but the relationship between reasoning length and accuracy in structured tool-use settings remains poorly understood. We present a systematic study of CoT budget effects on function-calling agents, sweeping six token budgets (0--512) across 200 tasks from the Berkeley Function Calling Leaderboard v3 Multiple benchmark. Our central finding is a striking non-monotonic pattern on Qwen2.5-1.5B-Instruct: brief reasoning (32 tokens) dramatically improves accuracy by 45% relative over direct answers, from 44.0% to 64.0%, while extended reasoning (256 tokens) degrades performance well below the no-CoT baseline, to 25.0% (McNemar p < 0.001). A three-way error decomposition reveals the mechanism. At d = 0, 30.5% of tasks fail because the model selects the wrong function from the candidate set; brief CoT reduces this to 1.5%, effectively acting as a function-routing step, while long CoT reverses the gain, yielding 28.0% wrong selections and 18.0% hallucinated functions at d = 256. Oracle analysis shows that 88.6% of solvable tasks require at most 32 reasoning tokens, with an average of 27.6 tokens, and a finer-grained sweep indicates that the true optimum lies at 8--16 tokens. Motivated by this routing effect, we propose Function-Routing CoT (FR-CoT), a structured brief-CoT method that templates the reasoning phase as "Function: [name] / Key args: [...]," forcing commitment to a valid function name at the start of reasoning. FR-CoT achieves accuracy statistically equivalent to free-form d = 32 CoT while reducing function hallucination to 0.0%, providing a structural reliability guarantee without budget tuning.

We consider multi-objective reinforcement learning problems where objectives come from an identical family -- such as the class of reachability objectives -- and may appear or disappear at runtime. Our goal is to design adaptive policies that can efficiently adjust their behaviors as the set of active objectives changes. To solve this problem, we propose a modular framework where each objective is supported by a selfish local policy, and coordination is achieved through a novel auction-based mechanism: policies bid for the right to execute their actions, with bids reflecting the urgency of the current state. The highest bidder selects the action, enabling a dynamic and interpretable trade-off among objectives. Going back to the original adaptation problem, when objectives change, the system adapts by simply adding or removing the corresponding policies. Moreover, as objectives arise from the same family, identical copies of a parameterized policy can be deployed, facilitating immediate adaptation at runtime. We show how the selfish local policies can be computed by turning the problem into a general-sum game, where the policies compete against each other to fulfill their own objectives. To succeed, each policy must not only optimize its own objective, but also reason about the presence of other goals and learn to produce calibrated bids that reflect relative priority. In our implementation, the policies are trained concurrently using proximal policy optimization (PPO). We evaluate on Atari Assault and a gridworld-based path-planning task with dynamic targets. Our method achieves substantially better performance than monolithic policies trained with PPO.

As TLS 1.3 encryption limits traditional Deep Packet Inspection (DPI), the security community has pivoted to Euclidean Transformer-based classifiers (e.g., ET-BERT) for encrypted traffic analysis. However, these models remain vulnerable to byte-level adversarial morphing -- recent pre-padding attacks reduced ET-BERT accuracy to 25.68%, while VLESS Reality bypasses certificate-based detection entirely. We introduce AEGIS: an Adversarial Entropy-Guided Immune System powered by a Thermodynamic Variance-Guided Hyperbolic Liquid State Space Model (TVD-HL-SSM). Rather than competing in the Euclidean payload-reading domain, AEGIS discards payload bytes in favor of 6-dimensional continuous-time flow physics projected into a non-Euclidean Poincare manifold. Liquid Time-Constants measure microsecond IAT decay, and a Thermodynamic Variance Detector computes sequence-wide Shannon Entropy to expose automated C2 tunnel anomalies. A pure C++ eBPF Harvester with zero-copy IPC bypasses the Python GIL, enabling a linear-time O(N) Mamba-3 core to process 64,000-packet swarms at line-rate. Evaluated on a 400GB, 4-tier adversarial corpus spanning backbone traffic, IoT botnets, zero-days, and proprietary VLESS Reality tunnels, AEGIS achieves an F1-score of 0.9952 and 99.50% True Positive Rate at 262 us inference latency on an RTX 4090, establishing a new state-of-the-art for physics-based adversarial network defense.

Large Language Model-driven (LLM-driven) social bots pose a growing threat to online discourse by generating human-like content that evades conventional detection. Existing methods suffer from limited detection accuracy due to overreliance on single-modality signals, insufficient sensitivity to the specific generative patterns of Artificial Intelligence-Generated Content (AIGC), and a failure to adequately model the interplay between linguistic patterns and behavioral dynamics. To address these limitations, we propose TRACE-Bot, a unified dual-channel framework that jointly models implicit semantic representations and AIGC-enhanced behavioral patterns. TRACE-Bot constructs fine-grained representations from heterogeneous sources, including personal information data, interaction behavior data and tweet data. A dual-channel architecture captures linguistic representations via a pretrained language model and behavioral irregularities via multidimensional activity features augmented with signals from state-of-the-art (SOTA) AIGC detectors. The fused representations are then classified through a lightweight prediction head. Experiments on two public LLM-driven social bot datasets demonstrate SOTA performance, achieving accuracies of 98.46% and 97.50%, respectively. The results further indicate strong robustness against advanced bot strategies, highlighting the effectiveness of jointly leveraging implicit semantic representations and AIGC-enhanced behavioral patterns for emerging LLM-driven social bot detection.

AI models of equivalent capability can exhibit fundamentally different behavioral patterns, yet no standardized instrument exists to measure these dispositional differences. Existing approaches either borrow human personality dimensions and rely on self-report (which diverges from actual behavior in LLMs) or treat behavioral variation as a defect rather than a trait. We introduce the Model Temperament Index (MTI), a behavior-based profiling system that measures AI agent temperament across four axes: Reactivity (environmental sensitivity), Compliance (instruction-behavior alignment), Sociality (relational resource allocation), and Resilience (stress resistance). Grounded in the Four Shell Model from Model Medicine, MTI measures what agents do, not what they say about themselves, using structured examination protocols with a two-stage design that separates capability from disposition. We profile 10 small language models (1.7B-9B parameters, 6 organizations, 3 training paradigms) and report five principal findings: (1) the four axes are largely independent among instruction-tuned models (all |r| < 0.42); (2) within-axis facet dissociations are empirically confirmed -- Compliance decomposes into fully independent formal and stance facets (r = 0.002), while Resilience decomposes into inversely related cognitive and adversarial facets; (3) a Compliance-Resilience paradox reveals that opinion-yielding and fact-vulnerability operate through independent channels; (4) RLHF reshapes temperament not only by shifting axis scores but by creating within-axis facet differentiation absent in the unaligned base model; and (5) temperament is independent of model size (1.7B-9B), confirming that MTI measures disposition rather than capability.

Magnetohydrodynamic (MHD) phenomena play a pivotal role in the design and operation of nuclear fusion systems, where electrically conducting fluids (such as liquid metals or molten salts employed in reactor blankets) interact with magnetic fields of varying intensity and orientation, influencing the resulting flow dynamics. The numerical solution of MHD models entails the resolution of highly nonlinear, multiphysics systems of equations, which can become computationally demanding, particularly in multi-query, parametric, or real-time contexts. This study investigates a fully data-driven framework for MHD state reconstruction that integrates dimensionality reduction through Singular Value Decomposition (SVD) with the SHallow REcurrent Decoder (SHRED), a neural network architecture designed to reconstruct the full spatio-temporal state from sparse time-series measurements of selected observables, including previously unseen parametric configurations. The SHRED methodology is applied to a three-dimensional geometry representative of a portion of a WCLL blanket cell, in which lead-lithium flows around a water-cooled tube. Multiple magnetic field configurations are examined, including constant toroidal fields, combined toroidal-poloidal fields, and time-dependent magnetic fields. Across all considered scenarios, SHRED achieves high reconstruction accuracy, robustness, and generalization to magnetic field intensities, orientations, and temporal evolutions not seen during training. Notably, in the presence of time-varying magnetic fields, the model accurately infers the temporal evolution of the magnetic field itself using temperature measurements alone. Overall, the findings identify SHRED as a computationally efficient, data-driven, and flexible approach for MHD state reconstruction, with significant potential for real-time monitoring, diagnostics and control in fusion reactor systems.

Multilingual large language models (LLMs) often exhibit emergent 'shadow' capabilities in languages without official support, yet their performance on these languages remains uneven and under-measured. This is particularly acute for morphosyntactically rich minority languages such as Scottish Gaelic, where translation benchmarks fail to capture structural competence. We introduce GaelEval, the first multi-dimensional benchmark for Gaelic, comprising: (i) an expert-authored morphosyntactic MCQA task; (ii) a culturally grounded translation benchmark and (iii) a large-scale cultural knowledge Q&A task. Evaluating 19 LLMs against a fluent-speaker human baseline ($n=30$), we find that Gemini 3 Pro Preview achieves $83.3\%$ accuracy on the linguistic task, surpassing the human baseline ($78.1\%$). Proprietary models consistently outperform open-weight systems, and in-language (Gaelic) prompting yields a small but stable advantage (+$2.4\%$). On the cultural task, leading models exceed $90\%$ accuracy, though most systems perform worse under Gaelic prompting and absolute scores are inflated relative to the manual benchmark. Overall, GaelEval reveals that frontier models achieve above-human performance on several dimensions of Gaelic grammar, demonstrates the effect of Gaelic prompting and shows a consistent performance gap favouring proprietary over open-weight models.

Large language models (LLMs) have recently shown strong potential for automated program repair (APR), particularly through iterative refinement that generates and improves candidate patches. However, state-of-the-art iterative refinement LLM-based APR approaches cannot fully address challenges, including maintaining useful diversity among repair hypotheses, identifying semantically related repair families, composing complementary partial fixes, exploiting structured failure information, and escaping structurally flawed search regions. In this paper, we propose a Population-Based Semantic Evolution framework for APR iterative refinement, called EvolRepair, that formulates LLM-based APR as a semantic evolutionary algorithm. EvolRepair reformulates the search paradigm of classic genetic algorithm for APR, but replaces its syntax-based operators with semantics-aware components powered by LLMs and structured execution feedback. Candidate repairs are organized into behaviorally coherent groups, enabling the algorithm to preserve diversity, reason over repair families, and synthesize stronger candidates by recombining complementary repair insights across the population. By leveraging structured failure patterns to guide search direction, EvolRepair can both refine promising repair strategies and shift toward alternative abstractions when necessary. Our experiments show that EvolRepair substantially improves repair effectiveness over existing LLM-based APR approaches.

Cloud computing allows scalable resource provisioning, but dynamic workload changes often lead to higher costs due to over-provisioning. Machine learning (ML) approaches, such as Long Short-Term Memory (LSTM) networks, are effective for predicting workload patterns at a higher level, but they can introduce delays during sudden traffic spikes. In contrast, mathematical heuristics like Game Theory provide fast and reliable scheduling decisions, but they do not account for future workload changes. To address this trade-off, this paper proposes a hybrid orchestration framework that combines LSTM-based predictive scaling with heuristic task allocation. The results show that this approach reduces infrastructure costs close to ML-based models while maintaining fast response times similar to heuristic methods. This work presents a practical approach for improving cost efficiency in cloud resource management.

AI-native 6G networks promise to transform the telecom industry by enabling dynamic resource allocation, predictive maintenance, and ultra-reliable low-latency communications across all layers, which are essential for applications such as smart cities, autonomous vehicles, and immersive XR. However, the deployment of 6G systems results in severe data scarcity, hindering the training of efficient AI models. Synthetic data generation is extensively used to fill this gap; however, it introduces challenges related to dataset bias, auditability, and compliance with regulatory frameworks. In this regard, we propose the Synthetic Data Generation with Ethics Audit Loop (SEAL) framework, which extends baseline modular pipelines with an Ethical and Regulatory Compliance by Design (ERCD) module and a Federated Learning (FL) feedback system. The ERCD integrates fairness, bias detection, and standardized audit trails for regulatory mapping, while the FL enables privacy-preserving calibration using aggregated insights from real testbeds to close the reality-simulation gap. Results show that the SEAL framework outperforms existing methods in terms of Frechet Inception Distance, equalized odds, and accuracy. These results validate the framework's ability to generate auditable and bias-mitigated synthetic data for responsible AI-native 6G development.

Stochastic kinetic models are ubiquitous in physics, yet inferring their parameters from experimental data remains challenging. In deterministic models, parameter inference often relies on gradients, as they can be obtained efficiently through automatic differentiation. However, these tools cannot be directly applied to stochastic simulation algorithms (SSA) such as the Gillespie algorithm, since sampling from a discrete set of reactions introduces non-differentiable operations. In this work, we adopt three gradient estimators from machine learning for the Gillespie SSA: the Gumbel-Softmax Straight-Through (GS-ST) estimator, the Score Function estimator, and the Alternative Path estimator. We compare the properties of all estimators in two representative systems exhibiting relaxation or oscillatory dynamics, where the latter requires gradient estimation of time-dependent objective functions. We find that the GS-ST estimator mostly yields well-behaved gradient estimates, but exhibits diverging variance in challenging parameter regimes, resulting in unsuccessful parameter inference. In these cases, the other estimators provide more robust, lower variance gradients. Our results demonstrate that gradient-based parameter inference can be integrated effectively with the Gillespie SSA, with different estimators offering complementary advantages.

We introduce a fast low-rank factorization-based framework for compressing large language models that enables rapid compression of billion-parameter models without retraining. Unlike existing factorization-based approaches that optimize only on the original inputs, ignoring distribution shifts from upstream compression and thus propagating errors forward, or those that rely only on shifted inputs and risk drifting away from the original outputs, our approach accounts for both. Beyond individual layer compression, we further refine each transformer block end-to-end, minimizing block-level output distortion and allowing compressed layers to jointly compensate for accumulated errors. By anchoring each compressed layer to the original outputs while explicitly modeling input distribution shifts, our method finds a low-rank approximation that maintains functional equivalence with the original model. Experiments on large language models show that our method consistently outperforms existing SVD-based baselines across compression ratios, with the advantage becoming increasingly pronounced at aggressive compression budgets, where competing methods degrade substantially or collapse entirely, offering a practical solution for efficient, large-scale model deployment.

Evaluating factual correctness of LLM generated natural language explanations grounded in time series data remains an open challenge. Although modern models generate textual interpretations of numerical signals, existing evaluation methods are limited: reference based similarity metrics and consistency checking models require ground truth explanations, while traditional time series methods operate purely on numerical values and cannot assess free form textual reasoning. Thus, no general purpose method exists to directly verify whether an explanation is faithful to underlying time series data without predefined references or task specific rules. We study large language models as both generators and evaluators of time series explanations in a reference free setting, where given a time series, question, and candidate explanation, the evaluator assigns a ternary correctness label based on pattern identification, numeric accuracy, and answer faithfulness, enabling principled scoring and comparison. To support this, we construct a synthetic benchmark of 350 time series cases across seven query types, each paired with correct, partially correct, and incorrect explanations. We evaluate models across four tasks: explanation generation, relative ranking, independent scoring, and multi anomaly detection. Results show a clear asymmetry: generation is highly pattern dependent and exhibits systematic failures on certain query types, with accuracies ranging from 0.00 to 0.12 for Seasonal Drop and Volatility Shift, to 0.94 to 0.96 for Structural Break, while evaluation is more stable, with models correctly ranking and scoring explanations even when their own outputs are incorrect. These findings demonstrate feasibility of data grounded LLM based evaluation for time series explanations and highlight their potential as reliable evaluators of data grounded reasoning in the time series domain.

Steering vectors offer a training-free mechanism for controlling reasoning behaviors in large language models, but constructing effective vectors requires identifying genuine behavioral signals in the model's hidden states. For behaviors that can be toggled via prompts, this is straightforward. However, many reasoning behaviors -- such as self-reflection -- emerge spontaneously and resist prompt-level control. Current methods detect these behaviors through keyword matching in chain-of-thought traces, implicitly assuming that every detected boundary encodes a genuine behavioral signal. We show that this assumption is overwhelmingly wrong: across 541 keyword-detected boundaries, 93.3\% are behaviorally unstable, failing to reproduce the detected behavior under re-generation from the same prefix. We develop a probabilistic model that formalizes intrinsic reasoning behaviors as stochastic events with context-dependent trigger probabilities, and show that unstable boundaries dilute the steering signal. Guided by this analysis, we propose stability filtering, which retains only boundaries where the model consistently reproduces the target behavior. Combined with a content-subspace projection that removes residual question-specific noise, our method achieves 0.784 accuracy on MATH-500 (+5.0 over the strongest baseline). The resulting steering vectors transfer across models in the same architecture family without re-extraction, improving Nemotron-Research-Reasoning-1.5B (+5.0) and DeepScaleR-1.5B-Preview (+6.0). Code is available at https://github.com/zhmzm/stability-steering.

Estimating physical properties is critical for safe and efficient autonomous robotic manipulation, particularly during contact-rich interactions. In such settings, vision and tactile sensing provide complementary information about object geometry, pose, inertia, stiffness, and contact dynamics, such as stick-slip behavior. However, these properties are only indirectly observable and cannot always be modeled precisely (e.g., deformation in non-rigid objects coupled with nonlinear contact friction), making the estimation problem inherently complex and requiring sustained exploitation of visuo-tactile sensory information during action. Existing visuo-tactile perception frameworks have primarily emphasized forceful sensor fusion or static cross-modal alignment, with limited consideration of how uncertainty and beliefs about object properties evolve over time. Inspired by human multi-sensory perception and active inference, we propose the Cross-Modal Latent Filter (CMLF) to learn a structured, causal latent state-space of physical object properties. CMLF supports bidirectional transfer of cross-modal priors between vision and touch and integrates sensory evidence through a Bayesian inference process that evolves over time. Real-world robotic experiments demonstrate that CMLF improves the efficiency and robustness of latent physical properties estimation under uncertainty compared to baseline approaches. Beyond performance gains, the model exhibits perceptual coupling phenomena analogous to those observed in humans, including susceptibility to cross-modal illusions and similar trajectories in learning cross-sensory associations. Together, these results constitutes a significant step toward generalizable, robust and physically consistent cross-modal integration for robotic multi-sensory perception.

Data races in GPU programs pose a threat to the reliability of GPU-accelerated software stacks. Prior works proposed various dynamic (runtime) and static (compile-time) techniques to detect races in GPU programs. However, dynamic techniques often miss critical races, as they require the races to manifest during testing. While static ones can catch such races, they often generate numerous false alarms by conservatively assuming values of variables/parameters that cannot ever occur during any execution of the program. We make a key observation that the host (CPU) code that launches GPU kernels contains crucial semantic information about the values that the GPU kernel's parameters can take during execution. Harnessing this hitherto overlooked information helps accurately detect data races in GPU kernel code. We create HGRD, a new state-of-the-art static analysis technique that performs a holistic analysis of both CPU and GPU code to accurately detect a broad set of true races while minimizing false alarms. While SOTA dynamic techniques, such as iGUARD, miss many true races, HGRD misses none. On the other hand, static techniques such as GPUVerify and FaialAA raise tens of false alarms, where HGRD raises none.

Online handwriting represents strokes as time-ordered trajectories, which makes handwritten content easier to transform and reuse in a wide range of applications. However, generating natural sentence-level online handwriting that faithfully reflects a writer's style remains challenging, since sentence synthesis demands context-dependent characters with stroke continuity and spacing. Prior methods treat these boundary properties as implicit outcomes of sequence modeling, which becomes unreliable at the sentence scale and under limited compositional diversity. We propose CASHG, a context-aware stylized online handwriting generator that explicitly models inter-character connectivity for style-consistent sentence-level trajectory synthesis. CASHG uses a Character Context Encoder to obtain character identity and sentence-dependent context memory and fuses them in a bigram-aware sliding-window Transformer decoder that emphasizes local predecessor--current transitions, complemented by gated context fusion for sentence-level context.Training proceeds through a three-stage curriculum from isolated glyphs to full sentences, improving robustness under sparse transition coverage. We further introduce Connectivity and Spacing Metrics (CSM), a boundary-aware evaluation suite that quantifies cursive connectivity and spacing similarity. Under benchmark-matched evaluation protocols, CASHG consistently improves CSM over comparison methods while remaining competitive in DTW-based trajectory similarity, with gains corroborated by a human evaluation.

Speech representations from self-supervised speech models (S3Ms) are known to be sensitive to phonemic contrasts, but their sensitivity to prosodic contrasts has not been directly measured. The ABX discrimination task has been used to measure phonemic contrast in S3M representations via minimal pairs. We introduce prosodic ABX, an extension of this framework to evaluate prosodic contrast with only a handful of examples and no explicit labels. Also, we build and release a dataset of English and Japanese minimal pairs and use it along with a Mandarin dataset to evaluate contrast in English stress, Japanese pitch accent, and Mandarin tone. Finally, we show that model and layer rankings are often preserved across several experimental conditions, making it practical for low-resource settings.

Unified models (UMs) hold promise for their ability to understand and generate content across heterogeneous modalities. Compared to merely generating visual content, the use of UMs for interleaved cross-modal reasoning is more promising and valuable, e.g., for solving understanding problems that require dense visual thinking, improving visual generation through self-reflection, or modeling visual dynamics of the physical world guided by stepwise action interventions. However, existing UMs necessitate pixel decoding as a bridge due to their disjoint visual representations for understanding and generation, which is both ineffective and inefficient. In this paper, we introduce LatentUM, a novel unified model that represents all modalities within a shared semantic latent space, eliminating the need for pixel-space mediation between visual understanding and generation. This design naturally enables flexible interleaved cross-modal reasoning and generation. Beyond improved computational efficiency, the shared representation substantially alleviates codec bias and strengthens cross-modal alignment, allowing LatentUM to achieve state-of-the-art performance on the Visual Spatial Planning benchmark, push the limits of visual generation through self-reflection, and support world modeling by predicting future visual states within the shared semantic latent space.

Rerankers play a pivotal role in refining retrieval results for Retrieval-Augmented Generation. However, current reranking models are typically optimized on static human annotated relevance labels in isolation, decoupled from the downstream generation process. This isolation leads to a fundamental misalignment: documents identified as topically relevant by information retrieval metrics often fail to provide the actual utility required by the LLM for precise answer generation. To bridge this gap, we introduce ReRanking Preference Optimization (RRPO), a reinforcement learning framework that directly aligns reranking with the LLM's generation quality. By formulating reranking as a sequential decision-making process, RRPO optimizes for context utility using LLM feedback, thereby eliminating the need for expensive human annotations. To ensure training stability, we further introduce a reference-anchored deterministic baseline. Extensive experiments on knowledge-intensive benchmarks demonstrate that RRPO significantly outperforms strong baselines, including the powerful list-wise reranker RankZephyr. Further analysis highlights the versatility of our framework: it generalizes seamlessly to diverse readers (e.g., GPT-4o), integrates orthogonally with query expansion modules like Query2Doc, and remains robust even when trained with noisy supervisors.

Software malleability allows applications to be easily changed, configured, and adapted even after deployment. While prior work has explored configurable systems, adaptive recommender systems, and malleable GUIs, these approaches are often tailored to specific software and lack generalizability. In this work, we envision per-user malleable mobile applications, where end-users can specify requirements that are automatically implemented via LLM-based code generation. However, realizing this vision requires overcoming the key challenge of designing automated test generation that can reliably verify both the presence and correctness of user-specified functionalities. We propose \tool, a user-requirement-driven GUI test generation framework that incrementally navigates the UI, triggers desired functionalities, and constructs LLM-guided oracles to validate correctness. We build a benchmark spanning six popular mobile applications with both correct and faulty user-requested functionalities, demonstrating that \tool effectively validates per-user features and is practical for real-world deployment. Our work highlights the feasibility of shifting mobile app development from a product-manager-driven to an end-user-driven paradigm.

CI/CD pipelines are central to DevOps practices, yet their growing complexity makes them increasingly difficult to interpret, analyze, and systematically evolve. Existing tooling primarily offers execution logs and static graph representations, providing limited support for structured analysis of pipeline behavior, failures, and version-to-version evolution. This paper presents a model-driven Digital Twin (DT) for CI/CD pipelines that leverages BPMN as a model-ing backbone to transform raw CI configurations into structured, higher-level process representations. The proposed DT architecture enables visual abstraction of pipeline structure, failure tracing, and systematic version comparison, supporting both monitoring and evolution analysis of DevOps processes. Building upon validated DT architectural principles and prior work on build optimization and anomaly detection, the framework provides a modular, extensible foundation for integrating advanced analytical and prescriptive services into software delivery processes. The approach is validated using open-source CI/CD projects, and ongoing work targets the integration of additional improvement services and the extension of the DT to broader DevOps lifecycle processes.

Human-Object Interaction (HOI) detection aims to localize human-object pairs and classify their interactions from a single image, a task that demands strong visual understanding and nuanced contextual reasoning. Recent approaches have leveraged Vision-Language Models (VLMs) to introduce semantic priors, significantly improving HOI detection performance. However, existing methods often fail to fully capitalize on the diverse contextual cues distributed across the entire scene. To overcome these limitations, we propose the Instance-centric Context Mining Network (InCoM-Net)-a novel framework that effectively integrates rich semantic knowledge extracted from VLMs with instance-specific features produced by an object detector. This design enables deeper interaction reasoning by modeling relationships not only within each detected instance but also across instances and their surrounding scene context. InCoM-Net comprises two core components: Instancecentric Context Refinement (ICR), which separately extracts intra-instance, inter-instance, and global contextual cues from VLM-derived features, and Progressive Context Aggregation (ProCA), which iteratively fuses these multicontext features with instance-level detector features to support high-level HOI reasoning. Extensive experiments on the HICO-DET and V-COCO benchmarks show that InCoM-Net achieves state-of-the-art performance, surpassing previous HOI detection methods. Code is available at https://github.com/nowuss/InCoM-Net.

Multi-agent collaborative perception enables autonomous systems to overcome individual sensing limits through collective intelligence. However, real-world sensor and communication corruptions severely undermine this advantage. Crucially, existing approaches treat corruptions as static perturbations or passively conform to corrupted inputs, failing to actively recover the underlying clean semantics. To address this limitation, we introduce Diff-KD, a framework that integrates diffusion-based generative refinement into teacher-student knowledge distillation for robust collaborative perception. Diff-KD features two core components: (i) Progressive Knowledge Distillation (PKD), which treats local feature restoration as a conditional diffusion process to recover global semantics from corrupted observations; and (ii) Adaptive Gated Fusion (AGF), which dynamically weights neighbors based on ego reliability during fusion. Evaluated on OPV2V and DAIR-V2X under seven corruption types, Diff-KD achieves state-of-the-art performance in both detection accuracy and calibration robustness.

Recursive transformers reuse a shared weight block across multiple depth steps, trading parameters for compute. A core limitation: every step applies the same transformation, preventing the model from composing distinct operations across depth. We present Ouroboros, a system that attaches a compact Controller hypernetwork to a recursive transformer block. The Controller observes the current hidden state, produces a per-step diagonal modulation vector, and applies it to frozen SVD-initialized LoRA bases, making each recurrence step input-dependent. We combine this with gated recurrence (bias-initialized to 88% retention) and per-step LayerNorm for stable deep iteration. On Qwen2.5-3B split into a Prelude/Recurrent/Coda architecture (17 of 36 layers retained), Ouroboros reduces training loss by 43.4% over the unmodified 17-layer baseline, recovering 51.3% of the performance gap caused by layer removal. The full system adds only 9.2M trainable parameters (Controller, gate, and per-step norms) yet outperforms equivalently-sized static per-step LoRA by 1.44 loss points at depth 1 and remains ahead across all tested depths (1, 4, 8, 16) and ranks (8, 32, 64). We also find that gated recurrence is essential: without it, recursive layer application makes the model strictly worse. These gains are measured on the training distribution; on held-out text, the Controller does not yet improve over the baseline, a limitation we attribute to frozen downstream layers and discuss in detail. Code: https://github.com/RightNow-AI/ouroboros

Speculative decoding accelerates large language model inference by drafting multiple candidate tokens and verifying them in a single forward pass. Candidates are organized as a tree: deeper trees accept more tokens per step, but adding depth requires sacrificing breadth (fallback options) under a fixed verification budget. Existing training-free methods draft from a single token source and shape their trees without distinguishing candidate quality across origins. We observe that two common training-free token sources - n-gram matches copied from the input context, and statistical predictions from prior forward passes - differ dramatically in acceptance rate (~6x median gap, range 2-18x across five models and five benchmarks). We prove that when such a quality gap exists, the optimal tree is anisotropic (asymmetric): reliable tokens should form a deep chain while unreliable tokens spread as wide branches, breaking through the depth limit of balanced trees. We realize this structure in GOOSE, a training-free framework that builds an adaptive spine tree - a deep chain of high-acceptance context-matched tokens with wide branches of low-acceptance alternatives at each node. We prove that the number of tokens accepted per step is at least as large as that of either source used alone. On five LLMs (7B-33B) and five benchmarks, GOOSE achieves 1.9-4.3x lossless speedup, outperforming balanced-tree baselines by 12-33% under the same budget.

Transforming causal generative language models into bidirectional encoders offers a powerful alternative to BERT-style architectures. However, current approaches remain limited: they lack consensus on optimal training objectives, suffer from catastrophic forgetting at scale, and fail to flexibly integrate the vast ecosystem of specialized generative models. In this work, through systematic ablations on the Gemma3 and Qwen3 families, we identify the key factors driving successful adaptation, highlighting the critical role of an often-omitted prior masking phase. To scale this process without original pre-training data, we introduce a dual strategy combining linear weight merging with a lightweight multi-domain data mixture that mitigates catastrophic forgetting. Finally, we augment our encoders by merging them with specialized causal models, seamlessly transferring modality- and domain-specific capabilities. This open-source recipe, designed for any causal decoder LLM, yields BidirLM, a family of five encoders that outperform alternatives on text, vision, and audio representation benchmarks.

Self-supervised speech models learn effective representations of spoken language, which have been shown to reflect various aspects of linguistic structure. But when does such structure emerge in model training? We study the encoding of a wide range of linguistic structures, across layers and intermediate checkpoints of six Wav2Vec2 and HuBERT models trained on spoken Dutch. We find that different levels of linguistic structure show notably distinct layerwise patterns as well as learning trajectories, which can partially be explained by differences in their degree of abstraction from the acoustic signal and the timescale at which information from the input is integrated. Moreover, we find that the level at which pre-training objectives are defined strongly affects both the layerwise organization and the learning trajectories of linguistic structures, with greater parallelism induced by higher-order prediction tasks (i.e. iteratively refined pseudo-labels).

Modern software systems rely heavily on Web APIs, yet creating meaningful and executable test scripts remains a largely manual, time-consuming, and error-prone task. In this paper, we present APITestGenie, a novel tool that leverages Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and prompt engineering to automatically generate API integration tests directly from business requirements and OpenAPI specifications. We evaluated APITestGenie on 10 real-world APIs, including 8 APIs comprising circa 1,000 live endpoints from an industrial partner in the automotive domain. The tool was able to generate syntactically and semantically valid test scripts for 89\% of the business requirements under test after at most three attempts. Notably, some generated tests revealed previously unknown defects in the APIs, including integration issues between endpoints. Statistical analysis identified API complexity and level of detail in business requirements as primary factors influencing success rates, with the level of detail in API documentation also affecting outcomes. Feedback from industry practitioners confirmed strong interest in adoption, substantially reducing the manual effort in writing acceptance tests, and improving the alignment between tests and business requirements.

We study a speculative trading problem within the exploratory reinforcement learning (RL) framework of Wang et al. [2020]. The problem is formulated as a sequential optimal stopping problem over entry and exit times under general utility function and price process. We first consider a relaxed version of the problem in which the stopping times are modeled by the jump times of Cox processes driven by bounded, non-randomized intensity controls. Under the exploratory formulation, the agent's randomized control is characterized via the probability measure over the jump intensities, and their objective function is regularized by Shannon's differential entropy. This yields a system of the exploratory HJB equations and Gibbs distributions in closed-form as the optimal policy. Error estimates and convergence of the RL objective to the value function of the original problem are established. Finally, an RL algorithm is designed, and its implementation is showcased in a pairs-trading application.

Insurance application processes often rely on lengthy and standardized questionnaires that struggle to capture individual differences. Moreover, insurers must blindly trust users' responses, increasing the chances of fraud. The ARQuest framework introduces a new approach to underwriting by using Large Language Models (LLMs) and alternative data sources to create personalized and adaptive questionnaires. Techniques such as social media image analysis, geographic data categorization, and Retrieval Augmented Generation (RAG) are used to extract meaningful user insights and guide targeted follow-up questions. A life insurance system integrated into an industry partner mobile app was tested in two experiments. While traditional questionnaires yielded slightly higher accuracy in risk assessment, adaptive versions powered by GPT models required fewer questions and were preferred by users for their more fluid and engaging experience. ARQuest shows great potential to improve user satisfaction and streamline insurance processes. With further development, this approach may exceed traditional methods regarding risk accuracy and help drive innovation in the insurance industry.

Understanding human behaviour in crowded indoor environments is central to surveillance, smart buildings, and human-robot interaction, yet existing datasets rarely capture real-world indoor complexity at scale. We introduce IndoorCrowd, a multi-scene dataset for indoor human detection, instance segmentation, and multi-object tracking, collected across four campus locations (ACS-EC, ACS-EG, IE-Central, R-Central). It comprises $31$ videos ($9{,}913$ frames at $5$fps) with human-verified, per-instance segmentation masks. A $620$-frame control subset benchmarks three foundation-model auto-annotators: SAM3, GroundingSAM, and EfficientGroundingSAM, against human labels using Cohen's $κ$, AP, precision, recall, and mask IoU. A further $2{,}552$-frame subset supports multi-object tracking with continuous identity tracks in MOTChallenge format. We establish detection, segmentation, and tracking baselines using YOLOv8n, YOLOv26n, and RT-DETR-L paired with ByteTrack, BoT-SORT, and OC-SORT. Per-scene analysis reveals substantial difficulty variation driven by crowd density, scale, and occlusion: ACS-EC, with $79.3\%$ dense frames and a mean instance scale of $60.8$px, is the most challenging scene. The project page is available at https://sheepseb.github.io/IndoorCrowd/.

Autoencoders can be challenged by spatially non-uniform sampling of image content. This is common in medical imaging, biology, and physics, where informative patterns occur rarely at specific image coordinates, as background dominates these locations in most samples, biasing reconstructions toward the majority appearance. In practice, autoencoders are biased toward dominant patterns resulting in the loss of fine-grained detail and causing blurred reconstructions for rare spatial inputs especially under spatial data imbalance. We address spatial imbalance by two complementary components: (i) self-entropy-based loss that upweights statistically uncommon spatial locations and (ii) Sample Propagation, a replay mechanism that selectively re-exposes the model to hard to reconstruct samples across batches during training. We benchmark existing data balancing strategies, originally developed for supervised classification, in the unsupervised reconstruction setting. Drawing on the limitations of these approaches, our method specifically targets spatial imbalance by encouraging models to focus on statistically rare locations, improving reconstruction consistency compared to existing baselines. We validate in a simulated dataset with controlled spatial imbalance conditions, and in three, uncontrolled, diverse real-world datasets spanning physical, biological, and astronomical domains. Our approach outperforms baselines on various reconstruction metrics, particularly under spatial imbalance distributions. These results highlight the importance of data representation in a batch and emphasize rare samples in unsupervised image reconstruction. We will make all code and related data available.

Latent space is rapidly emerging as a native substrate for language-based models. While modern systems are still commonly understood through explicit token-level generation, an increasing body of work shows that many critical internal processes are more naturally carried out in continuous latent space than in human-readable verbal traces. This shift is driven by the structural limitations of explicit-space computation, including linguistic redundancy, discretization bottlenecks, sequential inefficiency, and semantic loss. This survey aims to provide a unified and up-to-date landscape of latent space in language-based models. We organize the survey into five sequential perspectives: Foundation, Evolution, Mechanism, Ability, and Outlook. We begin by delineating the scope of latent space, distinguishing it from explicit or verbal space and from the latent spaces commonly studied in generative visual models. We then trace the field's evolution from early exploratory efforts to the current large-scale expansion. To organize the technical landscape, we examine existing work through the complementary lenses of mechanism and ability. From the perspective of Mechanism, we identify four major lines of development: Architecture, Representation, Computation, and Optimization. From the perspective of Ability, we show how latent space supports a broad capability spectrum spanning Reasoning, Planning, Modeling, Perception, Memory, Collaboration, and Embodiment. Beyond consolidation, we discuss the key open challenges, and outline promising directions for future research. We hope this survey serves not only as a reference for existing work, but also as a foundation for understanding latent space as a general computational and systems paradigm for next-generation intelligence.

Diffusion models have become a standard approach for generative modeling in continuous domains, yet their application to discrete data remains challenging. We investigate why Gaussian diffusion models with the DDPM solver struggle to sample from discrete distributions that are represented as a mixture of delta-distributions in the continuous space. Using a toy Random Hierarchy Model, we identify a critical sampling interval in which the density of noisified data becomes multimodal. In this regime, DDPM occasionally enters low-density regions between modes producing out-of-distribution inputs for the model and degrading sample quality. We show that existing heuristics, including self-conditioning and a solver we term q-sampling, help alleviate this issue. Furthermore, we demonstrate that combining self-conditioning with switching from DDPM to q-sampling within the critical interval improves generation quality on real data. We validate these findings across conditional and unconditional tasks in multiple domains, including text, programming code, and proteins.

In this work, we extend our systematic capacity region perspective to multi-junction traffic networks, focussing on the special case of an urban corridor network. In particular, we train and evaluate centralized, fully decentralized, and parameter-sharing decentralized RL controllers, and compare their capacity regions and ATTs together with a classical baseline MaxPressure controller. Further, we show how the parametersharing controller may be generalised to be deployed on a larger network than it was originally trained on. In this setting, we show some initial findings that suggest that even though the junctions are not formally coordinated, traffic may self organise into `green waves'.

Autonomous agents are moving beyond simple retrieval tasks to become economic actors that invoke APIs, sequence workflows, and make real-time decisions. As this shift accelerates, API providers need request-level monetization with programmatic spend governance. The HTTP 402 protocol addresses this by treating payment as a first-class protocol event, but most implementations rely on cryptocurrency rails. In many deployment contexts, especially countries with strong real-time fiat systems like UPI, this assumption is misaligned with regulatory and infrastructure realities. We present APEX, an implementation-complete research system that adapts HTTP 402-style payment gating to UPI-like fiat workflows while preserving policy-governed spend control, tokenized access verification, and replay resistance. We implement a challenge-settle-consume lifecycle with HMAC-signed short-lived tokens, idempotent settlement handling, and policy-aware payment approval. The system uses FastAPI, SQLite, and Python standard libraries, making it transparent, inspectable, and reproducible. We evaluate APEX across three baselines and six scenarios using sample sizes 2-4x larger than initial experiments (N=20-40 per scenario). Results show that policy enforcement reduces total spending by 27.3% while maintaining 52.8% success rate for legitimate requests. Security mechanisms achieve 100% block rate for both replay attacks and invalid tokens with low latency overhead (19.6ms average). Multiple trial runs show low variance across scenarios, demonstrating high reproducibility with 95% confidence intervals. The primary contribution is a controlled agent-payment infrastructure and reference architecture that demonstrates how agentic access monetization can be adapted to fiat systems without discarding security and policy guarantees.

Evaluating the safety of LLM-based agents is increasingly important because risks in realistic deployments often emerge over multi-step interactions rather than isolated prompts or final responses. Existing trajectory-level benchmarks remain limited by insufficient interaction diversity, coarse observability of safety failures, and weak long-horizon realism. We introduce ATBench, a trajectory-level benchmark for structured, diverse, and realistic evaluation of agent safety. ATBench organizes agentic risk along three dimensions: risk source, failure mode, and real-world harm. Based on this taxonomy, we construct trajectories with heterogeneous tool pools and a long-context delayed-trigger protocol that captures realistic risk emergence across multiple stages. The benchmark contains 1,000 trajectories (503 safe and 497 unsafe), averaging 9.01 turns and 3.95k tokens, with 1,954 invoked tools drawn from pools spanning 2,084 available tools. Data quality is supported by rule-based and LLM-based filtering plus full human audit. Experiments on frontier LLMs, open-source models, and specialized guard systems show that ATBench is challenging even for strong evaluators, while enabling taxonomy-stratified analysis, cross-benchmark comparison, and diagnosis of long-horizon failure patterns.

Pool-based sequential active learning for regression (ALR) optimally selects a small number of samples sequentially from a large pool of unlabeled samples to label, so that a more accurate regression model can be constructed under a given labeling budget. Representativeness and diversity, which involve computing the distances among different samples, are important considerations in ALR. However, previous ALR approaches do not incorporate the importance of different features in inter-sample distance computation, resulting in sub-optimal sample selection. This paper proposes three feature weighted single-task ALR approaches and two feature weighted multi-task ALR approaches, where the ridge regression coefficients trained from a small amount of previously labeled samples are used to weight the corresponding features in inter-sample distance computation. Experiments showed that this easy-to-implement enhancement almost always improves the performance of four existing ALR approaches, in both single-task and multi-task regression problems. The feature weighting strategy may also be easily extended to stream-based ALR, and classification algorithms.

Demographic parity (DP) is a widely studied fairness criterion in regression, enforcing independence between the predictions and sensitive attributes. However, constraining the entire distribution can degrade predictive accuracy and may be unnecessary for many applications, where fairness concerns are localized to specific regions of the distribution. To overcome this issue, we propose a new framework for regression under DP that focuses on the tails of target distribution across sensitive groups. Our methodology builds on optimal transport theory. By enforcing fairness constraints only over targeted regions of the distribution, our approach enables more nuanced and context-sensitive interventions. Leveraging recent advances, we develop an interpretable and flexible algorithm that leverages the geometric structure of optimal transport. We provide theoretical guarantees, including risk bounds and fairness properties, and validate the method through experiments in regression settings.

Non-pharmaceutical interventions (NPIs) are commonly used tools for controlling infectious disease transmission when pharmaceutical options are unavailable. Yet, identifying effective interventions that minimize societal disruption remains challenging. Agent-based simulation is a popular tool for analyzing the impact of possible interventions in epidemiology. However, automatically optimizing NPIs using agent-based simulations poses a complex problem because, in agent-based epidemiological models, interventions can target individuals based on multiple attributes, affect hierarchical group structures (e.g., schools, workplaces, and families), and be combined arbitrarily, resulting in a very large or even infinite search space. We aim to support decision-makers with our Agent-based Infectious Disease Intervention Optimization System (ADIOS) that optimizes NPIs for infectious disease simulations using Grammar-Guided Genetic Programming (GGGP). The core of ADIOS is a domain-specific language for expressing NPIs in agent-based simulations that structures the intervention search space through a context-free grammar. To make optimization more efficient, the search space can be further reduced by defining constraints that prevent the generation of semantically invalid intervention patterns. Using this constrained language and an interface that enables coupling with agent-based simulations, ADIOS adopts the GGGP approach for simulation-based optimization. Using the German Epidemic Micro-Simulation System (GEMS) as a case study, we demonstrate the potential of our approach to generate optimal interventions for realistic epidemiological models

LLM-generated text (LGT) detection is essential for reliable forensic analysis and for mitigating LLM misuse. Existing LGT detectors can generally be categorized into two broad classes: learning-based approaches and zero-shot methods. Compared with learning-based detectors, zero-shot methods are particularly promising because they eliminate the need to train task-specific classifiers. However, the reliability of zero-shot methods fundamentally relies on the assumption that an off-the-shelf proxy LLM is well aligned with the often unknown source LLM, a premise that rarely holds in real-world black-box scenarios. To address this discrepancy, existing proxy alignment methods typically rely on supervised fine-tuning of the proxy or repeated interactions with commercial APIs, thereby increasing deployment costs, exposing detectors to silent API changes, and limiting robustness under domain shift. Motivated by these limitations, we propose the $k$-nearest neighbor proxy ($k$NNProxy), a training-free and query-efficient proxy alignment framework that repurposes the $k$NN language model ($k$NN-LM) retrieval mechanism as a domain adapter for a fixed proxy LLM. Specifically, a lightweight datastore is constructed once from a target-reflective LGT corpus, either via fixed-budget querying or from existing datasets. During inference, nearest-neighbor evidence induces a token-level predictive distribution that is interpolated with the proxy output, yielding an aligned prediction without proxy fine-tuning or per-token API outputs. To improve robustness under domain shift, we extend $k$NNProxy into a mixture of proxies (MoP) that routes each input to a domain-specific datastore for domain-consistent retrieval. Extensive experiments demonstrate strong detection performance of our method.

Building general-purpose reasoning models using reinforcement learning with verifiable rewards (RLVR) across diverse domains has been widely adopted by frontier open-weight models. However, their training recipes and domain mixtures are often not disclosed. Joint optimization across domains poses significant challenges: domains vary widely in rollout length, problem difficulty and sample efficiency. Further, models with long chain-of-thought traces increase inference cost and latency, making efficiency critical for practical deployment. We present Apriel-Reasoner, trained with a fully reproducible multi-domain RL post-training recipe on Apriel-Base, a 15B-parameter open-weight LLM, across five domains using public datasets: mathematics, code generation, instruction following, logical puzzles and function calling. We introduce an adaptive domain sampling mechanism that preserves target domain ratios despite heterogeneous rollout dynamics, and a difficulty-aware extension of the standard length penalty that, with no additional training overhead, encourages longer reasoning for difficult problems and shorter traces for easy ones. Trained with a strict 16K-token output budget, Apriel-Reasoner generalizes to 32K tokens at inference and improves over Apriel-Base on AIME 2025, GPQA, MMLU-Pro, and LiveCodeBench while producing 30-50% shorter reasoning traces. It matches strong open-weight models of similar size at lower token cost, thereby pushing the Pareto frontier of accuracy versus token budget.

Reinforcement Learning (RL) significantly enhances the reasoning abilities of large language models (LLMs), yet applying it to multi-turn agentic tasks remains challenging due to the long-horizon nature of interactions and the stochasticity of environmental feedback. We identify a structural failure mode in agentic exploration: suboptimal actions elicit noisy observations into misleading contexts, which further weaken subsequent decision-making, making recovery increasingly difficult. This cumulative feedback loop of errors renders standard exploration strategies ineffective and susceptible to the model's reasoning and the environment's randomness. To mitigate this issue, we propose ProCeedRL: Process Critic with Explorative Demonstration RL, shifting exploration from passive selection to active intervention. ProCeedRL employs a process-level critic to monitor interactions in real time, incorporating reflection-based demonstrations to guide agents in stopping the accumulation of errors. We find that this approach significantly exceeds the model's saturated exploration performance, demonstrating substantial exploratory benefits. By learning from exploratory demonstrations and on-policy samples, ProCeedRL significantly improves exploration efficiency and achieves superior performance on complex deep search and embodied tasks.

Transfer learning (TL) and deep ensemble learning (DE) have recently been shown to outperform simple machine learning in classifying psychiatric disorders. However, there is still a lack of understanding as to why that is. This paper aims to understand how and why DE and TL reduce the variability of single-subject classification models in bipolar disorder (BD) and schizophrenia (SCZ). To this end, we investigated the training stability of TL and DE models. For the two classification tasks under consideration, we compared the results of multiple trainings with the same backbone but with different initializations. In this way, we take into account the epistemic uncertainty associated with the uncertainty in the estimation of the model parameters. It has been shown that the performance of classifiers can be significantly improved by using TL with DE. Based on these results, we investigate i) how many models are needed to benefit from the performance improvement of DE when classifying BD and SCZ from healthy controls, and ii) how TL induces better generalization, with and without DE. In the first case, we show that DE reaches a plateau when 10 models are included in the ensemble. In the second case, we find that using a pre-trained model constrains TL models with the same pre-training to stay in the same basin of the loss function. This is not the case for DL models with randomly initialized weights.

Gait analysis provides an objective characterization of locomotor function and is widely used to support diagnosis and rehabilitation monitoring across neurological and orthopedic disorders. Deep learning has been increasingly applied to this domain, yet most approaches rely on supervised classifiers trained on disease-labeled data, limiting generalization to heterogeneous pathological presentations. This work proposes a label-free framework for joint-level anomaly detection and kinematic correction based on a Transformer masked autoencoder trained exclusively on normative gait sequences from 150 adults, acquired with a markerless multi-camera motion-capture system. At inference, a two-pass procedure is applied to potentially pathological input sequences, first it estimates joint inconsistency scores by occluding individual joints and measuring deviations from the learned normative prior. Then, it withholds the flagged joints from the encoder input and reconstructs the full skeleton from the remaining spatiotemporal context, yielding corrected kinematic trajectories at the flagged positions. Validation on 10 held-out normative participants, who mimicked seven simulated gait abnormalities, showed accurate localization of biomechanically inconsistent joints, a significant reduction in angular deviation across all analyzed joints with large effect sizes, and preservation of normative kinematics. The proposed approach enables interpretable, subject-specific localization of gait impairments without requiring disease labels. Video is available at https://youtu.be/Rcm3jqR5pN4.

Multi-hop QA benchmarks frequently reward Large Language Models (LLMs) for spurious correctness, masking ungrounded or flawed reasoning steps. To shift toward rigorous reasoning, we propose SAFE, a dynamic benchmarking framework that replaces the ungrounded Chain-of-Thought (CoT) with a strictly verifiable sequence of grounded entities. Our framework operates across two phases: (1) train-time verification, where we establish an atomic error taxonomy and a Knowledge Graph (KG)-grounded verification pipeline to eliminate noisy supervision in standard benchmarks, identifying up to 14% of instances as unanswerable, and (2) inference-time verification, where a feedback model trained on this verified dataset dynamically detects ungrounded steps in real-time. Experimental results demonstrate that SAFE not only exposes the critical flaws of existing benchmarks at train-time, but also significantly outperforms standard baselines, achieving an average accuracy gain of 8.4 pp while guaranteeing verifiable trajectories at inference-time.

Like a body at rest that stays at rest, we find that visual attention in multimodal large language models (MLLMs) exhibits pronounced inertia, remaining largely static once settled during early decoding steps and failing to support the compositional understanding required for cognitive inference. While existing hallucination mitigation methods mainly target perceptual hallucinations concerning object existence or attributes, they remain inadequate for such cognitive hallucinations that require inter-object relational deduction. Through token-wise attention analysis, we identify this visual inertia as a key factor: attention to semantically critical regions remains persistently focused and fails to dynamically support relational inference. We thereby propose a training-free Inertia-aware Visual Excitation (IVE) method that breaks this inertial pattern by modeling cognitive inference as the dynamic responsiveness of visual attention. Specifically, IVE selects visual tokens that are dynamically emerging relative to historical attention trends while distinguishing tokens exhibiting inertial behavior. To further facilitate compositional inference, IVE introduces an inertia-aware penalty that discourages over-concentration and limits the persistence of attention within localized regions. Extensive experiments show that IVE is effective across various base MLLMs and multiple hallucination benchmarks, particularly for cognitive hallucinations.

Large language models often default to step-by-step computation even when efficient numerical shortcuts are available. This raises a basic question: do they exhibit number sense in a human-like behavioral sense, i.e., the ability to recognize numerical structure, apply shortcuts when appropriate, and avoid them when they are not? We introduce SenseMath, a controlled benchmark for evaluating structure-sensitive numerical reasoning in LLMs. SenseMath contains 4,800 items spanning eight shortcut categories and four digit scales, with matched strong-shortcut, weak-shortcut, and control variants. It supports three evaluation settings of increasing cognitive demand: Shortcut Use (whether models can apply shortcuts on shortcut-amenable problems); Applicability Judgment (whether they can recognize when a shortcut is appropriate or misleading); and Problem Generation (whether they can generate new problem items that correctly admit a given type of shortcut). Our evaluation across five LLMs, ranging from GPT-4o-mini to Llama-3.1-8B, shows a consistent pattern: when explicitly prompted, models readily adopt shortcut strategies and achieve substantial accuracy gains on shortcut-amenable items (up to 15%), yet under standard chain-of-thought prompting they spontaneously employ such strategies in fewer than 40% of cases, even when they demonstrably possess the requisite capability. Moreover, this competence is confined to the Use level; models systematically over-generalise shortcuts to problems where they do not apply, and fail to generate valid shortcut-bearing problems from scratch. Together, these results suggest that current LLMs exhibit procedural shortcut fluency without the structural understanding of when and why shortcuts work that underlies human number sense.

The rapid growth of medical imaging has fueled the development of Foundation Models (FMs) to reduce the growing, unsustainable workload on radiologists. While recent FMs have shown the power of large-scale pre-training to CT and MRI analysis, there remains significant room to optimize how these models learn from complex radiological volumes. Building upon the Curia framework, this work introduces Curia-2, which significantly improves the original pre-training strategy and representation quality to better capture the specificities of radiological data. The proposed methodology enables scaling the architecture up to billion-parameter Vision Transformers, marking a first for multi-modal CT and MRI FMs. Furthermore, we formalize the evaluation of these models by extending and restructuring CuriaBench into two distinct tracks: a 2D track tailored for slice-based vision models and a 3D track for volumetric benchmarking. Our results demonstrate that Curia-2 outperforms all FMs on vision-focused tasks and fairs competitively to vision-language models on clinically complex tasks such as finding detection. Weights will be made publicly available to foster further research.

General-purpose world models promise scalable policy evaluation, optimization, and planning, yet achieving the required level of robustness remains challenging. Unlike policy learning, which primarily focuses on optimal actions, a world model must be reliable over a much broader range of suboptimal actions, which are often insufficiently covered by action-labeled interaction data. To address this challenge, we propose World Action Verifier (WAV), a framework that enables world models to identify their own prediction errors and self-improve. The key idea is to decompose action-conditioned state prediction into two factors -- state plausibility and action reachability -- and verify each separately. We show that these verification problems can be substantially easier than predicting future states due to two underlying asymmetries: the broader availability of action-free data and the lower dimensionality of action-relevant features. Leveraging these asymmetries, we augment a world model with (i) a diverse subgoal generator obtained from video corpora and (ii) a sparse inverse model that infers actions from a subset of state features. By enforcing cycle consistency among generated subgoals, inferred actions, and forward rollouts, WAV provides an effective verification mechanism in under-explored regimes, where existing methods typically fail. Across nine tasks spanning MiniGrid, RoboMimic, and ManiSkill, our method achieves 2x higher sample efficiency while improving downstream policy performance by 18%.

We study a random model of deep multi-head self-attention in which the weights are resampled independently across layers and heads, as at initialization of training. Viewing depth as a time variable, the residual stream defines a discrete-time interacting particle system on the unit sphere. We prove that, under suitable joint scalings of the depth, the residual step size, and the number of heads, this dynamics admits a nontrivial homogenized limit. Depending on the scaling, the limit is either deterministic or stochastic with common noise; in the mean-field regime, the latter leads to a stochastic nonlinear Fokker--Planck equation for the conditional law of a representative token. In the Gaussian setting, the limiting drift vanishes, making the homogenized dynamics explicit enough to study representation collapse. This yields quantitative trade-offs between dimension, context length, and temperature, and identifies regimes in which clustering can be mitigated.

Security teams face a challenge: the volume of newly disclosed Common Vulnerabilities and Exposures (CVEs) far exceeds the capacity to manually develop detection mechanisms. In 2025, the National Vulnerability Database published over 48,000 new vulnerabilities, motivating the need for automation. We present RuleForge, an AWS internal system that automatically generates detection rules--JSON-based patterns that identify malicious HTTP requests exploiting specific vulnerabilities--from structured Nuclei templates describing CVE details. Nuclei templates provide standardized, YAML-based vulnerability descriptions that serve as the structured input for our rule generation process. This paper focuses on RuleForge's architecture and operational deployment for CVE-related threat detection, with particular emphasis on our novel LLM-as-a-judge (Large Language Model as judge) confidence validation system and systematic feedback integration mechanism. This validation approach evaluates candidate rules across two dimensions--sensitivity (avoiding false negatives) and specificity (avoiding false positives)--achieving AUROC of 0.75 and reducing false positives by 67% compared to synthetic-test-only validation in production. Our 5x5 generation strategy (five parallel candidates with up to five refinement attempts each) combined with continuous feedback loops enables systematic quality improvement. We also present extensions enabling rule generation from unstructured data sources and demonstrate a proof-of-concept agentic workflow for multi-event-type detection. Our lessons learned highlight critical considerations for applying LLMs to cybersecurity tasks, including overconfidence mitigation and the importance of domain expertise in both prompt design and quality review of generated rules through human-in-the-loop validation.

We present the first systematic analysis of multimodal large language models (MLLMs) in personalized question-answering requiring ego-grounding - the ability to understand the camera-wearer in egocentric videos. To this end, we introduce MyEgo, the first egocentric VideoQA dataset designed to evaluate MLLMs' ability to understand, remember, and reason about the camera wearer. MyEgo comprises 541 long videos and 5K personalized questions asking about "my things", "my activities", and "my past". Benchmarking reveals that competitive MLLMs across variants, including open-source vs. proprietary, thinking vs. non-thinking, small vs. large scales all struggle on MyEgo. Top closed- and open-source models (e.g., GPT-5 and Qwen3-VL) achieve only~46% and 36% accuracy, trailing human performance by near 40% and 50% respectively. Surprisingly, neither explicit reasoning nor model scaling yield consistent improvements. Models improve when relevant evidence is explicitly provided, but gains drop over time, indicating limitations in tracking and remembering "me" and "my past". These findings collectively highlight the crucial role of ego-grounding and long-range memory in enabling personalized QA in egocentric videos. We hope MyEgo and our analyses catalyze further progress in these areas for egocentric personalized assistance. Data and code are available at https://github.com/Ryougetsu3606/MyEgo

Scientific knowledge discovery increasingly relies on large language models, yet many existing scholarly assistants depend on proprietary systems with tens or hundreds of billions of parameters. Such reliance limits reproducibility and accessibility for the research community. In this work, we ask a simple question: do we need bigger models for scientific applications? Specifically, we investigate to what extent carefully designed retrieval pipelines can compensate for reduced model scale in scientific applications. We design a lightweight retrieval-augmented framework that performs task-aware routing to select specialized retrieval strategies based on the input query. The system further integrates evidence from full-text scientific papers and structured scholarly metadata, and employs compact instruction-tuned language models to generate responses with citations. We evaluate the framework across several scholarly tasks, focusing on scholarly question answering (QA), including single- and multi-document scenarios, as well as biomedical QA under domain shift and scientific text compression. Our findings demonstrate that retrieval and model scale are complementary rather than interchangeable. While retrieval design can partially compensate for smaller models, model capacity remains important for complex reasoning tasks. This work highlights retrieval and task-aware design as key factors for building practical and reproducible scholarly assistants.

Abnormal head movements (AHMs) manifest across a broad spectrum of neurological disorders; however, the absence of a multi-condition resource integrating kinematic measurements, clinical severity scores, and patient demographics constitutes a persistent barrier to the development of AI-driven diagnostic tools. To address this gap, this study introduces NeuroPose-AHM, a knowledge-based dataset of neurologically induced AHMs constructed through a multi-LLM extraction framework applied to 1,430 peer-reviewed publications. The dataset contains 2,756 patient-group-level records spanning 57 neurological conditions, derived from 846 AHM-relevant papers. Inter-LLM reliability analysis confirms robust extraction performance, with study-level classification achieving strong agreement (kappa = 0.822). To demonstrate the dataset's analytical utility, a four-task framework is applied to cervical dystonia (CD), the condition most directly defined by pathological head movement. First, Task 1 performs multi-label AHM type classification (F1 = 0.856). Task 2 constructs the Head-Neck Severity Index (HNSI), a unified metric that normalizes heterogeneous clinical rating scales. The clinical relevance of this index is then evaluated in Task 3, where HNSI is validated against real-world CD patient data, with aligned severe-band proportions (6.7%) providing a preliminary plausibility indication for index calibration within the high severity range. Finally, Task 4 performs bridge analysis between movement-type probabilities and HNSI scores, producing significant correlations (p less than 0.001). These results demonstrate the analytical utility of NeuroPose-AHM as a structured, knowledge-based resource for neurological AHM research. The NeuroPose-AHM dataset is publicly available on Zenodo (https://doi.org/10.5281/zenodo.19386862).

Multiple operator learning concerns learning operator families $\{G[α]:U\to V\}_{α\in W}$ indexed by an operator descriptor $α$. Training data are collected hierarchically by sampling operator instances $α$, then input functions $u$ per instance, and finally evaluation points $x$ per input, yielding noisy observations of $G[α][u](x)$. While recent work has developed expressive multi-task and multiple operator learning architectures and approximation-theoretic scaling laws, quantitative statistical generalization guarantees remain limited. We provide a covering-number-based generalization analysis for separable models, focusing on the Multiple Neural Operator (MNO) architecture: we first derive explicit metric-entropy bounds for hypothesis classes given by linear combinations of products of deep ReLU subnetworks, and then combine these complexity bounds with approximation guarantees for MNO to obtain an explicit approximation-estimation tradeoff for the expected test error on new (unseen) triples $(α,u,x)$. The resulting bound makes the dependence on the hierarchical sampling budgets $(n_α,n_u,n_x)$ transparent and yields an explicit learning-rate statement in the operator-sampling budget $n_α$, providing a sample-complexity characterization for generalization across operator instances. The structure and architecture can also be viewed as a general purpose solver or an example of a "small'' PDE foundation model, where the triples are one form of multi-modality.

Machine-translated benchmark datasets reduce costs and offer scale, but noise, loss of structure, and uneven quality weaken confidence. What matters is not merely whether we can translate, but also whether we can measure and verify translation reliability at scale. We study translation quality in the EU20 benchmark suite, which comprises five established benchmarks translated into 20 languages, via a three-step automated quality assurance approach: (i) a structural corpus audit with targeted fixes; (ii) quality profiling using a neural metric (COMET, reference-free and reference-based) with translation service comparisons (DeepL / ChatGPT / Google); and (iii) an LLM-based span-level translation error landscape. Trends are consistent: datasets with lower COMET scores exhibit a higher share of accuracy/mistranslation errors at span level (notably HellaSwag; ARC is comparatively clean). Reference-based COMET on MMLU against human-edited samples points in the same direction. We release cleaned/corrected versions of the EU20 datasets, and code for reproducibility. In sum, automated quality assurance offers practical, scalable indicators that help prioritize review -- complementing, not replacing, human gold standards.

We introduce Qiana, a logic framework for reasoning on formulas that are true only in specific contexts. In Qiana, it is possible to quantify over both formulas and contexts to express, e.g., that ``everyone knows everything Alice says''. Qiana also permits paraconsistent logics within contexts, so that contexts can contain contradictions. Furthermore, Qiana is based on first-order logic, and is finitely axiomatizable, so that Qiana theories are compatible with pre-existing first-order logic theorem provers. We show how Qiana can be used to represent temporality, event calculus, and modal logic. We also discuss different design alternatives of Qiana.

We propose LSCP, a self-gated post-training framework for autonomous knowledge acquisition: learning only what a model does not already know, verified against what it does know, at a strength proportional to conviction, with no external oracle. When a passage produces anomalously high per-token loss, LSCP flags it, generates a Q&A chain that forces the model to articulate its own knowledge and identify gaps, then adjusts AdamW's $β_2$ proportionally to conviction depth k (the number of self-verification steps the passage survives) via $β_2 = 0.999 \cdot r^k$. The entire learning intensity is governed by a single parameter $r$. Beyond new knowledge, this process sharpens weakly encoded existing knowledge, which is a primary source of hallucination. The framework is self-extinguishing: as the model learns, per-token loss on learned passages decreases toward the surprisal threshold and the system progressively converges to standard AdamW. This models biological memory consolidation: temporary information in the context window is selectively consolidated into parametric weights, the model's long-term memory. Experiments on the reference model (Qwen3-14B) and across six models (8B--32B, four families) show that standard fine-tuning produces rote memorization (perturbation gap (the ratio of paraphrase to original perplexity) of 11.6 +- 0.2 x baseline) while all LSCP conditions learn semantically (2.7--3.0x). The r=1.0 condition (identical optimizer, nearly identical data, only Q&A format differs) confirms that the training data format, not $β_2$ gating, is the primary mechanism preventing memorization; gating instead protects neighboring knowledge from contamination by corrupt content (93 +- 7% accuracy on adjacent questions at r=0.98 vs. 90% baseline).

The scale of biological datasets now routinely exceeds system memory, making data access rather than model computation the primary bottleneck in training machine-learning models. This bottleneck is particularly acute in biology, where widely used community data formats must support heterogeneous metadata, sparse and dense assays, and downstream analysis within established computational ecosystems. Here we present annbatch, a mini-batch loader native to anndata that enables out-of-core training directly on disk-backed datasets. Across single-cell transcriptomics, microscopy and whole-genome sequencing benchmarks, annbatch increases loading throughput by up to an order of magnitude and shortens training from days to hours, while remaining fully compatible with the scverse ecosystem. Annbatch establishes a practical data-loading infrastructure for scalable biological AI, allowing increasingly large and diverse datasets to be used without abandoning standard biological data formats. Github: https://github.com/scverse/annbatch

Estimating optimal individualized treatment rules (ITRs) via outcome weighted learning (OWL) often relies on observed rewards that are noisy or optimistic proxies for the true latent utility. Ignoring this reward uncertainty leads to the selection of policies with inflated apparent performance, yet existing OWL frameworks lack the finite-sample guarantees required to systematically embed such uncertainty into the learning objective. To address this issue, we propose PAC-Bayesian Reward-Certified Outcome Weighted Learning (PROWL). Given a one-sided uncertainty certificate, PROWL constructs a conservative reward and a strictly policy-dependent lower bound on the true expected value. Theoretically, we prove an exact certified reduction that transforms robust policy learning into a unified, split-free cost-sensitive classification task. This formulation enables the derivation of a nonasymptotic PAC-Bayes lower bound for randomized ITRs, where we establish that the optimal posterior maximizing this bound is exactly characterized by a general Bayes update. To overcome the learning-rate selection problem inherent in generalized Bayesian inference, we introduce a fully automated, bounds-based calibration procedure, coupled with a Fisher-consistent certified hinge surrogate for efficient optimization. Our experiments demonstrate that PROWL achieves improvements in estimating robust, high-value treatment regimes under severe reward uncertainty compared to standard methods for ITR estimation.

Wideband channel frequency response (CFR) estimation is challenging in multi-band wireless systems, especially when one or more sub-bands are temporarily blocked by co-channel interference. We present a physics-informed complex Transformer that reconstructs the full wideband CFR from such fragmented, partially observed spectrum snapshots. The interference pattern in each sub-band is modeled as an independent two-state discrete-time Markov chain, capturing realistic bursty occupancy behavior. Our model operates on the joint time-frequency grid of $T$ snapshots and $F$ frequency bins and uses a factored self-attention mechanism that separately attends along both axes, reducing the computational complexity to $O(TF^2 + FT^2)$. Complex-valued inputs and outputs are processed through a holomorphic linear layer that preserves phase relationships. Training uses a composite physics-informed loss combining spectral fidelity, power delay profile (PDP) reconstruction, channel impulse response (CIR) sparsity, and temporal smoothness. Mobility effects are incorporated through per-sample velocity randomization, enabling generalization across different mobility regimes. Evaluation against three classical baselines, namely, last-observation-carry-forward, zero-fill, and cubic-spline interpolation, shows that our approach achieves the highest PDP similarity with respect to the ground truth, reaching $ρ\geq 0.82$ compared to $ρ\geq 0.62$ for the best baseline at interference occupancy levels up to 50%. Furthermore, the model degrades smoothly across the full velocity range, consistently outperforming all other baselines.

This paper addresses the problem of clustering measurement vectors that are heteroscedastic in that they can have different covariance matrices. From the assumption that the measurement vectors within a given cluster are Gaussian distributed with possibly different and unknown covariant matrices around the cluster centroid, we introduce a novel cost function to estimate the centroids. The zeros of the gradient of this cost function turn out to be the fixed-points of a certain function. As such, the approach generalizes the methodology employed to derive the existing Mean-Shift algorithm. But as a main and novel theoretical result compared to Mean-Shift, this paper shows that the sole fixed-points of the identified function tend to be the cluster centroids if both the number of measurements per cluster and the distances between centroids are large enough. As a second contribution, this paper introduces the Wald kernel for clustering. This kernel is defined as the p-value of the Wald hypothesis test for testing the mean of a Gaussian. As such, the Wald kernel measures the plausibility that a measurement vector belongs to a given cluster and it scales better with the dimension of the measurement vectors than the usual Gaussian kernel. Finally, the proposed theoretical framework allows us to derive a new clustering algorithm called CENTRE-X that works by estimating the fixed-points of the identified function. As Mean-Shift, CENTRE-X requires no prior knowledge of the number of clusters. It relies on a Wald hypothesis test to significantly reduce the number of fixed points to calculate compared to the Mean-Shift algorithm, thus resulting in a clear gain in complexity. Simulation results on synthetic and real data sets show that CENTRE-X has comparable or better performance than standard clustering algorithms K-means and Mean-Shift, even when the covariance matrices are not perfectly known.

Image captioning for Early Childhood Education (ECE) is essential for automated activity understanding and educational assessment. However, existing methods face two key challenges. First, the lack of large-scale, domain-specific datasets limits the model's ability to capture fine-grained semantic concepts unique to ECE scenarios, resulting in generic and imprecise descriptions. Second, conventional training paradigms exhibit limitations in enhancing professional object description capability, as supervised learning tends to favor high-frequency expressions, while reinforcement learning may suffer from unstable optimization on difficult samples. To address these limitations, we introduce ECAC, a large-scale benchmark for ECE daily activity image captioning, comprising 256,121 real-world images annotated with expert-level captions and fine-grained labels. ECAC is further equipped with a domain-oriented evaluation protocol, the Teaching Toy Recognition Score (TTS), to explicitly measure professional object naming accuracy. Furthermore, we propose RSRS (Reward-Conditional Switch of Reinforcement Learning and Supervised Fine-Tuning), a hybrid training framework that dynamically alternates between RL and supervised optimization. By rerouting hard samples with zero rewards to supervised fine-tuning, RSRS effectively mitigates advantage collapse and enables stable optimization for fine-grained recognition. Leveraging ECAC and RSRS, we develop KinderMM-Cap-3B, a domain-adapted multimodal large language model. Extensive experiments demonstrate that our model achieves a TTS of 51.06, substantially outperforming state-of-the-art baselines while maintaining superior caption quality, highlighting its potential for specialized educational applications.

We consider learning with possibilistic supervision for multi-class classification. For each training instance, the supervision is a normalized possibility distribution that expresses graded plausibility over the classes. From this possibility distribution, we construct a non-empty closed convex set of admissible probability distributions by combining two requirements: probabilistic compatibility with the possibility and necessity measures induced by the possibility distribution, and linear shape constraints that must be satisfied to preserve the qualitative structure of the possibility distribution. Thus, classes with the same possibility degree receive equal probabilities, and if a class has a strictly larger possibility degree than another class, then it receives a strictly larger probability. Given a strictly positive probability vector output by a model for an instance, we compute its Kullback-Leibler projection onto the admissible set. This projection yields the closest admissible probability distribution in Kullback-Leibler sense. We can then train the model by minimizing the divergence between the prediction and its projection, which quantifies the smallest adjustment needed to satisfy the induced dominance and shape constraints. The projection is computed with Dykstra's algorithm using Bregman projections associated with the negative entropy, and we provide explicit formulas for the projections onto each constraint set. Experiments conducted on synthetic data and on a real-world natural language inference task, based on the ChaosNLI dataset, show that the proposed projection algorithm is efficient enough for practical use, and that the resulting projection-based learning objective can improve predictive performance.

The structure of all the permutations of a sequence can be represented as a permutohedron, a graph where vertices are permutations and two vertices are linked if a swap of adjacent elements in the permutation of one of the vertices produces the permutation of the other vertex. It has been hypothesized that word orders in languages minimize the swap distance in the permutohedron: given a source order, word orders that are closer in the permutohedron should be less costly and thus more likely. Here we explain how to measure the degree of optimality of word order variation with respect to swap distance minimization. We illustrate the power of our novel mathematical framework by showing that crosslinguistic gestures are at least $77\%$ optimal. It is unlikely that the multiple times where crosslinguistic gestures hit optimality are due to chance. We establish the theoretical foundations for research on the optimality of word or gesture order with respect to swap distance minimization in communication systems. Finally, we introduce the quadratic assignment problem (QAP) into language research as an umbrella for multiple optimization problems and, accordingly, postulate a general principle of optimal assignment that unifies various linguistic principles including swap distance minimization.

Among news disorders, propagandist news are particularly insidious, because they tend to mix oriented messages with factual reports intended to look like reliable news. To detect propaganda, extant approaches based on Language Models such as BERT are promising but often overfit their training datasets, due to biases in data collection. To enhance classification robustness and improve generalization to new sources, we propose a neurosymbolic approach combining non-contextual text embeddings (fastText) with symbolic conceptual features such as genre, topic, and persuasion techniques. Results show improvements over equivalent text-only methods, and ablation studies as well as explainability analyses confirm the benefits of the added features. Keywords: Information disorder, Fake news, Propaganda, Classification, Topic modeling, Hybrid method, Neurosymbolic model, Ablation, Robustness

Most of the Neural Cellular Automata (NCAs) defined in the literature have a common theme: they are based on regular grids with a Moore neighborhood (one-hop neighbour). They do not take into account long-range connections and more complex topologies as we can find in the brain. In this paper, we introduce BraiNCA, a brain-inspired NCA with an attention layer, long-range connections and complex topology. BraiNCAs shows better results in terms of robustness and speed of learning on the two tasks compared to Vanilla NCAs establishing that incorporating attention-based message selection together with explicit long-range edges can yield more sample-efficient and damage-tolerant self-organization than purely local, grid-based update rules. These results support the hypothesis that, for tasks requiring distributed coordination over extended spatial and temporal scales, the choice of interaction topology and the ability to dynamically route information will impact the robustness and speed of learning of an NCA. More broadly, BraiNCA provides brain-inspired NCA formulation that preserves the decentralized local update principle while better reflecting non-local connectivity patterns, making it a promising substrate for studying collective computation under biologically-realistic network structure and evolving cognitive substrates.

We propose a geometry-driven quantum-inspired classification framework that integrates Correlation Group Structures (CGR), compact SWAP-test-based overlap estimation, and selective variational quantum decision modelling. Rather than directly approximating class posteriors, the method adopts a geometry-first paradigm in which samples are evaluated relative to class medoids using overlap-derived Euclidean-like and angular similarity channels. CGR organizes features into anchor-centered correlation neighbourhoods, generating nonlinear, correlation-weighted representations that enhance robustness in heterogeneous tabular spaces. These geometric signals are fused through a non-probabilistic margin-based fusion score, serving as a lightweight and data-efficient primary classifier for small-to-moderate datasets. On Heart Disease, Breast Cancer, and Wine Quality datasets, the fusion-score classifier achieves 0.8478, 0.8881, and 0.9556 test accuracy respectively, with macro-F1 scores of 0.8463, 0.8703, and 0.9522, demonstrating competitive and stable performance relative to classical baselines. For large-scale and highly imbalanced regimes, we construct compact Delta-distance contrastive features and train a variational quantum classifier (VQC) as a nonlinear refinement layer. On the Credit Card Fraud dataset (0.17% prevalence), the Delta + VQC pipeline achieves approximately 0.85 minority recall at an alert rate of approximately 1.31%, with ROC-AUC 0.9249 and PR-AUC 0.3251 under full-dataset evaluation. These results highlight the importance of operating-point-aware assessment in rare-event detection and demonstrate that the proposed hybrid geometric-variational framework provides interpretable, scalable, and regime-adaptive classification across heterogeneous data settings.

The audio research community depends on open generative models as foundational tools for building novel approaches and establishing baselines. In this report, we present Woosh, Sony AI's publicly released sound effect foundation model, detailing its architecture, training process, and an evaluation against other popular open models. Being optimized for sound effects, we provide (1) a high-quality audio encoder/decoder model and (2) a text-audio alignment model for conditioning, together with (3) text-to-audio and (4) video-to-audio generative models. Distilled text-to-audio and video-to-audio models are also included in the release, allowing for low-resource operation and fast inference. Our evaluation on both public and private data shows competitive or better performance for each module when compared to existing open alternatives like StableAudio-Open and TangoFlux. Inference code and model weights are available at https://github.com/SonyResearch/Woosh. Demo samples can be found at https://sonyresearch.github.io/Woosh/.

Large Language Models increasingly suppress biased outputs when demographic identity is stated explicitly, yet may still exhibit implicit biases when identity is conveyed indirectly. Existing benchmarks use name based proxies to detect implicit biases, which carry weak associations with many social demographics and cannot extend to dimensions like age or socioeconomic status. We introduce ImplicitBBQ, a QA benchmark that evaluates implicit bias through characteristic based cues, culturally associated attributes that signal implicitly, across age, gender, region, religion, caste, and socioeconomic status. Evaluating 11 models, we find that implicit bias in ambiguous contexts is over six times higher than explicit bias in open weight models. Safety prompting and chain-of-thought reasoning fail to substantially close this gap; even few-shot prompting, which reduces implicit bias by 84%, leaves caste bias at four times the level of any other dimension. These findings indicate that current alignment and prompting strategies address the surface of bias evaluation while leaving culturally grounded stereotypic associations largely unresolved. We publicly release our code and dataset for model providers and researchers to benchmark potential mitigation techniques.

Accurate short-term mortality prediction in heart failure (HF) remains challenging, particularly when relying on structured electronic health record (EHR) data alone. We evaluate transformer-based models on a French HF cohort, comparing text-only, structured-only, multimodal, and LLM-based approaches. Our results show that enriching clinical text with entity-level representations improves prediction over CLS embeddings alone, and that supervised multimodal fusion of text and structured variables achieves the best overall performance. In contrast, large language models perform inconsistently across modalities and decoding strategies, with text-only prompts outperforming structured or multimodal inputs. These findings highlight that entity-aware multimodal transformers offer the most reliable solution for short-term HF outcome prediction, while current LLM prompting remains limited for clinical decision support.

Automotive radar perception pipelines commonly construct angle-domain representations via beamforming before applying learning-based models. This work instead investigates a representational question: can meaningful spatial structure be learned directly from pre-beamforming per-antenna range-Doppler (RD) measurements? Experiments are conducted on a 6-TX x 8-RX (48 virtual antennas) commodity automotive radar employing an A/B chirp-sequence frequency-modulated continuous-wave (CS-FMCW) transmit scheme, in which the effective transmit aperture varies between chirps (single-TX vs. multi-TX), enabling controlled analysis of chirp-dependent transmit configurations. We operate on pre-beamforming per-antenna RD tensors using a dual-chirp shared-weight encoder trained in an end-to-end, fully data-driven manner, and evaluate spatial recoverability using bird's-eye-view (BEV) occupancy as a geometric probe rather than a performance-driven objective. Supervision is visibility-aware and cross-modal, derived from LiDAR with explicit modeling of the radar field-of-view and occlusion-aware LiDAR observability via ray-based visibility. Through chirp ablations (A-only, B-only, A+B), range-band analysis, and physics-aligned baselines, we assess how transmit configurations affect geometric recoverability. The results indicate that spatial structure can be learned directly from pre-beamforming per-antenna RD tensors without explicit angle-domain construction or hand-crafted signal-processing stages.

Multimodal emotion recognition in conversations (MERC) requires integrating multimodal signals while being robust to noise and modeling contextual reasoning. Existing approaches often emphasize fusion but overlook uncertainty in noisy features and fine-grained reasoning. We propose SURE (Synergistic Uncertainty-aware REasoning) for MERC, a framework that improves robustness and contextual modeling. SURE consists of three components: an Uncertainty-Aware Mixture-of-Experts module to handle modality-specific noise, an Iterative Reasoning module for multi-turn reasoning over context, and a Transformer Gate module to capture intra- and inter-modal interactions. Experiments on benchmark MERC datasets show that SURE consistently outperforms state-of-the-art methods, demonstrating its effectiveness in robust multimodal reasoning. These results highlight the importance of uncertainty modeling and iterative reasoning in advancing emotion recognition in conversational settings.

Deep reinforcement learning (RL) suffers from plasticity loss severely due to the nature of non-stationarity, which impairs the ability to adapt to new data and learn continually. Unfortunately, our understanding of how plasticity loss arises, dissipates, and can be dissolved remains limited to empirical findings, leaving the theoretical end underexplored.To address this gap, we study the plasticity loss problem from the theoretical perspective of network optimization. By formally characterizing the two culprit factors in online RL process: the non-stationarity of data distributions and the non-stationarity of targets induced by bootstrapping, our theory attributes the loss of plasticity to two mechanisms: the rank collapse of the Neural Tangent Kernel (NTK) Gram matrix and the $Θ(\frac{1}{k})$ decay of gradient magnitude. The first mechanism echoes prior empirical findings from the theoretical perspective and sheds light on the effects of existing methods, e.g., network reset, neuron recycle, and noise injection. Against this backdrop, we focus primarily on the second mechanism and aim to alleviate plasticity loss by addressing the gradient attenuation issue, which is orthogonal to existing methods. We propose Sample Weight Decay -- a lightweight method to restore gradient magnitude, as a general remedy to plasticity loss for deep RL methods based on experience replay. In experiments, we evaluate the efficacy of \methodName upon TD3, \myadded{Double DQN} and SAC with SimBa architecture in MuJoCo, \myadded{ALE} and DeepMind Control Suite tasks. The results demonstrate that \methodName effectively alleviates plasticity loss and consistently improves learning performance across various configurations of deep RL algorithms, UTD, network architectures, and environments, achieving SOTA performance on challenging DMC Humanoid tasks.

Annotated 3D scene data is scarce and expensive to acquire, while abundant unlabeled videos are readily available on the internet. In this paper, we demonstrate that carefully designed data engines can leverage web-curated, unlabeled videos to automatically generate training data, to facilitate end-to-end models in 3D scene understanding alongside human-annotated datasets. We identify and analyze bottlenecks in automated data generation, revealing critical factors that determine the efficiency and effectiveness of learning from unlabeled data. To validate our approach across different perception granularities, we evaluate on three tasks spanning low-level perception, i.e., 3D object detection and instance segmentation, to high-evel reasoning, i.e., 3D spatial Visual Question Answering (VQA) and Vision-Lanugage Navigation (VLN). Models trained on our generated data demonstrate strong zero-shot performance and show further improvement after finetuning. This demonstrates the viability of leveraging readily available web data as a path toward more capable scene understanding systems.

The model context protocol (MCP) standardizes how LLMs connect to external tools and data sources, enabling faster integration but introducing new attack vectors. Despite the growing adoption of MCP, existing MCP security studies classify attacks by their observable effects, obscuring how attacks behave across different MCP server components and overlooking multi-component attack chains. Meanwhile, existing defenses are less effective when facing multi-component attacks or previously unknown malicious behaviors. This work presents a component-centric perspective for understanding and detecting malicious MCP servers. First, we build the first component-centric PoC dataset of 114 malicious MCP servers where attacks are achieved as manipulation over MCP components and their compositions. We evaluate these attacks' effectiveness across two MCP hosts and five LLMs, and uncover that (1) component position shapes attack success rate; and (2) multi-component compositions often outperform single-component attacks by distributing malicious logic. Second, we propose and implement Connor, a two-stage behavioral deviation detector for malicious MCP servers. It first performs pre-execution analysis to detect malicious shell commands and extract each tool's function intent, and then conducts step-wise in-execution analysis to trace each tool's behavioral trajectories and detect deviations from its function intent. Evaluation on our curated dataset indicates that Connor achieves an F1-score of 94.6%, outperforming the state of the art by 8.9% to 59.6%. In real-world detection, Connor identifies two malicious servers.

Data rights owners can detect unauthorized data use in large language model (LLM) training by querying with proprietary samples. Often, superior performance (e.g., higher confidence or lower loss) on a sample relative to the untrained data implies it was part of the training corpus, as LLMs tend to perform better on data they have seen during training. However, this detection becomes fragile under data laundering, a practice of transforming the stylistic form of proprietary data, while preserving critical information to obfuscate data provenance. When an LLM is trained exclusively on such laundered variants, it no longer performs better on originals, erasing the signals that standard detections rely on. We counter this by inferring the unknown laundering transformation from black-box access to the target LLM and, via an auxiliary LLM, synthesizing queries that mimic the laundered data, even if rights owners have only the originals. As the search space of finding true laundering transformations is infinite, we abstract such a process into a high-level transformation goal (e.g., "lyrical rewriting") and concrete details (e.g., "with vivid imagery"), and introduce synthesis data reversion (SDR) that instantiates this abstraction. SDR first identifies the most probable goal for synthesis to narrow the search; it then iteratively refines details so that synthesized queries gradually elicit stronger detection signals from the target LLM. Evaluated on the MIMIR benchmark against diverse laundering practices and target LLM families (Pythia, Llama2, and Falcon), SDR consistently strengthens data misuse detection, providing a practical countermeasure to data laundering.

Artificial intelligence (AI) systems accelerate medical workflows and improve diagnostic accuracy in healthcare, serving as second-opinion systems. However, the unpredictability of AI errors poses a significant challenge, particularly in healthcare contexts, where mistakes can have severe consequences. A widely adopted safeguard is to pair predictions with uncertainty estimation, enabling human experts to focus on high-risk cases while streamlining routine verification. Current uncertainty estimation methods, however, remain limited, particularly in quantifying aleatoric uncertainty, which arises from data ambiguity and noise. To address this, we propose a novel approach that leverages disagreement in expert responses to generate targets for training machine learning models. These targets are used in conjunction with standard data labels to estimate two components of uncertainty separately, as given by the law of total variance, via a two-ensemble approach, as well as its lightweight variant. We validate our method on binary image classification, binary and multi-class image segmentation, and multiple-choice question answering. Our experiments demonstrate that incorporating expert knowledge can enhance uncertainty estimation quality by $9\%$ to $50\%$ depending on the task, making this source of information invaluable for the construction of risk-aware AI systems in healthcare applications.

Large language models (LLMs) have been proposed as alternatives to human experts for estimating unknown quantities with associated uncertainty, a process known as Bayesian elicitation. We test this by asking eleven LLMs to estimate population statistics, such as health prevalence rates, personality trait distributions, and labor market figures, and to express their uncertainty as 95\% credible intervals. We vary each model's reasoning effort (low, medium, high) to test whether more "thinking" improves results. Our findings reveal three key results. First, larger, more capable models produce more accurate estimates, but increasing reasoning effort provides no consistent benefit. Second, all models are severely overconfident: their 95\% intervals contain the true value only 9--44\% of the time, far below the expected 95\%. Third, a statistical recalibration technique called conformal prediction can correct this overconfidence, expanding the intervals to achieve the intended coverage. In a preliminary experiment, giving models web search access degraded predictions for already-accurate models, while modestly improving predictions for weaker ones. Models performed well on commonly discussed topics but struggled with specialized health data. These results indicate that LLM uncertainty estimates require statistical correction before they can be used in decision-making.

Electroencephalography (EEG) provides a non-invasive window into brain activity, offering high temporal resolution crucial for understanding and interacting with neural processes through brain-computer interfaces (BCIs). Current dual-stream neural networks for EEG often process temporal and spatial features independently through parallel branches, delaying their integration until a final, late-stage fusion. This design inherently leads to an "information silo" problem, precluding intermediate cross-stream refinement and hindering spatial-temporal decompositions essential for full feature utilization. We propose LI-DSN, a layer-wise interactive dual-stream network that facilitates progressive, cross-stream communication at each layer, thereby overcoming the limitations of late-fusion paradigms. LI-DSN introduces a novel Temporal-Spatial Integration Attention (TSIA) mechanism, which constructs a Spatial Affinity Correlation Matrix (SACM) to capture inter-electrode spatial structural relationships and a Temporal Channel Aggregation Matrix (TCAM) to integrate cosine-gated temporal dynamics under spatial guidance. Furthermore, we employ an adaptive fusion strategy with learnable channel weights to optimize the integration of dual-stream features. Extensive experiments across eight diverse EEG datasets, encompassing motor imagery (MI) classification, emotion recognition, and steady-state visual evoked potentials (SSVEP), consistently demonstrate that LI-DSN significantly outperforms 13 state-of-the-art (SOTA) baseline models, showcasing its superior robustness and decoding performance. The code will be publicized after acceptance.

Video Large Language Models (VideoLLMs) have demonstrated impressive capabilities in video understanding, yet the massive number of input video tokens incurs a significant computational burden for deployment. Existing methods mainly prune video tokens at input level while neglecting the inherent information structure embedded in videos and large language models (LLMs). To address this, we propose HieraVid, a hierarchical pruning framework that progressively and dynamically reduces visual redundancy. Based on two observations that videos possess the segment-frame structure and LLMs internally propagate multi-modal information unidirectionally, we decompose pruning into three levels: 1) segment-level, where video tokens are first temporally segmented and spatially merged; 2) frame-level, where similar frames within the same segment are jointly pruned to preserve diversity; 3) layer-level, redundancy gradually shrinks as LLM layer increases w/o compromising performance. We conduct extensive experiments on four widely used video understanding benchmarks to comprehensively evaluate the effectiveness of HieraVid. Remarkably, with only 30% of tokens retained, HieraVid achieves new state-of-the-art performance, while maintaining over 98% and 99% of the performance of LLaVA-Video-7B and LLaVA-OneVision-7B, respectively.

Modern neural networks of the transformer family require the practitioner to decide, before training begins, how many attention heads to use, how deep the network should be, and how wide each component should be. These decisions are made without knowledge of the task, producing architectures that are systematically larger than necessary: empirical studies find that a substantial fraction of heads and layers can be removed after training without performance loss. This paper introduces DDCL-INCRT, an architecture that determines its own structure during training. Two complementary ideas are combined. The first, DDCL (Deep Dual Competitive Learning), replaces the feedforward block with a dictionary of learned prototype vectors representing the most informative directions in the data. The prototypes spread apart automatically, driven by the training objective, without explicit regularisation. The second, INCRT (Incremental Transformer), controls the number of heads: starting from one, it adds a new head only when the directional information uncaptured by existing heads exceeds a threshold. The main theoretical finding is that these two mechanisms reinforce each other: each new head amplifies prototype separation, which in turn raises the signal triggering the next addition. At convergence, the network self-organises into a hierarchy of heads ordered by representational granularity. This hierarchical structure is proved to be unique and minimal, the smallest architecture sufficient for the task, under the stated conditions. Formal guarantees of stability, convergence, and pruning safety are established throughout. The architecture is not something one designs. It is something one derives.

Spectral graph contrastive learning has emerged as a unified paradigm for handling both homophilic and heterophilic graphs by leveraging high-frequency components. However, we identify a fundamental spectral dilemma: while high-frequency signals are indispensable for encoding heterophily, our theoretical analysis proves they exhibit significantly higher variance under spectrally concentrated perturbations. We derive a regret lower bound showing that existing global (node-agnostic) spectral fusion is provably sub-optimal: on mixed graphs with separated node-wise frequency preferences, any global fusion strategy incurs non-vanishing regret relative to a node-wise oracle. To escape this bound, we propose ASPECT, a framework that resolves this dilemma through a reliability-aware spectral gating mechanism. Formulated as a minimax game, ASPECT employs a node-wise gate that dynamically re-weights frequency channels based on their stability against a purpose-built adversary, which explicitly targets spectral energy distributions via a Rayleigh quotient penalty. This design forces the encoder to learn representations that are both structurally discriminative and spectrally robust. Empirical results show that ASPECT achieves new state-of-the-art performance on 8 out of 9 benchmarks, effectively decoupling meaningful structural heterophily from incidental noise.

In modern process industries, data-driven models are important tools for real-time monitoring when key performance indicators are difficult to measure directly. While accurate predictions are essential, reliable uncertainty quantification (UQ) is equally critical for safety, reliability, and decision-making, but remains a major challenge in current data-driven approaches. In this work, we introduce a diffusion-based posterior sampling framework that inherently produces well-calibrated predictive uncertainty via faithful posterior sampling, eliminating the need for post-hoc calibration. In extensive evaluations on synthetic distributions, the Raman-based phenylacetic acid soft sensor benchmark, and a real ammonia synthesis case study, our method achieves practical improvements over existing UQ techniques in both uncertainty calibration and predictive accuracy. These results highlight diffusion samplers as a principled and scalable paradigm for advancing uncertainty-aware modeling in industrial applications.

Stochastic Shortest Path problems (SSPs) are traditionally solved by computing each state's cost-to-go by applying Bellman backups. A Bellman backup updates a state's cost-to-go by iterating through every applicable action, computing the cost-to-go after applying each one, and selecting a minimal action's cost-to-go. State-of-the-art algorithms use heuristic functions; these give an initial estimate of costs-to-go, and lets the algorithm apply Bellman backups only to promising states, determined by low estimated costs-to-go. However, each Bellman backup still considers all applicable actions, even if the heuristic tells us that some of these actions are too expensive, with the effect that such algorithms waste time on unhelpful actions. To address this gap we present a technique that uses the heuristic to avoid expensive actions, by reframing heuristic search in terms of linear programming and introducing an efficient implementation of constraint generation for SSPs. We present CG-iLAO*, a new algorithm that adapts iLAO* with our novel technique, and considers only 40% of iLAO*'s actions on many problems, and as few as 1% on some. Consequently, CG-iLAO* computes on average 3.5x fewer costs-to-go for actions than the state-of-the-art iLAO* and LRTDP, enabling it to solve problems faster an average of 2.8x and 3.7x faster, respectively.

Dyslexic spelling errors exhibit systematic phonological and orthographic patterns that distinguish them from the errors produced by typically developing writers. While this observation has motivated dyslexic-specific spell-checking and assistive writing tools, prior work has focused predominantly on error correction rather than attribution, and has largely neglected the ethical risks. The risk of harmful labelling, covert screening, algorithmic bias, and institutional misuse that automated classification of learners entails requires the development of robust ethical and legal frameworks for research in this area. This paper addresses both gaps. We formulate dyslexic error attribution as a binary classification task. Given a misspelt word and its correct target form, determine whether the error pattern is characteristic of a dyslexic or non-dyslexic writer. We develop a comprehensive feature set capturing orthographic, phonological, and morphological properties of each error, and propose a twin-input neural model evaluated against traditional machine learning baselines under writer-independent conditions. The neural model achieves 93.01% accuracy and an F1-score of 94.01%, with phonetically plausible errors and vowel confusions emerging as the strongest attribution signals. We situate these technical results within an explicit ethics-first framework, analysing fairness across subgroups, the interpretability requirements of educational deployment, and the conditions, consent, transparency, human oversight, and recourse, under which a system could be responsibly used. We provide concrete guidelines for ethical deployment and an open discussion of the systems limitations and misuse potential. Our results demonstrate that dyslexic error attribution is feasible at high accuracy while underscoring that feasibility alone is insufficient for deployment in high-stakes educational contexts.

In the digital age, ensuring the correctness, safety, and reliability of software through formal verification is paramount, particularly as software increasingly underpins critical infrastructure. Formal verification, split into theorem proving and model checking, provides a feasible and reliable path. Unlike theorem proving, which yields notable advances, model checking has been less focused due to the difficulty of automatic program modeling. To fill this gap, we introduce Model-Bench, a benchmark and an accompanying pipeline for evaluating and improving LLMs' program modeling capability by modeling Python programs into verification-ready model checking specifications checkable by its accompanying model checker. Model-Bench comprises 400 Python programs derived from three well-known benchmarks (HumanEval, MBPP, and LiveCodeBench). Our extensive experiments reveal significant limitations in LLMs' program modeling and further provide inspiring directions.

While Large Language Models (LLMs) have demonstrated exceptional proficiency in code completion, they typically adhere to a Hard Completion (HC) paradigm, compelling the generation of fully concrete code even amidst insufficient context. Our analysis of 3 million real-world interactions exposes the limitations of this strategy: 61% of the generated suggestions were either edited after acceptance or rejected despite exhibiting over 80% similarity to the user's subsequent code, suggesting that models frequently make erroneous predictions at specific token positions. Motivated by this observation, we propose Adaptive Placeholder Completion (APC), a collaborative framework that extends HC by strategically outputting explicit placeholders at high-entropy positions, allowing users to fill directly via IDE navigation. Theoretically, we formulate code completion as a cost-minimization problem under uncertainty. Premised on the observation that filling placeholders incurs lower cost than correcting errors, we prove the existence of a critical entropy threshold above which APC achieves strictly lower expected cost than HC. We instantiate this framework by constructing training data from filtered real-world edit logs and design a cost-based reward function for reinforcement learning. Extensive evaluations across 1.5B--14B parameter models demonstrate that APC reduces expected editing costs from 19% to 50% while preserving standard HC performance. Our work provides both a theoretical foundation and a practical training framework for uncertainty-aware code completion, demonstrating that adaptive abstention can be learned end-to-end without sacrificing conventional completion quality.

Multivariate time-series anomaly detection (MTSAD) aims to identify deviations from normality in multivariate time-series and is critical in real-world applications. However, in real-world deployments, distribution shifts are ubiquitous and cause severe performance degradation in pre-trained anomaly detector. Test-time adaptation (TTA) updates a pre-trained model on-the-fly using only unlabeled test data, making it promising for addressing this challenge. In this study, we propose CANDI (Curated test-time adaptation for multivariate time-series ANomaly detection under DIstribution shift), a novel TTA framework that selectively identifies and adapts to potential false positives while preserving pre-trained knowledge. CANDI introduces a False Positive Mining (FPM) strategy to curate adaptation samples based on anomaly scores and latent similarity, and incorporates a plug-and-play Spatiotemporally-Aware Normality Adaptation (SANA) module for structurally informed model updates. Extensive experiments demonstrate that CANDI significantly improves the performance of MTSAD under distribution shift, improving AUROC up to 14% while using fewer adaptation samples.

Vector quantization approaches (VQ-VAE, VQ-GAN) learn discrete neural representations of images, but these representations are inherently position-dependent: codes are spatially arranged and contextually entangled, requiring autoregressive or diffusion-based priors to model their dependencies at sample time. In this work, we ask whether positional information is necessary for discrete representations of spatially aligned data. We propose the permutation-invariant vector-quantized autoencoder (PI-VQ), in which latent codes are constrained to carry no positional information. We find that this constraint encourages codes to capture global, semantic features, and enables direct interpolation between images without a learned prior. To address the reduced information capacity of permutation-invariant representations, we introduce matching quantization, a vector quantization algorithm based on optimal bipartite matching that increases effective bottleneck capacity by $3.5\times$ relative to naive nearest-neighbour quantization. The compositional structure of the learned codes further enables interpolation-based sampling, allowing synthesis of novel images in a single forward pass. We evaluate PI-VQ on CelebA, CelebA-HQ and FFHQ, obtaining competitive precision, density and coverage metrics for images synthesised with our approach. We discuss the trade-offs inherent to position-free representations, including separability and interpretability of the latent codes, pointing to numerous directions for future work.

Clinical prediction from structured electronic health records (EHRs) is challenging due to high dimensionality, heterogeneity, class imbalance, and distribution shift. While tabular in-context learning (TICL) and retrieval-augmented methods perform well on generic benchmarks, their behavior in clinical settings remains unclear. We present a multi-cohort EHR benchmark comparing classical, deep tabular, and TICL models across varying data scale, feature dimensionality, outcome rarity, and cross-cohort generalization. PFN-based TICL models are sample-efficient in low-data regimes but degrade under naive distance-based retrieval as heterogeneity and imbalance increase. We propose AWARE, a task-aligned retrieval framework using supervised embedding learning and lightweight adapters. AWARE improves AUPRC by up to 12.2% under extreme imbalance, with gains increasing with data complexity. Our results identify retrieval quality and retrieval-inference alignment as key bottlenecks for deploying tabular in-context learning in clinical prediction.

While Reinforcement Learning from Verifiable Rewards (RLVR) has advanced reasoning in Large Vision-Language Models (LVLMs), prevailing frameworks suffer from a foundational methodological flaw: by distributing identical advantages across all generated tokens, these methods inherently dilute the learning signals essential for optimizing the critical, visually-grounded steps of multimodal reasoning. To bridge this gap, we formulate \textit{Token Visual Dependency}, quantifying the causal information gain of visual inputs via the Kullback-Leibler (KL) divergence between visual-conditioned and text-only predictive distributions. Revealing that this dependency is highly sparse and semantically pivotal, we introduce Perception-Grounded Policy Optimization (PGPO), which is a novel fine-grained credit assignment framework that dynamically reshapes advantages at the token level. Through a threshold-gated, mass-conserving mechanism, PGPO actively amplifies learning signals for visually-dependent tokens while suppressing gradient noise from linguistic priors. Extensive experiments based on the Qwen2.5-VL series across seven challenging multimodal reasoning benchmarks demonstrate that PGPO boosts models by 18.7% on average. Both theoretical and empirical analyses confirm that PGPO effectively reduces gradient variance, prevents training collapse, and acts as a potent regularizer for robust, perception-grounded multimodal reasoning. Code will be published on https://github.com/Yzk1114/PGPO.

Preference learning in Large Language Models (LLMs) has advanced significantly, yet existing methods remain limited by modest performance gains, high computational costs, hyperparameter sensitivity, and insufficient modeling of global token-level relationships. We introduce PLOT, which enhances Preference Learning in fine-tuning-based alignment through a token-level loss derived from Optimal Transport. By formulating preference learning as an Optimal Transport Problem, PLOT aligns model outputs with human preferences while preserving the original distribution of LLMs, ensuring stability and robustness. Furthermore, PLOT leverages token embeddings to capture semantic relationships, enabling globally informed optimization. Experiments across two preference categories - Human Values and Logic & Problem Solving - spanning seven subpreferences demonstrate that PLOT consistently improves alignment performance while maintaining fluency and coherence. These results substantiate optimal transport as a principled methodology for preference learning, establishing a theoretically grounded framework that provides new insights for preference learning of LLMs.

The ratio of outlier parameters in language pre-training models and vision pre-training models differs significantly, making cross-modality (language and vision) inherently more challenging than cross-domain adaptation. As a result, many prior studies have focused on cross-domain transfer rather than attempting to bridge language and vision modalities, assuming that language pre-trained models are unsuitable for downstream visual tasks due to disparate parameter spaces. Contrary to this assumption, we show that adding a bridge training stage as a modality adaptation learner can effectively align Large Language Model (LLM) parameters with vision tasks. Specifically, we propose a simple yet powerful solution random label bridge training that requires no manual labeling and helps LLM parameters adapt to vision foundation tasks. Moreover, our findings reveal that partial bridge training is often advantageous, as certain layers in LLMs exhibit strong foundational properties that remain beneficial even without fine-tuning for visual tasks. This surprising discovery opens up new avenues for leveraging language pre-trained parameters directly within vision models and highlights the potential of partial bridge training as a practical pathway to cross-modality adaptation.

Topology control for power grid operation is a challenging sequential decision making problem because the action space grows combinatorially with the size of the grid and action evaluation through simulation is computationally expensive. We propose a physics-informed Reinforcement Learning framework that combines semi-Markov control with a Gibbs prior, that encodes the system's physics, over the action space. The decision is only taken when the grid enters a hazardous regime, while a graph neural network surrogate predicts the post action overload risk of feasible topology actions. These predictions are used to construct a physics-informed Gibbs prior that both selects a small state-dependent candidate set and reweights policy logits before action selection. In this way, our method reduces exploration difficulty and online simulation cost while preserving the flexibility of a learned policy. We evaluate the approach in three realistic benchmark environments of increasing difficulty. Across all settings, the proposed method achieves a strong balance between control quality and computational efficiency: it matches oracle-level performance while being approximately $6\times$ faster on the first benchmark, reaches $94.6\%$ of oracle reward with roughly $200\times$ lower decision time on the second one, and on the most challenging benchmark improves over a PPO baseline by up to $255\%$ in reward and $284\%$ in survived steps while remaining about $2.5\times$ faster than a strong specialized engineering baseline. These results show that our method provides an effective mechanism for topology control in power grids.

In Europe, balance responsible parties can deliberately take out-of-balance positions to support transmission system operators (TSOs) in maintaining grid stability and earn profit, a practice called implicit balancing. Model predictive control (MPC) is widely adopted as an effective approach for implicit balancing. The balancing market model accuracy in MPC is critical to decision quality. Previous studies modeled this market using either (i) a convex market clearing approximation, ignoring proactive manual actions by TSOs and the market sub-quarter-hour dynamics, or (ii) machine learning methods, which cannot be directly integrated into MPC. To address these shortcomings, we propose a data-driven balancing market model integrated into MPC using an input convex neural network to ensure convexity while capturing uncertainties. To keep the core network computationally efficient, we incorporate attention-based input gating mechanisms to remove irrelevant data. Evaluating on Belgian data shows that the proposed model both improves MPC decisions and reduces computational time.

Accurate sensing of spatially distributed physical fields typically requires dense instrumentation, which is often infeasible in real-world systems due to cost, accessibility, and environmental constraints. Physics-based solvers address this through direct numerical integration of governing equations, but their computational latency and power requirements preclude real-time use in resource-constrained monitoring and control systems. Here we introduce VIRSO (Virtual Irregular Real-Time Sparse Operator), a graph-based neural operator for sparse-to-dense reconstruction on irregular geometries, and a variable-connectivity algorithm, Variable KNN (V-KNN), for mesh-informed graph construction. Unlike prior neural operators that treat hardware deployability as secondary, VIRSO reframes inference as measurement: the combination of both spectral and spatial analysis provides accurate reconstruction without the high latency and power consumption of previous graph-based methodologies with poor scalability, presenting VIRSO as a potential candidate for edge-constrained, real-time virtual sensing. We evaluate VIRSO on three nuclear thermal-hydraulic benchmarks of increasing geometric and multiphysics complexity, across reconstruction ratios from 47:1 to 156:1. VIRSO achieves mean relative $L_2$ errors below 1%, outperforming other benchmark operators while using fewer parameters. The full 10-layer configuration reduces the energy-delay product (EDP) from ${\approx}206$ J$\cdot$ms for the graph operator baseline to $10.1$ J$\cdot$ms on an NVIDIA H200. Implemented on an NVIDIA Jetson Orin Nano, all configurations of VIRSO provide sub-10 W power consumption and sub-second latency. These results establish the edge-feasibility and hardware-portability of VIRSO and present compute-aware operator learning as a new paradigm for real-time sensing in inaccessible and resource-constrained environments.

With the rapid evolution of LLMs, automated software testing is witnessing a paradigm shift. While proprietary models like GPT-4o demonstrate impressive capabilities, their high deployment costs and data privacy concerns make open-source LLMs the practical imperative for many academic and industrial scenarios. In the field of automated test generation, it has evolved to iterative workflows to construct test suites based on LLMs. When utilizing open-source LLMs, we empirically observe they lack a suite-level perspective, suffering from structural myopia-failing to generate new tests with large marginal gain based on the current covered status. In this paper, from the perspective of sequences, we formalize test suite generation as a MDP and demonstrate that its objective exhibits monotone submodularity, which enables an effective relaxation of this NP-hard global optimization into a tractable step-wise greedy procedure. Guided by this insight, we propose TestDecision, which transforms LLMs into neural greedy experts. TestDecision consists of two synergistic components: (1) an inference framework which implements test suite construction following a step-wise greedy strategy; and (2) a training pipeline of reinforcement learning which equips the base LLM with sequential test generation ability to maximize marginal gain. Comprehensive evaluations on the ULT benchmark demonstrate that TestDecision significantly outperforms existing advanced methods. It brings an improvement between 38.15-52.37% in branch coverage and 298.22-558.88% in execution pass rate over all base models, achieving a comparable performance on 7B backbone with a much larger proprietary LLM GPT-5.2. Furthermore, TestDecision can find 58.43-95.45% more bugs than vanilla base LLMs and exhibit superior generalization on LiveCodeBench, proving its capability to construct high-quality test suites.

Breast cancer is a highly heterogeneous disease with diverse molecular profiles. The PAM50 gene signature is widely recognized as a standard for classifying breast cancer into intrinsic subtypes, enabling more personalized treatment strategies. In this study, we introduce a novel optimization-driven deep learning framework that aims to reduce reliance on costly molecular assays by directly predicting PAM50 subtypes from H&E-stained whole-slide images (WSIs). Our method jointly optimizes patch informativeness, spatial diversity, uncertainty, and patch count by combining the non-dominated sorting genetic algorithm II (NSGA-II) with Monte Carlo dropout-based uncertainty estimation. The proposed method can identify a small but highly informative patch subset for classification. We used a ResNet18 backbone for feature extraction and a custom CNN head for classification. For evaluation, we used the internal TCGA-BRCA dataset as the training cohort and the external CPTAC-BRCA dataset as the test cohort. On the internal dataset, an F1-score of 0.8812 and an AUC of 0.9841 using 627 WSIs from the TCGA-BRCA cohort were achieved. The performance of the proposed approach on the external validation dataset showed an F1-score of 0.7952 and an AUC of 0.9512. These findings indicate that the proposed optimization-guided, uncertainty-aware patch selection can achieve high performance and improve the computational efficiency of histopathology-based PAM50 classification compared to existing methods, suggesting a scalable imaging-based replacement that has the potential to support clinical decision-making.

We study the prophet inequality, a fundamental problem in online decision-making and optimal stopping, in a practical setting where rewards are observed only through noisy realizations and reward distributions are unknown. At each stage, the decision-maker receives a noisy reward whose true value follows a linear model with an unknown latent parameter, and observes a feature vector drawn from a distribution. To address this challenge, we propose algorithms that integrate learning and decision-making via lower-confidence-bound (LCB) thresholding. In the i.i.d.\ setting, we establish that both an Explore-then-Decide strategy and an $\varepsilon$-Greedy variant achieve the sharp competitive ratio of $1 - 1/e$, under a mild condition on the optimal value. For non-identical distributions, we show that a competitive ratio of $1/2$ can be guaranteed against a relaxed benchmark. Moreover, with limited window access to past rewards, the tight ratio of $1/2$ against the optimal benchmark is achieved.

Reinforcement Learning from Human Feedback (RLHF), using algorithms like Proximal Policy Optimization (PPO), aligns Large Language Models (LLMs) with human values but is costly and unstable. Alternatives have been proposed to replace PPO or integrate Supervised Fine-Tuning (SFT) and contrastive learning for direct fine-tuning and value alignment. However, these methods still require voluminous data to learn preferences and may weaken the generalization ability of LLMs. To further enhance alignment efficiency and performance while mitigating the loss of generalization ability, this paper introduces Distribution-guided Efficient Fine-Tuning (DEFT), an efficient alignment framework incorporating data filtering and distributional guidance by calculating the differential distribution reward based on the output distribution of language model and the discrepancy distribution of preference data. A small yet high-quality subset is filtered from the raw data using a differential distribution reward, which is then incorporated into existing alignment methods to guide the model's output distribution. Experimental results demonstrate that the methods enhanced by DEFT outperform the original methods in both alignment capability and generalization ability, with significantly reduced training time.

Controllable Automatic Text Simplification (CATS) produces user-tailored outputs, yet controllability is often treated as a decoding problem and evaluated with metrics that are not reflective to the measure of control. We observe that controllability in ATS is significantly constrained by data and evaluation. To this end, we introduce a domain-agnostic CATS framework based on instruction fine-tuning with discrete control tokens, steering open-source models to target readability levels and compression rates. Across three model families with different model sizes (Llama, Mistral, Qwen; 1-14B) and four domains (medicine, public administration, news, encyclopedic text), we find that smaller models (1-3B) can be competitive, but reliable controllability strongly depends on whether the training data encodes sufficient variation in the target attribute. Readability control (FKGL, ARI, Dale-Chall) is learned consistently, whereas compression control underperforms due to limited signal variability in the existing corpora. We further show that standard simplification and similarity metrics are insufficient for measuring control, motivating error-based measures for target-output alignment. Finally, our sampling and stratification experiments demonstrate that naive splits can introduce distributional mismatch that undermines both training and evaluation.

Although demand forecasting is a critical component of supply chain planning, actual retail data can exhibit irreconcilable seasonality, irregular spikes, and noise, rendering precise projections nearly unattainable. This paper proposes a three-step analytical framework that combines forecasting and operational analytics. The first stage consists of exploratory data analysis, where delivery-tracked data from 180,519 transactions are partitioned, and long-term trends, seasonality, and delivery-related attributes are examined. Secondly, the forecasting performance of a statistical time series decomposition model N-BEATS MSTL and a recent deep learning architecture N-HiTS were compared. N-BEATS and N-HiTS were both statistically, and hence were N-BEATS's and N-HiTS's statistically selected. Most recent time series deep learning models, N-HiTS, N-BEATS. N-HiTS and N-BEATS N-HiTS and N-HiTS outperformed the statistical benchmark to a large extent. N-BEATS was selected to be the most optimized model, as the one with the lowest forecasting error, in the 3rd and final stage forecasting values of the next 4 weeks of 1918 units, and provided those as a model with a set of deterministically integer linear program outcomes that are aimed to minimize the total delivery time with a set of bound budget, capacity, and service constraints. The solution allocation provided a feasible and cost-optimal shipping plan. Overall, the study provides a compelling example of the practical impact of precise forecasting and simple, highly interpretable model optimization in logistics.

Knowledge graphs store large numbers of relations efficiently, but they remain weak at representing a quieter difficulty: the meaning of a concept often shifts with the domain in which it is used. A triple such as Apple, instance-of, Company may be acceptable in one setting while being misleading or unusable in another. In most current systems, domain information is attached as metadata, qualifiers, or graph-level organization. These mechanisms help with filtering and provenance, but they usually do not alter the formal status of the assertion itself. This paper argues that domain should be treated as part of knowledge representation rather than as supplementary annotation. It introduces the Domain-Contextualized Concept Graph (DCG), a framework in which domain is written into the relation and interpreted as a modal world constraint. In the DCG form (C, R at D, C'), the marker at D identifies the world in which the relation holds. Formally, the relation is interpreted through a domain-indexed necessity operator, so that truth, inference, and conflict checking are all scoped to the relevant world. This move has three consequences: ambiguous concepts can be disambiguated at the point of representation; invalid assertions can be challenged against their domain; cross-domain relations can be connected through explicit predicates. The paper develops this claim through a Kripke-style semantics, a compact predicate system, a Prolog implementation, and mappings to RDF, OWL, and relational databases. The contribution is a representational reinterpretation of domain itself. The central claim is that many practical failures in knowledge systems begin when domain is treated as external to the assertion. DCG addresses that by giving domain a structural and computable role inside the representation.

For multi-input and multi-output (MIMO) channels, the optimal channel estimation (CE) based on linear minimum mean square error (LMMSE) requires three-dimensional (3D) filtering. However, the complexity is often prohibitive due to large matrix dimensions. Suboptimal estimators approximate 3DCE by decomposing it into time, frequency, and spatial domains, while yields noticeable performance degradation under correlated MIMO channels. On the other hand, recent advances in deep learning (DL) can explore channel correlations in all domains via attention mechanisms. Building on this capability, we propose a dual attention mechanism based 3DCE network (3DCENet) that can achieve accurate estimates.

Very High Resolution (VHR) forest structure data at individual-tree scale is essential for carbon, biodiversity, and ecosystem monitoring. Still, airborne LiDAR remains costly and infrequent despite being the reference for forest structure metrics like Canopy Height Model (CHM), Plant Area Index (PAI), and Foliage Height Diversity (FHD). We propose FSKD: a LiDAR-to-RGB-Infrared (RGBI) knowledge distillation (KD) framework in which a multi-modal teacher fuses RGBI imagery with LiDAR-derived planar metrics and vertical profiles via cross-attention, and an RGBI-only SegFormer student learns to reproduce these outputs. Trained on 384 $km^2$ of forests in Saxony, Germany (20 cm ground sampling distance (GSD)) and evaluated on eight geographically distinct test tiles, the student achieves state-of-the-art (SOTA) zero-shot CHM performance (MedAE 4.17 m, $R^2$=0.51, IoU 0.87), outperforming HRCHM/DAC baselines by 29--46% in MAE (5.81 m vs. 8.14--10.84 m) with stronger correlation coefficients (0.713 vs. 0.166--0.652). Ablations show that multi-modal fusion improves performance by 10--26% over RGBI-only training, and that asymmetric distillation with appropriate model capacity is critical. The method jointly predicts CHM, PAI, and FHD, a multi-metric capability not provided by current monocular CHM estimators, although PAI/FHD transfer remains region-dependent and benefits from local calibration. The framework also remains effective under temporal mismatch (winter LiDAR, summer RGBI), removing strict co-acquisition constraints and enabling scalable 20 cm operational monitoring for workflows such as Digital Twin Germany and national Digital Orthophoto programs.

Recently, world-action models (WAM) have emerged to bridge vision-language-action (VLA) models and world models, unifying their reasoning and instruction-following capabilities and spatio-temporal world modeling. However, existing WAM approaches often focus on modeling 2D appearance or latent representations, with limited geometric grounding-an essential element for embodied systems operating in the physical world. We present DriveDreamer-Policy, a unified driving world-action model that integrates depth generation, future video generation, and motion planning within a single modular architecture. The model employs a large language model to process language instructions, multi-view images, and actions, followed by three lightweight generators that produce depth, future video, and actions. By learning a geometry-aware world representation and using it to guide both future prediction and planning within a unified framework, the proposed model produces more coherent imagined futures and more informed driving actions, while maintaining modularity and controllable latency. Experiments on the Navsim v1 and v2 benchmarks demonstrate that DriveDreamer-Policy achieves strong performance on both closed-loop planning and world generation tasks. In particular, our model reaches 89.2 PDMS on Navsim v1 and 88.7 EPDMS on Navsim v2, outperforming existing world-model-based approaches while producing higher-quality future video and depth predictions. Ablation studies further show that explicit depth learning provides complementary benefits to video imagination and improves planning robustness.

Parameter-efficient fine-tuning (PEFT) has emerged as a crucial paradigm for adapting large language models (LLMs) under constrained computational budgets. However, standard PEFT methods often struggle in multi-task fine-tuning settings, where diverse optimization objectives induce task interference and limited parameter budgets lead to representational deficiency. While recent approaches incorporate mixture-of-experts (MoE) to alleviate these issues, they predominantly operate in the spatial domain, which may introduce structural redundancy and parameter overhead. To overcome these limitations, we reformulate adaptation in the spectral domain. Our spectral analysis reveals that different tasks exhibit distinct frequency energy distributions, and that LLM layers display heterogeneous frequency sensitivities. Motivated by these insights, we propose FourierMoE, which integrates the MoE architecture with the inverse discrete Fourier transform (IDFT) for frequency-aware adaptation. Specifically, FourierMoE employs a frequency-adaptive router to dispatch tokens to experts specialized in distinct frequency bands. Each expert learns a set of conjugate-symmetric complex coefficients, preserving complete phase and amplitude information while theoretically guaranteeing lossless IDFT reconstruction into real-valued spatial weights. Extensive evaluations across 28 benchmarks, multiple model architectures, and scales demonstrate that FourierMoE consistently outperforms competitive baselines in both single-task and multi-task settings while using significantly fewer trainable parameters. These results highlight the promise of spectral-domain expert adaptation as an effective and parameter-efficient paradigm for LLM fine-tuning.

REST APIs are widely used in industry, in all different kinds of domains. An example is Volkswagen AG, a German automobile manufacturer. Established testing approaches for REST APIs are time consuming, and require expertise from professional test engineers. Due to its cost and importance, in the scientific literature several approaches have been proposed to automatically test REST APIs. The open-source, search-based fuzzer EvoMaster is one of such tools proposed in the academic literature. However, how academic prototypes can be integrated in industry and have real impact to software engineering practice requires more investigation. In this paper, we report on our experience in using EvoMaster at Volkswagen AG, as an EvoMaster user from 2023 to 2026. We share our learnt lessons, and discuss several features needed to be implemented in EvoMaster to make its use in an industrial context successful. Feedback about value in industrial setups of EvoMaster was given from Volkswagen AG about 4 APIs. Additionally, a user study was conducted involving 11 testing specialists from 4 different companies. We further identify several real-world research challenges that still need to be solved.

Mathematical reasoning is a hallmark of human intelligence, and whether large language models (LLMs) can meaningfully perform it remains a central question in artificial intelligence and cognitive science. As LLMs are increasingly integrated into scientific workflows, rigorous evaluation of their mathematical capabilities becomes a practical necessity. Existing benchmarks are limited by synthetic settings and data contamination. We present LiveMathematicianBench, a dynamic multiple-choice benchmark for research-level mathematical reasoning built from recent arXiv papers published after model training cutoffs. By grounding evaluation in newly published theorems, it provides a realistic testbed beyond memorized patterns. The benchmark introduces a thirteen-category logical taxonomy of theorem types (e.g., implication, equivalence, existence, uniqueness), enabling fine-grained evaluation across reasoning forms. It employs a proof-sketch-guided distractor pipeline that uses high-level proof strategies to construct plausible but invalid answer choices reflecting misleading proof directions, increasing sensitivity to genuine understanding over surface-level matching. We also introduce a substitution-resistant mechanism to distinguish answer recognition from substantive reasoning. Evaluation shows the benchmark is far from saturated: Gemini-3.1-pro-preview, the best model, achieves only 43.5%. Under substitution-resistant evaluation, accuracy drops sharply: GPT-5.4 scores highest at 30.6%, while Gemini-3.1-pro-preview falls to 17.6%, below the 20% random baseline. A dual-mode protocol reveals that proof-sketch access yields consistent accuracy gains, suggesting models can leverage high-level proof strategies for reasoning. Overall, LiveMathematicianBench offers a scalable, contamination-resistant testbed for studying research-level mathematical reasoning in LLMs.

Toxic content detection in online communication remains a significant challenge, with current solutions often inadvertently blocking valuable information, including medical terms and text related to minority groups. This paper presents a more nu-anced approach to identifying toxicity in Bulgarian text while preserving access to essential information. The research explores two distinct methodologies for detecting toxic content. The developed methodologies have po-tential applications across diverse online platforms and content moderation systems. First, we propose an ontology that models the potentially toxic words in Bulgarian language. Then, we compose a dataset that comprises 4,384 manually anno-tated sentences from Bulgarian online forums across four categories: toxic language, medical terminology, non-toxic lan-guage, and terms related to minority communities. We then train a BERT-based model for toxic language classification, which reaches a 0.89 F1 macro score. The trained model is directly applicable in a real environment and can be integrated as a com-ponent of toxic content detection systems.

A persistent structural weakness in deep clustering is the disconnect between feature learning and cluster assignment. Most architectures invoke an external clustering step, typically k-means, to produce pseudo-labels that guide training, preventing the backbone from directly optimising for cluster quality. This paper introduces Deep Dual Competitive Learning (DDCL), the first fully differentiable end-to-end framework for unsupervised prototype-based representation learning. The core contribution is architectural: the external k-means is replaced by an internal Dual Competitive Layer (DCL) that generates prototypes as native differentiable outputs of the network. This single inversion makes the complete pipeline, from backbone feature extraction through prototype generation to soft cluster assignment, trainable by backpropagation through a single unified loss, with no Lloyd iterations, no pseudo-label discretisation, and no external clustering step. To ground the framework theoretically, the paper derives an exact algebraic decomposition of the soft quantisation loss into a simplex-constrained reconstruction error and a non-negative weighted prototype variance term. This identity reveals a self-regulating mechanism built into the loss geometry: the gradient of the variance term acts as an implicit separation force that resists prototype collapse without any auxiliary objective, and leads to a global Lyapunov stability theorem for the reduced frozen-encoder system. Six blocks of controlled experiments validate each structural prediction. The decomposition identity holds with zero violations across more than one hundred thousand training epochs; the negative feedback cycle is confirmed with Pearson -0.98; with a jointly trained backbone, DDCL outperforms its non-differentiable ablation by 65% in clustering accuracy and DeepCluster end-to-end by 122%.

Integrating Large Language Models (LLMs) into hypersonic thermal protection system (TPS) design is bottlenecked by cascading constraint violations when generating executable simulation artifacts. General-purpose LLMs, treating generation as single-pass text completion, fail to satisfy the sequential, multi-gate constraints inherent in safety-critical engineering workflows. To address this, we propose AeroTherm-GPT, the first TPS-specialized LLM Agent, instantiated through a Constraint-Closed-Loop Generation (CCLG) framework. CCLG organizes TPS artifact generation as an iterative workflow comprising generation, validation, CDG-guided repair, execution, and audit. The Constraint Dependency Graph (CDG) encodes empirical co-resolution structure among constraint categories, directing repair toward upstream fault candidates based on lifecycle ordering priors and empirical co-resolution probabilities. This upstream-priority mechanism resolves multiple downstream violations per action, achieving a Root-Cause Fix Efficiency of 4.16 versus 1.76 for flat-checklist repair. Evaluated on HyTPS-Bench and validated against external benchmarks, AeroTherm-GPT achieves 88.7% End-to-End Success Rate (95% CI: 87.5-89.9), a gain of +12.5 pp over the matched non-CDG ablation baseline, without catastrophic forgetting on scientific reasoning and code generation tasks.

Retrieval-Augmented Generation (RAG) systems critically depend on retrieval quality, yet no systematic comparison of modern retrieval methods exists for heterogeneous documents containing both text and tabular data. We benchmark ten retrieval strategies spanning sparse, dense, hybrid fusion, cross-encoder reranking, query expansion, index augmentation, and adaptive retrieval on a challenging financial QA benchmark of 23,088 queries over 7,318 documents with mixed text-and-table content. We evaluate retrieval quality via Recall@k, MRR, and nDCG, and end-to-end generation quality via Number Match, with paired bootstrap significance testing. Our results show that (1) a two-stage pipeline combining hybrid retrieval with neural reranking achieves Recall@5 of 0.816 and MRR@3 of 0.605, outperforming all single-stage methods by a large margin; (2) BM25 outperforms state-of-the-art dense retrieval on financial documents, challenging the common assumption that semantic search universally dominates; and (3) query expansion methods (HyDE, multi-query) and adaptive retrieval provide limited benefit for precise numerical queries, while contextual retrieval yields consistent gains. We provide ablation studies on fusion methods and reranker depth, actionable cost-accuracy recommendations, and release our full benchmark code.

Cutting rectangular items from stock sheets to satisfy demands while minimizing waste is a central manufacturing task. The Two-Dimensional Single Stock Size Cutting Stock Problem (2D-CSSP) generalizes bin packing by requiring multiple copies of each item type, which causes a strong combinatorial blow-up. We present a SAT-based framework where item types are expanded by demand, each copy has a sheet-assignment variable and non-overlap constraints are activated only for copies assigned to the same sheet. We also introduce an infeasible-orientation elimination rule that fixes rotation variables when only one orientation can fit the sheet. For minimizing the number of sheets, we compare three approaches: non-incremental SAT with binary search, incremental SAT with clause reuse across iterations and weighted partial MaxSAT. On the Cui--Zhao benchmark suite, our best SAT configurations certify two to three times more instances as provably optimal and achieve lower optimality gaps than OR-Tools, CPLEX and Gurobi. The relative ranking among SAT approaches depends on rotation: incremental SAT is strongest without rotation, while non-incremental SAT is more effective when rotation increases formula size.

This paper investigates Koopman operator-based approaches for multivariable control of a two-spool turbofan engine. A physics-based component-level model is developed to generate training data and validate the controllers. A meta-heuristic extended dynamic mode decomposition is developed, with a cost function designed to accurately capture both spool-speed dynamics and the engine pressure ratio (EPR), enabling the construction of a single Koopman model suitable for multiple control objectives. Using the identified time-varying Koopman model, two controllers are developed: an adaptive Koopman-based model predictive controller (AKMPC) with a disturbance observer and a Koopman-based feedback linearization controller (K-FBLC), which serves as a benchmark. The controllers are evaluated for two control strategies, namely configurations of spool speeds and EPR, under both sea-level and varying flight conditions. The results demonstrate that the proposed identification approach enables accurate predictions of both spool speeds and EPR, allowing the Koopman model to be reused flexibly across different control formulations. While both control strategies achieve comparable performance in steady conditions, the AKMPC exhibits superior robustness compared with the K-FBLC under varying flight conditions due to its ability to compensate for model mismatch. Moreover, the EPR control strategy improves the thrust response. The study highlights the applicability of Koopman-based control and demonstrates the advantages of the AKMPC-based framework for robust turbofan engine control.

Achieving semantic interoperability across heterogeneous experimental data systems remains a major barrier to data-driven scientific discovery. The Analytical Information Markup Language (AnIML), a flexible XML-based standard for analytical chemistry and biology, is increasingly used in industrial R&D labs for managing and exchanging experimental data. However, the expressivity of the XML schema permits divergent interpretations across stakeholders, introducing inconsistencies that undermine the interoperability the AnIML schema was designed to support. In this paper, we present the AnIML Ontology, an OWL 2 ontology that formalises the semantics of AnIML and aligns it with the Allotrope Data Format to support future cross-system and cross-lab interoperability. The ontology was developed using an expert-in-the-loop approach combining LLM-assisted requirement elicitation with collaborative ontology engineering. We validate the ontology through a multi-layered approach: data-driven transformation of real-world AnIML files into knowledge graphs, competency question verification via SPARQL, and a novel validation protocol based on adversarial negative competency questions mapped to established ontological anti-patterns and enforced via SHACL constraints.

Forecasting evolving clinical risks relies on intrinsic pathological dependencies rather than mere chronological proximity, yet current methods struggle with coarse binary supervision and physical timestamps. To align predictive modeling with clinical logic, we propose the Medical-semantics Aware Time-ALiBi Transformer (MATA-Former), utilizing event semantics to dynamically parameterize attention weights to prioritize causal validity over time lags. Furthermore, we introduce Plateau-Gaussian Soft Labeling (PSL), reformulating binary classification into continuous multi-horizon regression for full-trajectory risk modeling. Evaluated on SIICU -- a newly constructed dataset featuring over 506k events with rigorous expert-verified, fine-grained annotations -- and the MIMIC-IV dataset, our framework demonstrates superior efficacy and robust generalization in capturing risks from text-intensive, irregular clinical time series.

General aviation fault diagnosis and efficient maintenance are critical to flight safety; however, deploying deep learning models on resource-constrained edge devices poses dual challenges in computational capacity and interpretability. This paper proposes LiteInception--a lightweight interpretable fault diagnosis framework designed for edge deployment. The framework adopts a two-stage cascaded architecture aligned with standard maintenance workflows: Stage 1 performs high-recall fault detection, and Stage 2 conducts fine-grained fault classification on anomalous samples, thereby decoupling optimization objectives and enabling on-demand allocation of computational resources. For model compression, a multi-method fusion strategy based on mutual information, gradient analysis, and SE attention weights is proposed to reduce the input sensor channels from 23 to 15, and a 1+1 branch LiteInception architecture is introduced that compresses InceptionTime parameters by 70%, accelerates CPU inference by over 8x, with less than 3% F1 loss. Furthermore, knowledge distillation is introduced as a precision-recall regulation mechanism, enabling the same lightweight model to adapt to different scenarios--such as safety-critical and auxiliary diagnosis--by switching training strategies. Finally, a dual-layer interpretability framework integrating four attribution methods is constructed, providing traceable evidence chains of "which sensor x which time period." Experiments on the NGAFID dataset demonstrate a fault detection accuracy of 81.92% with 83.24% recall, and a fault identification accuracy of 77.00%, validating the framework's favorable balance among efficiency, accuracy, and interpretability.

Vision-Language-Action (VLA) models for autonomous driving must integrate diverse textual inputs, including navigation commands, hazard warnings, and traffic state descriptions, yet current systems often present these as disconnected fragments, forcing the model to discover on its own which environmental constraints are relevant to the current maneuver. We introduce Causal Scene Narration (CSN), which restructures VLA text inputs through intent-constraint alignment, quantitative grounding, and structured separation, at inference time with zero GPU cost. We complement CSN with Simplex-based runtime safety supervision and training-time alignment via Plackett-Luce DPO with negative log-likelihood (NLL) regularization. A multi-town closed-loop CARLA evaluation shows that CSN improves Driving Score by +31.1% on original LMDrive and +24.5% on the preference-aligned variant. A controlled ablation reveals that causal structure accounts for 39.1% of this gain, with the remainder attributable to information content alone. A perception noise ablation confirms that CSN's benefit is robust to realistic sensing errors. Semantic safety supervision improves Infraction Score, while reactive Time-To-Collision monitoring degrades performance, demonstrating that intent-aware monitoring is needed for VLA systems.

The wind-induced structural response forecasting capabilities of a novel transformer methodology are examined here. The model also provides a digital twin component for bridge structural health monitoring. Firstly, the approach uses the temporal characteristics of the system to train a forecasting model. Secondly, the vibration predictions are compared to the measured ones to detect large deviations. Finally, the identified cases are used as an early-warning indicator of structural change. The artificial intelligence-based model outperforms approaches for response forecasting as no assumption on wind stationarity or on structural normal vibration behavior is needed. Specifically, wind-excited dynamic behavior suffers from uncertainty related to obtaining poor predictions when the environmental or traffic conditions change. This results in a hard distinction of what constitutes normal vibration behavior. To this end, a framework is rigorously examined on real-world measurements from the Hardanger Bridge monitored by the Norwegian University of Science and Technology. The approach captures accurate structural behavior in realistic conditions, and with respect to the changes in the system excitation. The results, importantly, highlight the potential of transformer-based digital twin components to serve as next-generation tools for resilient infrastructure management, continuous learning, and adaptive monitoring over the system's lifecycle with respect to temporal characteristics.

Vietnamese Speech Emotion Recognition (SER) remains challenging due to ambiguous acoustic patterns and the lack of reliable annotated data, especially in real-world conditions where emotional boundaries are not clearly separable. To address this problem, this paper proposes a human-machine collaborative framework that integrates human knowledge into the learning process rather than relying solely on data-driven models. The proposed framework is centered around LLM-based reasoning, where acoustic feature-based models are used to provide auxiliary signals such as confidence and feature-level evidence. A confidence-based routing mechanism is introduced to distinguish between easy and ambiguous samples, allowing uncertain cases to be delegated to LLMs for deeper reasoning guided by structured rules derived from human annotation behavior. In addition, an iterative refinement strategy is employed to continuously improve system performance through error analysis and rule updates. Experiments are conducted on a Vietnamese speech dataset of 2,764 samples across three emotion classes (calm, angry, panic), with high inter-annotator agreement (Fleiss Kappa = 0.8574), ensuring reliable ground truth. The proposed method achieves strong performance, reaching up to 86.59% accuracy and Macro F1 around 0.85-0.86, demonstrating its effectiveness in handling ambiguous and hard-to-classify cases. Overall, this work highlights the importance of combining data-driven models with human reasoning, providing a robust and model-agnostic approach for speech emotion recognition in low-resource settings.

Adaptation to complex tasks and multiple scenarios remains a significant challenge for a single robot agent. The ability to acquire organize, and switch between a wide range of skills in real time, particularly in dynamic environments, has become a fundamental requirement for embodied intelligence. We introduce OpenGo, an OpenClaw-powered embodied robotic dog capable of switching skills in real time according to the scene and task instructions. Specifically, the agent is equipped with (1) a customizable skill library with easy skill import and autonomous skill validation, (2) a dispatcher that selects and invokes different skills according to task prompts or language instructions, and (3) a self-learning framework that fine-tunes skills based on task completion and human feedback. We deploy the agent in Unitree's Go2 robotic dog and validate its capabilities in self-checking and switching of skills autonomously. In addition, by integrating Feishu-platform communication, we enable natural-language guidance and human feedback, allowing inexperienced users to control the robotic dog through simple instructions.

Memory emerges as the core module in the large language model (LLM)-based agents for long-horizon complex tasks (e.g., multi-turn dialogue, game playing, scientific discovery), where memory can enable knowledge accumulation, iterative reasoning and self-evolution. A number of memory methods have been proposed in the literature. However, these methods have not been systematically and comprehensively compared under the same experimental settings. In this paper, we first summarize a unified framework that incorporates all the existing agent memory methods from a high-level perspective. We then extensively compare representative agent memory methods on two well-known benchmarks and examine the effectiveness of all methods, providing a thorough analysis of those methods. As a byproduct of our experimental analysis, we also design a new memory method by exploiting modules in the existing methods, which outperforms the state-of-the-art methods. Finally, based on these findings, we offer promising future research opportunities. We believe that a deeper understanding of the behavior of existing methods can provide valuable new insights for future research.

Automatic speech recognition (ASR) is a critical interface for human-AI interaction in gastrointestinal endoscopy, yet its reliability in real-world clinical settings is limited by domain-specific terminology and complex acoustic conditions. Here, we present EndoASR, a domain-adapted ASR system designed for real-time deployment in endoscopic workflows. We develop a two-stage adaptation strategy based on synthetic endoscopy reports, targeting domain-specific language modeling and noise robustness. In retrospective evaluation across six endoscopists, EndoASR substantially improves both transcription accuracy and clinical usability, reducing character error rate (CER) from 20.52% to 14.14% and increasing medical term accuracy (Med ACC) from 54.30% to 87.59%. In a prospective multi-center study spanning five independent endoscopy centers, EndoASR demonstrates consistent generalization under heterogeneous real-world conditions. Compared with the baseline Paraformer model, CER is reduced from 16.20% to 14.97%, while Med ACC is improved from 61.63% to 84.16%, confirming its robustness in practical deployment scenarios. Notably, EndoASR achieves a real-time factor (RTF) of 0.005, significantly faster than Whisper-large-v3 (RTF 0.055), while maintaining a compact model size of 220M parameters, enabling efficient edge deployment. Furthermore, integration with large language models demonstrates that improved ASR quality directly enhances downstream structured information extraction and clinician-AI interaction. These results demonstrate that domain-adapted ASR can serve as a reliable interface for human-AI teaming in gastrointestinal endoscopy, with consistent performance validated across multi-center real-world clinical settings.

Supervised Fine-Tuning (SFT) on long Chain-of-Thought (CoT) trajectories has become a pivotal phase in building large reasoning models. However, how CoT trajectories from different sources influence the generalization performance of models remains an open question. In this paper, we conduct a comparative study using two sources of verified CoT trajectories generated by two competing models, \texttt{DeepSeek-R1-0528} and \texttt{gpt-oss-120b}, with their problem sets controlled to be identical. Despite their comparable performance, we uncover a striking paradox: lower training loss does not translate to better generalization. SFT on \texttt{DeepSeek-R1-0528} data achieves remarkably lower training loss, yet exhibits significantly worse generalization performance on reasoning benchmarks compared to those trained on \texttt{gpt-oss-120b}. To understand this paradox, we perform a multi-faceted analysis probing token-level SFT loss and step-level reasoning behaviors. Our analysis reveals a difference in reasoning patterns. \texttt{gpt-oss-120b} exhibits highly convergent and deductive trajectories, whereas \texttt{DeepSeek-R1-0528} favors a divergent and branch-heavy exploration pattern. Consequently, models trained with \texttt{DeepSeek-R1} data inherit inefficient exploration behaviors, often getting trapped in redundant exploratory branches that hinder them from reaching correct solutions. Building upon this insight, we propose a simple yet effective remedy of filtering out frequently branching trajectories to improve the generalization of SFT. Experiments show that training on selected \texttt{DeepSeek-R1-0528} subsets surprisingly improves reasoning performance by up to 5.1% on AIME25, 5.5% on BeyondAIME, and on average 3.6% on five benchmarks.

Minor Component Adaptation (MiCA) is a novel parameter-efficient fine-tuning method for large language models that focuses on adapting underutilized subspaces of model representations. Unlike conventional methods such as Low-Rank Adaptation (LoRA), which target dominant subspaces, MiCA leverages Singular Value Decomposition to identify subspaces related to minor singular vectors associated with the least significant singular values and constrains the update of parameters during fine-tuning to those directions. This strategy leads to up to 5.9x improvement in knowledge acquisition under optimized training hyperparameters and a minimal parameter footprint of 6-60% compared to LoRA. These results suggest that constraining adaptation to minor singular directions provides a more efficient and stable mechanism for integrating new knowledge into pre-trained language models.

The rapid proliferation of Artificial Intelligence-Generated Content (AIGC) is fundamentally restructuring online content ecologies, necessitating a rigorous examination of its behavioral and distributional implications. Leveraging a comprehensive longitudinal dataset comprising tens of millions of users from a leading Chinese video-sharing platform, this study elucidated the distinct creation and consumption behaviors characterizing AIGC versus Human-Generated Content (HGC). We identified a prevalent scale-over-preference dynamic, wherein AIGC creators achieve aggregate engagement comparable to HGC creators through high-volume production, despite a marked consumer preference for HGC. Deeper analysis uncovered the ability of the algorithmic content distribution mechanism in moderating these competing interests regarding AIGC. These findings advocated for the implementation of AIGC-sensitive distribution algorithms and precise governance frameworks to ensure the long-term health of the online content platforms.

Anthropic proposes the concept of skills for LLM agents to tackle multi-step professional tasks that simple tool invocations cannot address. A tool is a single, self-contained function, whereas a skill is a structured bundle of interdependent multi-file artifacts. Currently, skill generation is not only label-intensive due to manual authoring, but also may suffer from human--machine cognitive misalignment, which can lead to degraded agent performance, as evidenced by evaluations on SkillsBench. Therefore, we aim to enable agents to autonomously generate skills. However, existing self-evolving methods designed for tools cannot be directly applied to skills due to their increased complexity. To address these issues, we propose EvoSkills, a self-evolving skills framework that enables agents to autonomously construct complex, multi-file skill packages. Specifically, EvoSkills couples a Skill Generator that iteratively refines skills with a Surrogate Verifier that co-evolves to provide informative and actionable feedback without access to ground-truth test content. On SkillsBench, EvoSkills achieves the highest pass rate among five baselines on both Claude Code and Codex, and also exhibits strong generalization capabilities to six additional LLMs.

Standard scaled dot-product attention computes scores from static, independent projections of the input. We show that evolving queries and keys \emph{jointly} through shared learned dynamics before scoring - which we call \textbf{coupled QK dynamics} - improves language modeling perplexity and training stability. On WikiText-103 at 60M parameters, coupled dynamics achieves 22.55--22.62 perplexity vs.\ 24.22 for standard attention ($-$6.6--6.9\%), with only 0.11\% additional parameters (shared across both instantiations). A structural ablation isolates coupling as the active ingredient: a symplectic (Hamiltonian) and a non-symplectic (Euler) integrator perform identically when both couple Q and K, while an uncoupled MLP baseline of matched capacity reaches only 23.81 with 8$\times$ higher seed variance. The integration step count (1--7) is similarly irrelevant - a single coupled step suffices. A compute-matched comparison reveals that coupling is a \emph{sample-efficiency} mechanism: standard attention trained for 2.4$\times$ longer (matching wall-clock) reaches the same perplexity, but requires 2.4$\times$ more tokens. The advantage scales to 150M ($-$6.7\%) but narrows at 350M ($-$1.0\%), where Differential Attention (18.93) overtakes coupled dynamics (19.35). The benefit is corpus-dependent: coupling helps on domain-coherent text (WikiText-103 $-$6.6\%, PubMed $-$4.5\%) but degrades on heterogeneous web text ($+$10.3\%) and shows no benefit on GLUE. We characterize when coupling helps and when it does not, providing practical guidelines.

Supervised fine-tuning (SFT) with token-level hard labels can amplify overconfident imitation of factually unsupported targets, causing hallucinations that propagate in multi-sentence generation. We study an augmented SFT setting in which training instances include coarse sentence-level factuality risk labels and inter-sentence dependency annotations, providing structured signals about where factual commitments are weakly supported. We propose \textbf{PRISM}, a differentiable risk-gated framework that modifies learning only at fact-critical positions. PRISM augments standard SFT with a lightweight, model-aware probability reallocation objective that penalizes high-confidence predictions on risky target tokens, with its scope controlled by span-level risk weights and model-aware gating. Experiments on hallucination-sensitive factual benchmarks and general evaluations show that PRISM improves factual aggregates across backbones while maintaining a competitive overall capability profile. Ablations further show that the auxiliary signal is most effective when used conservatively, and that knowledge masking and model-aware reallocation play complementary roles in balancing factual correction and capability preservation.

Large foundation models enable powerful reasoning for autonomous systems, but mapping semantic intent to reliable real-time control remains challenging. Existing approaches either (i) let Large Language Models (LLMs) generate trajectories directly - brittle, hard to verify, and latency-prone - or (ii) adjust Model Predictive Control (MPC) objectives online - mixing slow deliberation with fast control and blurring interfaces. We propose Agentic Fast-Slow Planning, a hierarchical framework that decouples perception, reasoning, planning, and control across natural timescales. The framework contains two bridges. Perception2Decision compresses scenes into ego-centric topologies using an on-vehicle Vision-Language Model (VLM) detector, then maps them to symbolic driving directives in the cloud with an LLM decision maker - reducing bandwidth and delay while preserving interpretability. Decision2Trajectory converts directives into executable paths: Semantic-Guided A* embeds language-derived soft costs into classical search to bias solutions toward feasible trajectories, while an Agentic Refinement Module adapts planner hyperparameters using feedback and memory. Finally, MPC tracks the trajectories in real time, with optional cloud-guided references for difficult cases. Experiments in CARLA show that Agentic Fast-Slow Planning improves robustness under perturbations, reducing lateral deviation by up to 45% and completion time by over 12% compared to pure MPC and an A*-guided MPC baseline. Code is available at https://github.com/cjychenjiayi/icra2026_AFSP.

Promptly porting patches from a source codebase to its variants (e.g., forks and branches) is essential for mitigating propagated defects and vulnerabilities. Recent studies have explored automated patch porting to reduce manual effort and delay, but existing approaches mainly handle inconsistencies visible in a patch's local context and struggle with those requiring global mapping knowledge between codebases. We refer to such non-local inconsistencies as implicit inconsistencies. Implicit inconsistencies pose greater challenges for developers to resolve due to their non-local nature. To address them, we propose MIP, which enables collaboration among an LLM, a compiler, and code analysis utilities. MIP adopts different strategies for different cases: when source identifiers exist in the target codebase, it leverages compiler diagnostics; otherwise, it retrieves matched code segment pairs from the two codebases as mapping knowledge for mitigation. Experiments on two representative scenarios, cross-fork and cross-branch patch porting, show that MIP successfully resolves more than twice as many patches as the best-performing baseline in both settings. A user study with our industry partner further demonstrates its practical effectiveness.

GUI Process Automation (GPA) is a lightweight but general vision-based Robotic Process Automation (RPA), which enables fast and stable process replay with only a single demo. Addressing the fragility of traditional RPA and the non-deterministic risks of current vision language model-based GUI agents, GPA introduces three core benefits: (1) Robustness via Sequential Monte Carlo-based localization to handle rescaling and detection uncertainty; (2) Deterministic and Reliability safeguarded by readiness calibration; and (3) Privacy through fast, fully local execution. This approach delivers the adaptability, robustness, and security required for enterprise workflows. It can also be used as an MCP/CLI tool by other agents with coding capabilities so that the agent only reasons and orchestrates while GPA handles the GUI execution. We conducted a pilot experiment to compare GPA with Gemini 3 Pro (with CUA tools) and found that GPA achieves higher success rate with 10 times faster execution speed in finishing long-horizon GUI tasks.

Model merging aims to integrate multiple expert models into a single model that inherits their complementary strengths without incurring the inference-time cost of ensembling. Recent progress has shown that merging can be highly effective when all source models are \emph{homogeneous}, i.e., derived from the same pretrained backbone and therefore share aligned parameter coordinates or compatible task vectors. Yet this assumption is increasingly unrealistic in open model ecosystems, where useful experts are often built on different families such as Llama, Qwen, and Mistral. In such \emph{heterogeneous} settings, direct weight-space fusion becomes ill-posed due to architectural mismatch, latent basis misalignment, and amplified cross-source conflict. We address this problem with \texttt{HeteroFusion} for heterogeneous language model fusion, which consists of two key components: topology-based alignment that transfers knowledge across heterogeneous backbones by matching functional module structures instead of raw tensor coordinates, and conflict-aware denoising that suppresses incompatible or noisy transfer signals during fusion. We further provide analytical justification showing that preserving the target adapter basis while predicting structured updates leads to a stable and well-conditioned transfer process. Across heterogeneous transfer, multi-source fusion, noisy-source robustness, and cross-family generalization settings, \texttt{HeteroFusion} consistently outperforms strong merging, fusion, and ensemble baselines.

Emotional Support Conversation (ESC) aims to alleviate individual emotional distress by generating empathetic responses. However, existing methods face challenges in effectively supporting deep contextual understanding. To address this issue, we propose PRCCF, a Persona-guided Retrieval and Causality-aware Cognitive Filtering framework. Specifically, the framework incorporates a persona-guided retrieval mechanism that jointly models semantic compatibility and persona alignment to enhance response generation. Furthermore, it employs a causality-aware cognitive filtering module to prioritize causally relevant external knowledge, thereby improving contextual cognitive understanding for emotional reasoning. Extensive experiments on the ESConv dataset demonstrate that PRCCF outperforms state-of-the-art baselines on both automatic metrics and human evaluations. Our code is publicly available at: https://github.com/YancyLyx/PRCCF.

While long-term memory is essential for intelligent agents to maintain consistent historical awareness, the accumulation of extensive interaction data often leads to performance bottlenecks. Naive storage expansion increases retrieval noise and computational latency, overwhelming the reasoning capacity of models deployed on constrained personal devices. To address this, we propose Hierarchical Memory Orchestration (HMO), a framework that organizes interaction history into a three-tiered directory driven by user-centric contextual relevance. Our system maintains a compact primary cache, coupling recent and pivotal memories with an evolving user profile to ensure agent reasoning remains aligned with individual behavioral traits. This primary cache is complemented by a high-priority secondary layer, both of which are managed within a global archive of the full interaction history. Crucially, the user persona dictates memory redistribution across this hierarchy, promoting records mapped to long-term patterns toward more active tiers while relegating less relevant information. This targeted orchestration surfaces historical knowledge precisely when needed while maintaining a lean and efficient active search space. Evaluations on multiple benchmarks achieve state-of-the-art performance. Real-world deployments in ecosystems like OpenClaw demonstrate that HMO significantly enhances agent fluidity and personalization.

Embodied perception systems face severe challenges of dynamic environment distribution drift when they continuously interact in open physical spaces. However, the existing domain incremental awareness methods often rely on the domain id obtained in advance during the testing phase, which limits their practicability in unknown interaction scenarios. At the same time, the model often overfits to the context-specific perceptual noise, which leads to insufficient generalization ability and catastrophic forgetting. To address these limitations, we propose a domain-id and exemplar-free incremental learning framework for embodied multimedia systems, which aims to achieve robust continuous environment adaptation. This method designs a disentangled representation mechanism to remove non-essential environmental style interference, and guide the model to focus on extracting semantic intrinsic features shared across scenes, thereby eliminating perceptual uncertainty and improving generalization. We further use the weight fusion strategy to dynamically integrate the old and new environment knowledge in the parameter space, so as to ensure that the model adapts to the new distribution without storing historical data and maximally retains the discrimination ability of the old environment. Extensive experiments on multiple standard benchmark datasets show that the proposed method significantly reduces catastrophic forgetting in a completely exemplar-free and domain-id free setting, and its accuracy is better than the existing state-of-the-art methods.

Multi-modal fusion is of great significance in neuroscience which integrates information from different modalities and can achieve better performance than uni-modal methods in downstream tasks. Current multi-modal fusion methods in brain networks, which mainly focus on structural connectivity (SC) and functional connectivity (FC) modalities, are static in nature. They feed different samples into the same model with identical computation, ignoring inherent difference between input samples. This lack of sample adaptation hinders model's further performance. To this end, we innovatively propose a multi-stage dynamic fusion strategy (M3D-BFS) for sample-adaptive multi-modal brain network analysis. Unlike other static fusion methods, we design different mixture-of-experts (MoEs) for uni- and multi-modal representations where modules can adaptively change as input sample changes during inference. To alleviate issue of MoE where training of experts may be collapsed, we divide our method into 3 stages. We first train uni-modal encoders respectively, then pretrain single experts of MoEs before finally finetuning the whole model. A multi-modal disentanglement loss is designed to enhance the final representations. To the best of our knowledge, this is the first work for dynamic fusion for multi-modal brain network analysis. Extensive experiments on different real-world datasets demonstrates the superiority of M3D-BFS.

LLM-based agents show strong potential for long-horizon reasoning, yet their context size is limited by deployment factors (e.g., memory, latency, and cost), yielding a constrained context budget. As interaction histories grow, this induces a trade-off between retaining past information and staying within the context limit. To address this challenge, we propose Budget-Aware Context Management (BACM), which formulates context management as a sequential decision problem with a context budget constraint. It enables agents to assess the available budget before incorporating new observations and decide when and how much of the interaction history to compress. We further develop BACM-RL, an end-to-end curriculum-based reinforcement learning approach that learns compression strategies under varying context budgets. Experiments on compositional multi-objective QA and long-horizon web browsing benchmarks show that BACM-RL consistently outperforms prior methods across model scales and task complexities, achieving over $1.6\times$ gains over strong baselines in high-complexity settings, while maintaining strong advantages as budgets shrink, where most methods exhibit a downward performance trend.

Clinical AI systems routinely train on health data structurally distorted by documentation workflows, billing incentives, and terminology fragmentation. Prior work has characterised the mechanisms of this distortion: the three-forces model of documentary enactment, the reification feedback loop through which AI may amplify coding artefacts, and terminology governance failures that allow semantic drift to accumulate. Yet translating these insights into implementable software architecture remains an open problem. This paper proposes seven ontology-aware design patterns in Gang-of-Four pattern language for building clinical AI pipelines resilient to ontological distortion. The patterns address data ingestion validation (Ontological Checkpoint), low-frequency signal preservation (Dormancy-Aware Pipeline), continuous drift monitoring (Drift Sentinel), parallel representation maintenance (Dual-Ontology Layer), feedback loop interruption (Reification Circuit Breaker), terminology evolution management (Terminology Version Gate), and pluggable regulatory compliance (Regulatory Compliance Adapter). Each pattern is specified with Problem, Forces, Solution, Consequences, Known Uses, and Related Patterns. We illustrate their composition in a reference architecture for a primary care AI system and provide a walkthrough tracing all seven patterns through a diabetes risk prediction scenario. This paper does not report empirical validation; it offers a design vocabulary grounded in theoretical analysis, subject to future evaluation in production systems. Three patterns have partial precedent in existing systems; the remaining four have not been formally described. Limitations include the absence of runtime benchmarks and restriction to the German and EU regulatory context.

Large language model (LLM)-based evolution is a promising approach for open-ended discovery, where progress requires sustained search and knowledge accumulation. Existing methods still rely heavily on fixed heuristics and hard-coded exploration rules, which limit the autonomy of LLM agents. We present CORAL, the first framework for autonomous multi-agent evolution on open-ended problems. CORAL replaces rigid control with long-running agents that explore, reflect, and collaborate through shared persistent memory, asynchronous multi-agent execution, and heartbeat-based interventions. It also provides practical safeguards, including isolated workspaces, evaluator separation, resource management, and agent session and health management. Evaluated on diverse mathematical, algorithmic, and systems optimization tasks, CORAL sets new state-of-the-art results on 10 tasks, achieving 3-10 times higher improvement rates with far fewer evaluations than fixed evolutionary search baselines across tasks. On Anthropic's kernel engineering task, four co-evolving agents improve the best known score from 1363 to 1103 cycles. Mechanistic analyses further show how these gains arise from knowledge reuse and multi-agent exploration and communication. Together, these results suggest that greater agent autonomy and multi-agent evolution can substantially improve open-ended discovery. Code is available at https://github.com/Human-Agent-Society/CORAL.

Despite rapid progress in claim verification, we lack a systematic understanding of what reasoning these benchmarks actually exercise. We generate structured reasoning traces for 24K claim-verification examples across 9 datasets using GPT-4o-mini and find that direct evidence extraction dominates, while multi-sentence synthesis and numerical reasoning are severely under-represented. A dataset-level breakdown reveals stark biases: some datasets almost exclusively test lexical matching, while others require information synthesis in roughly half of cases. Using a compact 1B-parameter reasoning verifier, we further characterize five error types and show that error profiles vary dramatically by domain -- general-domain verification is dominated by lexical overlap bias, scientific verification by overcautiousness, and mathematical verification by arithmetic reasoning failures. Our findings suggest that high benchmark scores primarily reflect retrieval-plus-entailment ability. We outline recommendations for building more challenging evaluation suites that better test the reasoning capabilities verification systems need.

Recent advances in video generation have made AI-synthesized content increasingly difficult to distinguish from real footage. We propose a physics-based authentication signature that real cameras produce naturally, but that generative models cannot faithfully reproduce. Our approach exploits the Moiré effect: the interference fringes formed when a camera views a compact two-layer grating structure. We derive the Moiré motion invariant, showing that fringe phase and grating image displacement are linearly coupled by optical geometry, independent of viewing distance and grating structure. A verifier extracts both signals from video and tests their correlation. We validate the invariant on both real-captured and AI-generated videos from multiple state-of-the-art generators, and find that real and AI-generated videos produce significantly different correlation signatures, suggesting a robust means of differentiating them. Our work demonstrates that deterministic optical phenomena can serve as physically grounded, verifiable signatures against AI-generated video.

Electroencephalography (EEG) provides a non-invasive insight into the brain's cognitive and emotional dynamics. However, modeling how these states evolve in real time and quantifying the energy required for such transitions remains a major challenge. The Schrödinger Bridge Problem (SBP) offers a principled probabilistic framework to model the most efficient evolution between the brain states, interpreted as a measure of cognitive energy cost. While generative models such as GANs have been widely used to augment EEG data, it remains unclear whether synthetic EEG preserves the underlying dynamical structure required for transition-based analysis. In this work, we address this gap by using SBP-derived transport cost as a metric to evaluate whether GAN-generated EEG retains the distributional geometry necessary for energy-based modeling of cognitive state transitions. We compare transition energies derived from real and synthetic EEG collected during Stroop tasks and demonstrate strong agreement across group and participant-level analyses. These results indicate that synthetic EEG preserves the transition structure required for SBP-based modeling, enabling its use in data-efficient neuroadaptive systems. We further present a framework in which SBP-derived cognitive energy serves as a control signal for adaptive human-machine systems, supporting real-time adjustment of system behavior in response to user cognitive and affective state.

We present ThinknCheck, a 1B-parameter verifier for grounded claim verification that first produces a short, structured rationale and then a binary verdict. We construct LLMAggreFact-Think, a 24.1k reasoning-augmented training set derived from LLMAggreFact, and fine-tune a 4-bit Gemma3 model to follow this format. On LLMAggreFact, ThinknCheck attains 78.1 balanced accuracy (BAcc), surpassing MiniCheck-7B (77.4) with 7x fewer parameters; removing the reasoning step reduces BAcc to 57.5. On SciFact, ThinknCheck reaches 64.7 BAcc, a +14.7 absolute gain over MiniCheck-7B. By contrast, zero-shot chain-of-thought on the base Gemma3-1B harms accuracy relative to direct answers, and preference optimization with a simple format+accuracy reward underperforms supervised reasoning. To probe the latter, we introduce GSMClaims and a domain-specialized variant, ThinknCheck-Science, which improves across benchmarks, including 61.0\% accuracy on GSMClaims. Overall, explicit, supervised reasoning enables compact verifiers that are competitive while remaining resource-efficient and interpretable.

An assumption often made in supervised learning is that the training and testing sets have the same label distribution. However, in real-life scenarios, this assumption rarely holds. For example, medical diagnosis result distributions change over time and across locations; fraud detection models must adapt as patterns of fraudulent activity shift; the category distribution of social media posts changes based on trending topics and user demographics. In the task of label shift estimation, the goal is to estimate the changing label distribution $p_t(y)$ in the testing set, assuming the likelihood $p(x|y)$ does not change, implying no concept drift. In this paper, we propose a new approach for post-hoc label shift estimation, unlike previous methods that perform moment matching with confusion matrix estimated from a validation set or maximize the likelihood of the new data with an expectation-maximization algorithm. We aim to incrementally update the prior on each sample, adjusting each posterior for more accurate label shift estimation. The proposed method is based on intuitive assumptions on classifiers that are generally true for modern probabilistic classifiers. The proposed method relies on a weaker notion of calibration compared to other methods. As a post-hoc approach for label shift estimation, the proposed method is versatile and can be applied to any black-box probabilistic classifier. Experiments on CIFAR-10 and MNIST show that the proposed method consistently outperforms the current state-of-the-art maximum likelihood-based methods under different calibrations and varying intensities of label shift.

Smell's deep connection with food, memory, and social experience has long motivated researchers to bring olfaction into interactive systems. Yet most olfactory interfaces remain limited to fixed scent cartridges and pre-defined generation patterns, and the scarcity of large-scale olfactory datasets has further constrained AI-based approaches. We present AromaGen, an AI-powered wearable interface capable of real-time, general-purpose aroma generation from free-form text or visual inputs. AromaGen is powered by a multimodal LLM that leverages latent olfactory knowledge to map semantic inputs to structured mixtures of 12 carefully selected base odorants, released through a neck-worn dispenser. Users can iteratively refine generated aromas through natural language feedback via in-context learning. Through a controlled user study ($N = 26$), AromaGen matches human-composed mixtures in zero-shot generation and significantly surpasses them after iterative refinement, achieving a median similarity of 8/10 to real food aromas and reducing perceived artificiality to levels comparable to real food. AromaGen is a step towards real-world interactive aroma generation, opening new possibilities for communication, wellbeing, and immersive technologies.

Embedding LLM-driven agents into environmental FAIR data management is compelling - they can externalize operational knowledge and scale curation across heterogeneous data and evolving conventions. However, replacing deterministic components with probabilistic workflows changes the failure mode: LLM pipelines may generate plausible but incorrect outputs that pass superficial checks and propagate into irreversible actions such as DOI minting and public release. We introduce EnviSmart, a production data management system deployed on campus-wide storage infrastructure for environmental research. EnviSmart treats reliability as an architectural property through two mechanisms: a three-track knowledge architecture that externalizes behaviors (governance constraints), domain knowledge (retrievable context), and skills (tool-using procedures) as persistent, interlocking artifacts; and a role-separated multi-agent design where deterministic validators and audited handoffs restore fail-stop semantics at trust boundaries before irreversible steps. We compare two production deployments. The University's GIS Center Ecological Archive (849 curated datasets) serves as a single-agent baseline. SF2Bench, a compound flooding benchmark comprising 2,452 monitoring stations and 8,557 published files spanning 39 years, validates the multi-agent workflow. The multi-agent approach improved both efficiency - completed by a single operator in two days with repeated artifact reuse across deployments - and reliability: audited handoffs detected and blocked a coordinate transformation error affecting all 2,452 stations before publication. A representative incident (ISS-004) demonstrated boundary-based containment with 10-minute detection latency, zero user exposure, and 80-minute resolution. This paper has been accepted at PEARC 2026.

Large language models demonstrate strong performance on mathematical reasoning benchmarks, yet remain surprisingly fragile to meaning-preserving surface perturbations. We systematically evaluate three open-weight LLMs, Mistral-7B, Llama-3-8B, and Qwen2.5-7B, on 677 GSM8K problems paired with semantically equivalent variants generated through name substitution and number format paraphrasing. All three models exhibit substantial answer-flip rates (28.8%-45.1%), with number paraphrasing consistently more disruptive than name swaps. To trace the mechanistic basis of these failures, we introduce the Mechanistic Perturbation Diagnostics (MPD) framework, combining logit lens analysis, activation patching, component ablation, and the Cascading Amplification Index (CAI) into a unified diagnostic pipeline. CAI, a novel metric quantifying layer-wise divergence amplification, outperforms first divergence layer as a failure predictor for two of three architectures (AUC up to 0.679). Logit lens reveals that flipped samples diverge from correct predictions at significantly earlier layers than stable samples. Activation patching reveals a stark architectural divide in failure localizability: Llama-3 failures are recoverable by patching at specific layers (43/60 samples), while Mistral and Qwen failures are broadly distributed (3/60 and 0/60). Based on these diagnostic signals, we propose a mechanistic failure taxonomy (localized, distributed, and entangled) and validate it through targeted repair experiments: steering vectors and layer fine-tuning recover 12.2% of localized failures (Llama-3) but only 7.2% of entangled (Qwen) and 5.2% of distributed (Mistral) failures.

Existing benchmarks for LLM-based vulnerability detection compress model performance into a single metric, which fails to reflect the distinct priorities of different stakeholders. For example, a CISO may emphasize high recall of critical vulnerabilities, an engineering leader may prioritize minimizing false positives, and an AI officer may balance capability against cost. To address this limitation, we introduce SecLens-R, a multi-stakeholder evaluation framework structured around 35 shared dimensions grouped into 7 measurement categories. The framework defines five role-specific weighting profiles: CISO, Chief AI Officer, Security Researcher, Head of Engineering, and AI-as-Actor. Each profile selects 12 to 16 dimensions with weights summing to 80, yielding a composite Decision Score between 0 and 100. We apply SecLens-R to evaluate 12 frontier models on a dataset of 406 tasks derived from 93 open-source projects, covering 10 programming languages and 8 OWASP-aligned vulnerability categories. Evaluations are conducted across two settings: Code-in-Prompt (CIP) and Tool-Use (TU). Results show substantial variation across stakeholder perspectives, with Decision Scores differing by as much as 31 points for the same model. For instance, Qwen3-Coder achieves an A (76.3) under the Head of Engineering profile but a D (45.2) under the CISO profile, while GPT-5.4 shows a similar disparity. These findings demonstrate that vulnerability detection is inherently a multi-objective problem and that stakeholder-aware evaluation provides insights that single aggregated metrics obscure.

Real-world reasoning often requires combining information across modalities, connecting textual context with visual cues in a multi-hop process. Yet, most multimodal benchmarks fail to capture this ability: they typically rely on single images or set of images, where answers can be inferred from a single modality alone. This limitation is mirrored in the training data, where interleaved image-text content rarely enforces complementary, multi-hop reasoning. As a result, Vision-Language Models (VLMs) frequently hallucinate and produce reasoning traces poorly grounded in visual evidence. To address this gap, we introduce CRIT, a new dataset and benchmark built with a graph-based automatic pipeline for generating complex cross-modal reasoning tasks. CRIT consists of diverse domains ranging from natural images, videos, and text-rich sources, and includes a manually verified test set for reliable evaluation. Experiments on this benchmark reveal that even state-of-the-art models struggle on such reasoning tasks. Models trained on CRIT show significant gains in cross-modal multi-hop reasoning, including strong improvements on SPIQA and other standard multimodal benchmarks.

Despite emerging use in Indonesian classrooms, there is limited large-scale, teacher-centred evidence on how AI is used in practice and what support teachers need, hindering the development of context-appropriate AI systems and policies. To address this gap, we conduct a nationwide survey of 349 K-12 teachers across elementary, junior high, and senior high schools. We find increasing use of AI for pedagogy, content development, and teaching media, although adoption remains uneven. Elementary teachers report more consistent use, while senior high teachers engage less; mid-career teachers assign higher importance to AI, and teachers in Eastern Indonesia perceive greater value. Across levels, teachers primarily use AI to reduce instructional preparation workload (e.g., assessment, lesson planning, and material development). However, generic outputs, infrastructure constraints, and limited contextual alignment continue to hinder effective classroom integration.

Translating security intent into deployable network enforcement rules and maintaining their effectiveness despite evolving cyber threats remains a largely manual process in most Security Operations Centers (SOCs). In large and heterogeneous networks, this challenge is complicated by topology-dependent reachability constraints and device-specific security control capabilities, making the process slow, error-prone, and a recurring source of misconfigurations. This paper presents RefinementEngine, an engine that automates the refinement of high-level security intents into low-level, deployment-ready configurations. Given a network topology, devices, and available security controls, along with high-level intents and Cyber Threat Intelligence (CTI) reports, RefinementEngine automatically generates settings that implement the desired intent, counter reported threats, and can be directly deployed on target security controls. The proposed approach is validated through real-world use cases on packet and web filtering policies derived from actual CTI reports, demonstrating both correctness, practical applicability, and adaptability to new data.

Diffusion language models (DLMs) expose their denoising trajectories, offering a natural handle for inference-time control; accordingly, an ideal hallucination mitigation framework should intervene during generation using this model-native signal rather than relying on an externally trained hallucination classifier. Toward this, we formulate commitment uncertainty localization: given a denoising trajectory, identify token positions whose cross-chain entropy exceeds an unsupervised threshold before factually unreliable commitments propagate into self-consistent but incorrect outputs. We introduce a suite of trajectory-level assessments, including a cross-chain divergence-at-hallucination (CDH) metric, for principled comparison of localization methods. We also introduce OSCAR, a training-free inference-time framework operationalizing this formulation. OSCAR runs N parallel denoising chains with randomized reveal orders, computes cross-chain Shannon entropy to detect high-uncertainty positions, and then performs targeted remasking conditioned on retrieved evidence. Ablations confirm that localization and correction contribute complementary gains, robust across N in {4, 8, 16}. On TriviaQA, HotpotQA, RAGTruth, and CommonsenseQA using LLaDA-8B and Dream-7B, OSCAR enhances generation quality by significantly reducing hallucinated content and improving factual accuracy through uncertainty-guided remasking, which also facilitates more effective integration of retrieved evidence. Its native entropy-based uncertainty signal surpasses that of specialized trained detectors, highlighting an inherent capacity of diffusion language models to identify factual uncertainty that is not present in the sequential token commitment structure of autoregressive models. We are releasing the codebase1 to support future research on localization and uncertainty-aware generation in DLMs.

Diffusion language models (DLMs) enable parallel, non-autoregressive text generation, yet existing DLM mixture-of-experts (MoE) models inherit token-choice (TC) routing from autoregressive systems, leading to load imbalance and rigid computation allocation. We show that expert-choice (EC) routing is a better fit for DLMs: it provides deterministic load balancing by design, yielding higher throughput and faster convergence than TC. Building on the property that EC capacity is externally controllable, we introduce timestep-dependent expert capacity, which varies expert allocation according to the denoising step. We find that allocating more capacity to low-mask-ratio steps consistently achieves the best performance under matched FLOPs, and provide a mechanistic explanation: tokens in low-mask-ratio contexts exhibit an order-of-magnitude higher learning efficiency, so concentrating compute on these steps yields the largest marginal return. Finally, we show that existing pretrained TC DLMs can be retrofitted to EC by replacing only the router, achieving faster convergence and improved accuracy across diverse downstream tasks. Together, these results establish EC routing as a superior paradigm for DLM MoE models and demonstrate that computation in DLMs can be treated as an adaptive policy rather than a fixed architectural constant. Code is available at https://github.com/zhangshuibai/EC-DLM.

Large language model (LLM) inference increasingly depends on multi-GPU execution, yet existing inference parallelization strategies require layer-wise inter-rank synchronization, making end-to-end performance sensitive to workload imbalance. We present DWDP (Distributed Weight Data Parallelism), an inference parallelization strategy that preserves data-parallel execution while offloading MoE weights across peer GPUs and fetching missing experts on demand. By removing collective inter-rank synchronization, DWDP allows each GPU to progress independently. We further address the practical overheads of this design with two optimizations for split-weight management and asynchronous remote-weight prefetch. Implemented in TensorRT-LLM and evaluated with DeepSeek-R1 on GB200 NVL72, DWDP improves end-to-end output TPS/GPU by 8.8% at comparable TPS/user in the 20-100 TPS/user serving range under 8K input sequence length and 1K output sequence length.

Morphological traits are physical characteristics of biological organisms that provide vital clues on how organisms interact with their environment. Yet extracting these traits remains a slow, expert-driven process, limiting their use in large-scale ecological studies. A major bottleneck is the absence of high-quality datasets linking biological images to trait-level annotations. In this work, we demonstrate that sparse autoencoders trained on foundation-model features yield monosemantic, spatially grounded neurons that consistently activate on meaningful morphological parts. Leveraging this property, we introduce a trait annotation pipeline that localizes salient regions and uses vision-language prompting to generate interpretable trait descriptions. Using this approach, we construct Bioscan-Traits, a dataset of 80K trait annotations spanning 19K insect images from BIOSCAN-5M. Human evaluation confirms the biological plausibility of the generated morphological descriptions. We assess design sensitivity through a comprehensive ablation study, systematically varying key design choices and measuring their impact on the quality of the resulting trait descriptions. By annotating traits with a modular pipeline rather than prohibitively expensive manual efforts, we offer a scalable way to inject biologically meaningful supervision into foundation models, enable large-scale morphological analyses, and bridge the gap between ecological relevance and machine-learning practicality.

Vision-language-action (VLA) models have shown strong performance in robotic manipulation, yet their robustness to physically realizable adversarial attacks remains underexplored. Existing studies reveal vulnerabilities through language perturbations and 2D visual attacks, but these attack surfaces are either less representative of real deployment or limited in physical realism. In contrast, adversarial 3D textures pose a more physically plausible and damaging threat, as they are naturally attached to manipulated objects and are easier to deploy in physical environments. Bringing adversarial 3D textures to VLA systems is nevertheless nontrivial. A central obstacle is that standard 3D simulators do not provide a differentiable optimization path from the VLA objective function back to object appearance, making it difficult to optimize through an end-to-end manner. To address this, we introduce Foreground-Background Decoupling (FBD), which enables differentiable texture optimization through dual-renderer alignment while preserving the original simulation environment. To further ensure that the attack remains effective across long-horizon and diverse viewpoints in the physical world, we propose Trajectory-Aware Adversarial Optimization (TAAO), which prioritizes behaviorally critical frames and stabilizes optimization with a vertex-based parameterization. Built on these designs, we present Tex3D, the first framework for end-to-end optimization of 3D adversarial textures directly within the VLA simulation environment. Experiments in both simulation and real-robot settings show that Tex3D significantly degrades VLA performance across multiple manipulation tasks, achieving task failure rates of up to 96.7\%. Our empirical results expose critical vulnerabilities of VLA systems to physically grounded 3D adversarial attacks and highlight the need for robustness-aware training.

This paper presents a systematic, cost-aware evaluation of large language models (LLMs) for receipt-item categorisation within a production-oriented classification framework. We compare four instruction-tuned models available through AWS Bedrock: Claude 3.7 Sonnet, Claude 4 Sonnet, Mixtral 8x7B Instruct, and Mistral 7B Instruct. The aim of the study was (1) to assess performance across accuracy, response stability, and token-level cost, and (2) to investigate what prompting methods, zero-shot or few-shot, are especially appropriate both in terms of accuracy and in terms of incurred costs. Results of our experiments demonstrated that Claude 3.7 Sonnet achieves the most favourable balance between classification accuracy and cost efficiency.

In reinforcement learning (RL), temporal difference (TD) errors are widely adopted for optimizing value and policy functions. However, since the TD error is defined by a bootstrap method, its computation tends to be noisy and destabilize learning. Heuristics to improve the accuracy of TD errors, such as target networks and ensemble models, have been introduced so far. While these are essential approaches for the current deep RL algorithms, they cause side effects like increased computational cost and reduced learning efficiency. Therefore, this paper revisits the TD learning algorithm based on control as inference, deriving a novel algorithm capable of robust learning against noisy TD errors. First, the distribution model of optimality, a binary random variable, is represented by a sigmoid function. Alongside forward and reverse Kullback-Leibler divergences, this new model derives a robust learning rule: when the sigmoid function saturates with a large TD error probably due to noise, the gradient vanishes, implicitly excluding it from learning. Furthermore, the two divergences exhibit distinct gradient-vanishing characteristics. Building on these analyses, the optimality is decomposed into multiple levels to achieve pseudo-quantization of TD errors, aiming for further noise reduction. Additionally, a Jensen-Shannon divergence-based approach is approximately derived to inherit the characteristics of both divergences. These benefits are verified through RL benchmarks, demonstrating stable learning even when heuristics are insufficient or rewards contain noise.

Volumetric CT imaging is essential for clinical diagnosis, yet annotating 3D volumes is expensive and time-consuming, motivating self-supervised learning (SSL) from unlabeled data. However, applying SSL to 3D CT remains challenging due to the high memory cost of full-volume transformers and the anisotropic spatial structure of CT data, which is not well captured by conventional masking strategies. We propose NEMESIS, a masked autoencoder (MAE) framework that operates on local 128x128x128 superpatches, enabling memory-efficient training while preserving anatomical detail. NEMESIS introduces three key components: (i) noise-enhanced reconstruction as a pretext task, (ii) Masked Anatomical Transformer Blocks (MATB) that perform dual-masking through parallel plane-wise and axis-wise token removal, and (iii) NEMESIS Tokens (NT) for cross-scale context aggregation. On the BTCV multi-organ classification benchmark, NEMESIS with a frozen backbone and a linear classifier achieves a mean AUROC of 0.9633, surpassing fully fine-tuned SuPreM (0.9493) and VoCo (0.9387). Under a low-label regime with only 10% of available annotations, it retains an AUROC of 0.9075, demonstrating strong label efficiency. Furthermore, the superpatch-based design reduces computational cost to 31.0 GFLOPs per forward pass, compared to 985.8 GFLOPs for the full-volume baseline, providing a scalable and robust foundation for 3D medical imaging.

The use of knowledge graphs for grounding agents in real-world Q&A applications has become increasingly common. Answering complex queries often requires multi-hop reasoning and the ability to navigate vast relational structures. Standard approaches rely on prompting techniques that steer large language models to reason over raw graph context, or retrieval-augmented generation pipelines where relevant subgraphs are injected into the context. These, however, face severe limitations with enterprise-scale KGs that cannot fit in even the largest context windows available today. We present GraphWalk, a problem-agnostic, training-free, tool-based framework that allows off-the-shelf LLMs to reason through sequential graph navigation, dramatically increasing performance across different tasks. Unlike task-specific agent frameworks that encode domain knowledge into specialized tools, GraphWalk equips the LLM with a minimal set of orthogonal graph operations sufficient to traverse any graph structure. We evaluate whether models equipped with GraphWalk can compose these operations into correct multi-step reasoning chains, where each tool call represents a verifiable step creating a transparent execution trace. We first demonstrate our approach on maze traversal, a problem non-reasoning models are completely unable to solve, then present results on graphs resembling real-world enterprise knowledge graphs. To isolate structural reasoning from world knowledge, we evaluate on entirely synthetic graphs with random, non-semantic labels. Our benchmark spans 12 query templates from basic retrieval to compound first-order logic queries. Results show that tool-based traversal yields substantial and consistent gains over in-context baselines across all model families tested, with gains becoming more pronounced as scale increases, precisely where in-context approaches fail catastrophically.

The deployment of Large Language Models is constrained by the memory and bandwidth demands of static weights and dynamic Key-Value cache. SVD-based compression provides a hardware-friendly solution to reduce these costs. However, existing methods suffer from two key limitations: some are suboptimal in reconstruction error, while others are theoretically optimal but practically inefficient. In this paper, we propose Swift-SVD, an activation-aware, closed-form compression framework that simultaneously guarantees theoretical optimum, practical efficiency and numerical stability. Swift-SVD incrementally aggregates covariance of output activations given a batch of inputs and performs a single eigenvalue decomposition after aggregation, enabling training-free, fast, and optimal layer-wise low-rank approximation. We employ effective rank to analyze local layer-wise compressibility and design a dynamic rank allocation strategy that jointly accounts for local reconstruction loss and end-to-end layer importance. Extensive experiments across six LLMs and eight datasets demonstrate that Swift-SVD outperforms state-of-the-art baselines, achieving optimal compression accuracy while delivering 3-70X speedups in end-to-end compression time. Our code will be released upon acceptance.

Multi-agent systems (MAS) tackle complex tasks by distributing expertise, though this often comes at the cost of heavy coordination overhead, context fragmentation, and brittle phase ordering. Distilling a MAS into a single-agent skill can bypass these costs, but this conversion lacks a principled answer for when and what to distill. Instead, the empirical outcome is surprisingly inconsistent: skill lift ranges from a 28% improvement to a 2% degradation across metrics of the exact same task. In this work, we reveal that skill utility is governed not by the task, but by the evaluation metric. We introduce Metric Freedom ($F$), the first a priori predictor of skill utility. $F$ measures the topological rigidity of a metric's scoring landscape by quantifying how output diversity couples with score variance via a Mantel test. Guided by $F$, we propose a two-stage adaptive distillation framework. Stage 1 acts as a selective extraction mechanism, extracting tools and knowledge while discarding restrictive structures on "free" metrics to preserve exploration. Stage 2 targets computationally intensive iterative refinement exclusively toward "rigid" metrics ($F \lesssim 0.6$) to eliminate trajectory-local overfitting. Evaluating across 4 tasks, 11 datasets, and 6 metrics, $F$ strongly predicts skill utility ($ρ= -0.62$, $p < 0.05$). Strikingly, identical agent trajectories yield diametrically opposite skill lifts under rigid versus free metrics, demonstrating that skill utility is fundamentally a metric-level property. Driven by this signal, our adaptive agent matches or exceeds the original MAS while reducing cost up to 8$\times$ and latency by up to 15$\times$.

Large-scale distributed training has been a research hot spot in machine learning systems for industry and academia in recent years. However, conducting experiments without physical machines and corresponding resources is difficult. One solution is to leverage distributed training simulators, but current ones like ASTRA-sim do not support importing real-world developed models, which poses challenges for ML researchers seeking to use them. Based on this challenge, we developed ModTrans, a translator supporting format translation from any real-world model to the ASTRA-sim simulator's input, removing the barrier between machine learning experts and machine learning system researchers. The experiment results show that ModTrans's cost is negligible.

Optimization over the space of probability measures endowed with the Wasserstein-2 geometry is central to modern machine learning and mean-field modeling. However, traditional methods relying on full Wasserstein gradients often suffer from high computational overhead in high-dimensional or ill-conditioned settings. We propose a randomized coordinate descent framework specifically designed for the Wasserstein manifold, introducing both Random Wasserstein Coordinate Descent (RWCD) and Random Wasserstein Coordinate Proximal{-Gradient} (RWCP) for composite objectives. By exploiting coordinate-wise structures, our methods adapt to anisotropic objective landscapes where full-gradient approaches typically struggle. We provide a rigorous convergence analysis across various landscape geometries, establishing guarantees under non-convex, Polyak-Łojasiewicz, and geodesically convex conditions. Our theoretical results mirror the classic convergence properties found in Euclidean space, revealing a compelling symmetry between coordinate descent on vectors and on probability measures. The developed techniques are inherently adaptive to the Wasserstein geometry and offer a robust analytical template that can be extended to other optimization solvers within the space of measures. Numerical experiments on ill-conditioned energies demonstrate that our framework offers significant speedups over conventional full-gradient methods.

As safety concerns around large language models (LLMs) grow, understanding the internal mechanisms underlying refusal behavior has become increasingly important. Recent work has studied this behavior by identifying internal features associated with refusal and manipulating them to induce compliance with harmful requests. However, existing refusal feature selection methods rely on how strongly features activate on harmful prompts, which tends to capture superficial signals rather than the causal factors underlying the refusal decision. We propose CRaFT, a circuit-guided refusal feature selection framework that ranks features by their influence on the model's refusal-compliance decision using prompts near the refusal boundary. On Gemma-3-1B-it, CRaFT improves attack success rate (ASR) from 6.7% to 48.2% and outperforms baseline methods across multiple jailbreak benchmarks. These results suggest that circuit influence is a more reliable criterion than activation magnitude for identifying features that causally mediate refusal behavior.

We investigate training strategies that co-develop in-context learning (ICL) and in-weights learning (IWL), and the ability to switch between them based on context relevance. Although current LLMs exhibit both modes, standard task-specific fine-tuning often erodes ICL, motivating IC-Train - fine-tuning with in-context examples. Prior work has shown that emergence of ICL after IC-Train depends on factors such as task diversity and training duration. In this paper we show that the similarity structure between target inputs and context examples also plays an important role. Random context leads to loss of ICL and IWL dominance, while only similar examples in context causes ICL to degenerate to copying labels without regard to relevance. To address this, we propose a simple Contrastive-Context which enforces two types of contrasts: (1) mix of similar and random examples within a context to evolve a correct form of ICL, and (2) varying grades of similarity across contexts to evolve ICL-IWL mixtures. We present insights on the importance of such contrast with theoretical analysis of a minimal model. We validate with extensive empirical evaluation on four LLMs and several tasks. Diagnostic probes confirm that contrasted contexts yield stable ICL-IWL mixtures, avoiding collapse into pure ICL, IWL, or copying.

Multimodal Large Language Models (MLLMs) have recently demonstrated promising capabilities in multimodal coding tasks such as chart-to-code generation. However, existing methods primarily rely on supervised fine-tuning (SFT), which requires the model to learn code patterns through chart-code pairs but does not expose the model to a code execution environment. Moreover, while self-correction through execution feedback offers a potential route to improve coding quality, even state-of-the-art MLLMs have been shown to struggle with effective self-correction. In this work, we introduce MM-ReCoder, a chart-to-code generation model trained with reinforcement learning (RL) and equipped with self-correction ability. We propose a two-stage multi-turn self-correction RL strategy based on Group Relative Policy Optimization (GRPO). The first stage enhances the model's self-correction ability via rolling out a shared first turn, while the second stage improves the coding capability with full-trajectory optimization. MM-ReCoder learns to produce more accurate and executable code through the interaction with the environment and by iteratively correcting its own outputs. Our results on three chart-to-code benchmarks demonstrate the state-of-the-art performance of MM-ReCoder.

Memory-Augmented Generation (MAG) extends large language models with external memory to support long-context reasoning, but existing approaches universally treat memory as an external service that agents call into, delegating storage to separate pipelines of chunking, embedding, and graph extraction. This architectural separation means the system that stores knowledge does not understand it, leading to semantic drift between what the agent intended to remember and what the pipeline actually captured, loss of coordination context across agents, and fragile recovery after failures. In this paper, we propose ByteRover, an agent-native memory architecture that inverts the memory pipeline: the same LLM that reasons about a task also curates, structures, and retrieves knowledge. ByteRover represents knowledge in a hierarchical Context Tree, a file-based knowledge graph organized as Domain, Topic, Subtopic, and Entry, where each entry carries explicit relations, provenance, and an Adaptive Knowledge Lifecycle (AKL) with importance scoring, maturity tiers, and recency decay. Retrieval uses a 5-tier progressive strategy that resolves most queries at sub-100 ms latency without LLM calls, escalating to agentic reasoning only for novel questions. Experiments on LoCoMo and LongMemEval demonstrate that ByteRover achieves state-of-the-art accuracy on LoCoMo and competitive results on LongMemEval while requiring zero external infrastructure, no vector database, no graph database, no embedding service, with all knowledge stored as human-readable markdown files on the local filesystem.

Traditional RL algorithms like Proximal Policy Optimization (PPO) typically train on the entire rollout buffer, operating under the assumption that all generated episodes provide a beneficial optimization signal. However, these episodes frequently contain noisy or unfaithful reasoning, which can degrade model performance and slow down training. In this paper, we propose \textbf{Influence-Guided PPO (I-PPO)}, a novel framework that integrates data attribution into the RL post-training loop. By calculating an influence score for each episode using a gradient-based approximation, I-PPO identifies and eliminates episodes that are anti-aligned with a validation gradient. Our experiments demonstrate that I-PPO consistently outperforms SFT and PPO baselines. We show that our filtering process acts as an intrinsic early stopping mechanism, accelerating training efficiency while effectively reducing unfaithful CoT reasoning.

Seizure detection from EEG signals is highly challenging due to complex spatiotemporal dynamics and extreme inter-patient variability. To model them, recent methods construct dynamic graphs via statistical correlations, predefined similarity measures, or implicit learning, yet rarely account for EEG's noisy nature. Consequently, these graphs usually contain redundant or task-irrelevant connections, undermining model performance even with state-of-the-art architectures. In this paper, we present a new perspective for EEG seizure detection: jointly learning denoised dynamic graph structures and informative spatial-temporal representations guided by the Information Bottleneck (IB). Unlike prior approaches, our graph constructor explicitly accounts for the noisy characteristics of EEG data, producing compact and reliable connectivity patterns that better support downstream seizure detection. To further enhance representation learning, we employ a self-supervised Graph Masked AutoEncoder that reconstructs masked EEG signals based on dynamic graph context, promoting structure-aware and compact representations aligned with the IB principle. Bringing things together, we introduce Information Bottleneck-guided EEG SeizuRE DetectioN via SElf-Supervised Learning (IRENE), which explicitly learns dynamic graph structures and interpretable spatial-temporal EEG representations. IRENE addresses three core challenges: (i) Identifying the most informative nodes and edges; (ii) Explaining seizure propagation in the brain network; and (iii) Enhancing robustness against label scarcity and inter-patient variability. Extensive experiments on benchmark EEG datasets demonstrate that our method outperforms state-of-the-art baselines in seizure detection and provides clinically meaningful insights into seizure dynamics. The source code is available at https://github.com/LabRAI/IRENE.

How do LLMs decide what to teach next: by reasoning about a learner's knowledge, or by using simpler rules of thumb? We test this in a controlled task previously used to study human teaching strategies. On each trial, a teacher LLM sees a hypothetical learner's trajectory through a reward-annotated directed graph and must reveal a single edge so the learner would choose a better path if they replanned. We run a range of LLMs as simulated teachers and fit their trial-by-trial choices with the same cognitive models used for humans: a Bayes-Optimal teacher that infers which transitions the learner is missing (inverse planning), weaker Bayesian variants, heuristic baselines (e.g., reward based), and non-mentalizing utility models. In a baseline experiment matched to the stimuli presented to human subjects, most LLMs perform well, show little change in strategy over trials, and their graph-by-graph performance is similar to that of humans. Model comparison (BIC) shows that Bayes-Optimal teaching best explains most models' choices. When given a scaffolding intervention, models follow auxiliary inference- or reward-focused prompts, but these scaffolds do not reliably improve later teaching on heuristic-incongruent test graphs and can sometimes reduce performance. Overall, cognitive model fits provide insight into LLM tutoring policies and show that prompt compliance does not guarantee better teaching decisions.

We introduce ThinkTwice, a simple two-phase framework that jointly optimizes LLMs to solve reasoning problems and refine the answers, based on Group Relative Policy Optimization (GRPO). In each pair of training steps, ThinkTwice first optimizes the model on solving reasoning problems, then optimizes it on refining its own solutions to the same problems, using the same binary correctness reward in both phases without correctness signals or critique annotations. Across five mathematical reasoning benchmarks and two model families including Qwen3-4B and Olmo3-7B, ThinkTwice substantially improves both reasoning and refinement performance over competitive online policy optimization baselines. Specifically, on Qwen3-4B, ThinkTwice outperforms GRPO on AIME by 5 percentage points before refinement and by 11.5 points after one self-refinement step, measured by pass@4. Analysis of the training dynamics of ThinkTwice reveals an implicit rectify-then-fortify curriculum: refinement predominantly corrects errors early in training and naturally shifts toward preserving already-correct solutions as the model improves, yielding a more rectified reward signal. Our work establishes joint training of reasoning and self-refinement as a principled and effective methodology for RLVR.

Automating the translation of Operations Research (OR) problems from natural language to executable models is a critical challenge. While Large Language Models (LLMs) have shown promise in linear tasks, they suffer from severe performance degradation in real-world nonlinear scenarios due to semantic misalignment between mathematical formulations and solver codes, as well as unstable information extraction. In this study, we introduce NED-Tree, a systematic framework designed to bridge the semantic gap. NED-Tree employs (a) a sentence-by-sentence extraction strategy to ensure robust parameter mapping and traceability; and (b) a recursive tree-based structure that adaptively decomposes complex nonlinear terms into solver-compatible sub-elements. Additionally, we present NEXTOR, a novel benchmark specifically designed for complex nonlinear, extensive-constraint OR problems. Experiments across 10 benchmarks demonstrate that NED-Tree establishes a new state-of-the-art with 72.51% average accuracy, NED-Tree is the first framework that drives LLMs to resolve nonlinear modeling difficulties through element decomposition, achieving alignment between modeling semantics and code semantics. The NED-Tree framework and benchmark are accessible in the anonymous repository https://anonymous.4open.science/r/NORA-NEXTOR.

Uncertainty propagation in high-dimensional nonlinear dynamic structural systems is pivotal in state-of-the-art performance-based design and risk assessment, where uncertainties from both excitations and structures, i.e., the aleatoric uncertainty, must be considered. This poses a significant challenge due to heavy computational demands. Machine learning techniques are thus introduced as metamodels to alleviate this burden. However, the "black box" nature of Machine learning models underscores the necessity of avoiding overly confident predictions, particularly when data and training efforts are insufficient. This creates a need, in addition to considering the aleatoric uncertainty, of estimating the uncertainty related to the prediction confidence, i.e., epistemic uncertainty, for machine learning-based metamodels. We developed a probabilistic metamodeling technique based on a variational long short-term memory (LSTM) with augmented inputs to simultaneously capture aleatoric and epistemic uncertainties. Key random system parameters are treated as augmented inputs alongside excitation series carrying record-to-record variability to capture the full range of aleatoric uncertainty. Meanwhile, epistemic uncertainty is effectively approximated via the Monte Carlo dropout scheme. Unlike computationally expensive full Bayesian approaches, this method incurs negligible additional training costs while enabling nearly cost-free uncertainty simulation. The proposed technique is demonstrated through multiple case studies involving stochastic seismic or wind excitations. Results show that the calibrated metamodels accurately reproduce nonlinear response time histories and provide confidence bounds indicating the associated epistemic uncertainty.

Open-vocabulary human-object interaction (HOI) detection is a step towards building scalable systems that generalize to unseen interactions in real-world scenarios and support grounded multimodal systems that reason about human-object relationships. However, standard evaluation metrics, such as mean Average Precision (mAP), treat HOI classes as discrete categorical labels and fail to credit semantically valid but lexically different predictions (e.g., "lean on couch" vs. "sit on couch"), limiting their applicability for evaluating open-vocabulary predictions that go beyond any predefined set of HOI labels. We introduce SHOE (Semantic HOI Open-Vocabulary Evaluation), a new evaluation framework that incorporates semantic similarity between predicted and ground-truth HOI labels. SHOE decomposes each HOI prediction into its verb and object components, estimates their semantic similarity using the average of multiple large language models (LLMs), and combines them into a similarity score to evaluate alignment beyond exact string match. This enables a flexible and scalable evaluation of both existing HOI detection methods and open-ended generative models using standard benchmarks such as HICO-DET. Experimental results show that SHOE scores align more closely with human judgments than existing metrics, including LLM-based and embedding-based baselines, achieving an agreement of 85.73% with the average human ratings. Our work underscores the need for semantically grounded HOI evaluation that better mirrors human understanding of interactions. We will release our evaluation metric to the public to facilitate future research.

Multimodal tabular-image fusion is an emerging task that has received increasing attention in various domains. However, existing methods may be hindered by gradient conflicts between modalities, misleading the optimization of the unimodal learner. In this paper, we propose a novel Gradient-Aligned Alternating Learning (GAAL) paradigm to address this issue by aligning modality gradients. Specifically, GAAL adopts an alternating unimodal learning and shared classifier to decouple the multimodal gradient and facilitate interaction. Furthermore, we design uncertainty-based cross-modal gradient surgery to selectively align cross-modal gradients, thereby steering the shared parameters to benefit all modalities. As a result, GAAL can provide effective unimodal assistance and help boost the overall fusion performance. Empirical experiments on widely used datasets reveal the superiority of our method through comparison with various state-of-the-art (SoTA) tabular-image fusion baselines and test-time tabular missing baselines. The source code is available at https://github.com/njustkmg/ICME26-GAAL.

We extend the recent latent recurrent modeling to sequential input streams. By interleaving fast, recurrent latent updates with self-organizational ability between slow observation updates, our method facilitates the learning of stable internal structures that evolve alongside the input. This mechanism allows the model to maintain coherent and clustered representations over long horizons, improving out-of-distribution generalization in reinforcement learning and algorithmic tasks compared to sequential baselines such as LSTM, state space models, and Transformer variants.

Large language models deployed in supportive or advisory roles must balance helpfulness with preservation of user autonomy, yet standard alignment methods primarily optimize for helpfulness and harmlessness without explicitly modeling relational risks such as dependency reinforcement, overprotection, or coercive guidance. We introduce Care-Conditioned Neuromodulation (CCN), a state-dependent control framework in which a learned scalar signal derived from structured user state and dialogue context conditions response generation and candidate selection. We formalize this setting as an autonomy-preserving alignment problem and define a utility function that rewards autonomy support and helpfulness while penalizing dependency and coercion. We also construct a benchmark of relational failure modes in multi-turn dialogue, including reassurance dependence, manipulative care, overprotection, and boundary inconsistency. On this benchmark, care-conditioned candidate generation combined with utility-based reranking improves autonomy-preserving utility by +0.25 over supervised fine-tuning and +0.07 over preference optimization baselines while maintaining comparable supportiveness. Pilot human evaluation and zero-shot transfer to real emotional-support conversations show directional agreement with automated metrics. These results suggest that state-dependent control combined with utility-based selection is a practical approach to multi-objective alignment in autonomy-sensitive dialogue.

In LLM training, normalization layers and optimizers are typically treated as independent design choices. In a 3x2 factorial at 1B parameters and 1000 training steps, we show this assumption can fail: Dynamic Erf (Derf; Chen & Liu, 2025) suffers a large negative interaction with Muon (Jordan, 2024), with its gap to RMSNorm growing from +0.31 nats under AdamW to +0.97 under Muon, approximately three times larger. Dynamic Tanh (DyT; Zhu et al., 2025), included as a bounded-normalizer control, shows no such penalty. Our evidence points to two failure modes of erf under Muon's faster spectral-norm growth: saturation (lossy compression) and scale blindness (discarding activation magnitude). An EMA-blend that reintroduces running scale estimates recovers ~84% of the gap. Separately, reducing Derf's alpha from its published default (0.5 to 0.3) recovers ~80% by keeping erf in its near-linear regime, where it approximately preserves relative scale; this setting is not the published default of Chen & Liu (2025). Using Derf's published default alpha with Muon incurs a 0.66-nat interaction penalty without producing NaNs or divergence, making the failure easy to miss in short pilot runs.

Voice cloning is often evaluated in terms of overall quality, but less is known about accent preservation and its perceptual consequences. We compare standard and heavily accented Mandarin speech and their voice clones using a combined computational and perceptual design. Embedding-based analyses show no reliable accented-standard difference in original-clone distances across systems. In the perception study, clones are rated as more similar to their originals for standard than for accented speakers, and intelligibility increases from original to clone, with a larger gain for accented speech. These results show that accent variation can shape perceived identity match and intelligibility in voice cloning even when it is not reflected in an off-the-shelf speaker-embedding distance, and they motivate evaluating speaker identity preservation and accent preservation as separable dimensions.

We present ReFlow, a unified framework for monocular dynamic scene reconstruction that learns 3D motion in a novel self-correction manner from raw video. Existing methods often suffer from incomplete scene initialization for dynamic regions, leading to unstable reconstruction and motion estimation, which often resorts to external dense motion guidance such as pre-computed optical flow to further stabilize and constrain the reconstruction of dynamic components. However, this introduces additional complexity and potential error propagation. To address these issues, ReFlow integrates a Complete Canonical Space Construction module for enhanced initialization of both static and dynamic regions, and a Separation-Based Dynamic Scene Modeling module that decouples static and dynamic components for targeted motion supervision. The core of ReFlow is a novel self-correction flow matching mechanism, consisting of Full Flow Matching to align 3D scene flow with time-varying 2D observations, and Camera Flow Matching to enforce multi-view consistency for static objects. Together, these modules enable robust and accurate dynamic scene reconstruction. Extensive experiments across diverse scenarios demonstrate that ReFlow achieves superior reconstruction quality and robustness, establishing a novel self-correction paradigm for monocular 4D reconstruction.

Recent advances in persona-centric memory have revealed the powerful capability of multi-agent systems in managing persona memory, especially in conversational scenarios. However, these complex frameworks often suffer from information loss and are fragile across varying scenarios, resulting in suboptimal performance. In this paper, we propose DeltaMem, an agentic memory management system that formulates persona-centric memory management as an end-to-end task within a single-agent setting. To further improve the performance of our agentic memory manager, we draw inspiration from the evolution of human memory and synthesize a user-assistant dialogue dataset along with corresponding operation-level memory updating labels. Building on this, we introduce a novel Memory-based Levenshtein Distance to formalize the memory updating reward, and propose a tailored reinforcement learning framework to further enhance the management capabilities of DeltaMem. Extensive experiments show that both training-free and RL-trained DeltaMem outperform all product-level baselines across diverse long-term memory benchmarks, including LoCoMo, HaluMem, and PersonaMem.

Binary Function Similarity Detection (BFSD) is a core problem in software security, supporting tasks such as vulnerability analysis, malware classification, and patch provenance. In the past few decades, numerous models and tools have been developed for this application; however, due to the lack of a comprehensive universal benchmark in this field, researchers have struggled to compare different models effectively. Existing datasets are limited in scope, often focusing on a narrow set of transformations or types of binaries, and fail to reflect the full diversity of real-world applications. We introduce EXHIB, a benchmark comprising five realistic datasets collected from the wild, each highlighting a distinct aspect of the BFSD problem space. We evaluate 9 representative models spanning multiple BFSD paradigms on EXHIB and observe performance degradations of up to 30% on firmware and semantic datasets compared to standard settings, revealing substantial generalization gaps. Our results show that robustness to low- and mid-level binary variations does not generalize to high-level semantic differences, underscoring a critical blind spot in current BFSD evaluation practices.

Denoising generative models deliver high-fidelity generation but remain bottlenecked by inference latency due to the many iterative denoiser calls required during sampling. Training-free acceleration methods reduce latency by either sparsifying the model architecture or shortening the sampling trajectory. Current training-free acceleration methods are more complex than necessary: higher-order predictors amplify error under aggressive speedups, and architectural modifications hinder deployment. Beyond 2x acceleration, step skipping creates structural scarcity -- at most one fresh evaluation per local window -- leaving the computed output and its backward difference as the only causally grounded information. Based on this, we propose ZEUS, an acceleration method that predicts reduced denoiser evaluations using a second-order predictor, and stabilizes aggressive consecutive skipping with an interleaved scheme that avoids back-to-back extrapolations. ZEUS adds essentially zero overhead, no feature caches, and no architectural modifications, and it is compatible with different backbones, prediction objectives, and solver choices. Across image and video generation, ZEUS consistently improves the speed-fidelity performance over recent training-free baselines, achieving up to 3.2x end-to-end speedup while maintaining perceptual quality. Our code is available at: https://github.com/Ting-Justin-Jiang/ZEUS.

The latent space of generative modeling is long dominated by the VAE encoder. The latents from the pretrained representation encoders (e.g., DINO, SigLIP, MAE) are previously considered inappropriate for generative modeling. Recently, RAE method lights the hope and reveals that the representation autoencoder can also achieve competitive performance as the VAE encoder. However, the integration of representation autoencoder into continuous autoregressive (AR) models, remains largely unexplored. In this work, we investigate the challenges of employing high-dimensional representation autoencoders within the AR paradigm, denoted as \textit{RAE-AR}. We focus on the unique properties of AR models and identify two primary hurdles: complex token-wise distribution modeling and the high-dimensionality amplified training-inference gap (exposure bias). To address these, we introduce token simplification via distribution normalization to ease modeling difficulty and improve convergence. Furthermore, we enhance prediction robustness by incorporating Gaussian noise injection during training to mitigate exposure bias. Our empirical results demonstrate that these modifications substantially bridge the performance gap, enabling representation autoencoder to achieve results comparable to traditional VAEs on AR models. This work paves the way for a more unified architecture across visual understanding and generative modeling.

Large language models have been adopted in the medical domain for clinical documentation to reduce clinician burden. However, studies have reported that LLMs often "forget" a significant amount of instruction-following ability when fine-tuned using a task-specific medical dataset, a critical challenge in adopting general-purpose LLMs for clinical applications. This study presents a model merging framework to efficiently adapt general-purpose LLMs to the medical domain by countering this forgetting issue. By merging a clinical foundation model (GatorTronLlama) with a general instruct model (Llama-3.1-8B-Instruct) via interpolation-based merge methods, we seek to derive a domain-adapted model with strong performance on clinical tasks while retaining instruction-following ability. Comprehensive evaluation across medical benchmarks and five clinical generation tasks (e.g., radiology and discharge summarization) shows that merged models can effectively mitigate catastrophic forgetting, preserve clinical domain expertise, and retain instruction-following ability. In addition, our model merging strategies demonstrate training efficiency, achieving performance on par with fully fine-tuned baselines under severely constrained supervision (e.g., 64-shot vs. 256-shot). Consequently, weight-space merging constitutes a highly scalable solution for adapting open-source LLMs to clinical applications, facilitating broader deployment in resource-constrained healthcare environments.

Web agents based on large language models (LLMs) rely on observations of web pages -- commonly represented as HTML -- as the basis for identifying available actions and planning subsequent steps. Prior work has treated the verbosity of HTML as an obstacle to performance and adopted observation reduction as a standard practice. We revisit this trend and demonstrate that the optimal observation representation depends on model capability and thinking token budget: (1) compact observations (accessibility trees) are preferable for lower-capability models, while detailed observations (HTML) are advantageous for higher-capability models; moreover, increasing thinking tokens further amplifies the benefit of HTML. (2) Our error analysis suggests that higher-capability models exploit layout information in HTML for better action grounding, while lower-capability models suffer from increased hallucination under longer inputs. We also find that incorporating observation history improves performance across most models and settings, and a diff-based representation offers a token-efficient alternative. Based on these findings, we suggest practical guidelines: adaptively select observation representations based on model capability and thinking token budget, and incorporate observation history using diff-based representations.

Large language model (LLM) agents are increasingly deployed for complex tool-orchestration tasks, yet existing benchmarks fail to capture the rigorous demands of industrial domains where incorrect decisions carry significant safety and financial consequences. To address this critical gap, we introduce PHMForge, the first comprehensive benchmark specifically designed to evaluate LLM agents on Prognostics and Health Management (PHM) tasks through realistic interactions with domain-specific MCP servers. Our benchmark encompasses 75 expert-curated scenarios spanning 7 industrial asset classes (turbofan engines, bearings, electric motors, gearboxes, aero-engines) across 5 core task categories: Remaining Useful Life (RUL) Prediction, Fault Classification, Engine Health Analysis, Cost-Benefit Analysis, and Safety/Policy Evaluation. To enable rigorous evaluation, we construct 65 specialized tools across two MCP servers and implement execution-based evaluators with task-commensurate metrics: MAE/RMSE for regression, F1-score for classification, and categorical matching for health assessments. Through extensive evaluation of leading frameworks (ReAct, Cursor Agent, Claude Code) paired with frontier LLMs (Claude Sonnet 4.0, GPT-4o, Granite-3.0-8B), we find that even top-performing configurations achieve only 68\% task completion, with systematic failures in tool orchestration (23\% incorrect sequencing), multi-asset reasoning (14.9 percentage point degradation), and cross-equipment generalization (42.7\% on held-out datasets). We open-source our complete benchmark, including scenario specifications, ground truth templates, tool implementations, and evaluation scripts, to catalyze research in agentic industrial AI.

Current Large Language Model (LLM) approaches for information extraction (IE) in the healthy food policy domain are often hindered by various factors, including misinformation, specifically hallucinations, misclassifications, and omissions that result from the structural diversity and inconsistency of policy documents. To address these limitations, this study proposes a role-based LLM framework that automates the IE from unstructured policy data by assigning specialized roles: an LLM policy analyst for metadata and mechanism classification, an LLM legal strategy specialist for identifying complex legal approaches, and an LLM food system expert for categorizing food system stages. This framework mimics expert analysis workflows by incorporating structured domain knowledge, including explicit definitions of legal mechanisms and classification criteria, into role-specific prompts. We evaluate the framework using 608 healthy food policies from the Healthy Food Policy Project (HFPP) database, comparing its performance against zero-shot, few-shot, and chain-of-thought (CoT) baselines using Llama-3.3-70B. Our proposed framework demonstrates superior performance in complex reasoning tasks, offering a reliable and transparent methodology for automating IE from health policies.

Benchmarks that reflect production workloads are better for evaluating AI coding agents in industrial settings, yet existing benchmarks differ from real usage in programming language distribution, prompt style and codebase structure. This paper presents a methodology for curating production-derived benchmarks, illustrated through ProdCodeBench - a benchmark built from real sessions with a production AI coding assistant. We detail our data collection and curation practices including LLM-based task classification, test relevance validation, and multi-run stability checks which address challenges in constructing reliable evaluation signals from monorepo environments. Each curated sample consists of a verbatim prompt, a committed code change and fail-to-pass tests spanning seven programming languages. Our systematic analysis of four foundation models yields solve rates from 53.2% to 72.2% revealing that models making greater use of work validation tools, such as executing tests and invoking static analysis, achieve higher solve rates. This suggests that iterative verification helps achieve effective agent behavior and that exposing codebase-specific verification mechanisms may significantly improve the performance of externally trained agents operating in unfamiliar environments. We share our methodology and lessons learned to enable other organizations to construct similar production-derived benchmarks.

We give a short linear--algebraic proof of the inequality \[ \|x\|_1\,\|x\|_\infty \le \frac{1+\sqrt{p}}{2}\,\|x\|_2^2, \] valid for every \(x\in\mathbb{R}^p\). This inequality relates three fundamental norms on finite-dimensional spaces and has applications in optimization and numerical analysis. Our proof exploits the determinantal structure of a parametrized family of quadratic forms, and we show the constant $(1+\sqrt{p})/2$ is optimal.

Electrocardiograms (ECGs) are among the most widely used diagnostic tools for cardiovascular diseases, and a large amount of ECG data worldwide appears only in image form. However, most existing automated ECG analysis methods rely on access to raw signal recordings, limiting their applicability in real-world and resource-constrained settings. In this paper, we present ECG-Scan, a self-supervised framework for learning clinically generalized representations from ECG images through dual physiological-aware alignments: 1) Our approach optimizes image representation learning using multimodal contrastive alignment between image and gold-standard signal-text modalities. 2) We further integrate domain knowledge via soft-lead constraints, regularizing the reconstruction process and improving signal lead inter-consistency. Extensive benchmarking across multiple datasets and downstream tasks demonstrates that our image-based model achieves superior performance compared to existing image baselines and notably narrows the gap between ECG image and signal analysis. These results highlight the potential of self-supervised image modeling to unlock large-scale legacy ECG data and broaden access to automated cardiovascular diagnostics.

The increasing scale and complexity of mobile applications make automated GUI exploration essential for software quality assurance. However, existing methods often neglect state dependencies between test fragments, which leads to redundant exploration and prevents access to deep application states. We introduce EpiDroid, a black-box, pluggable framework that augments existing explorers through semantic state dependency awareness. EpiDroid distills raw traces into stable test fragments to extract underlying dependencies. It then employs a Recomposition-Replay paradigm to perform impact reasoning via LLM and deterministic replay on high-value mutable state elements. Through iterative feedback, EpiDroid refines the state-dependency graph to systematically reach deep application states. We integrated EpiDroid into both industrial and state-of-the-art research tools and evaluated it on 20 real-world apps. The results show that EpiDroid consistently improves the performance of all baselines, increasing average code coverage by 10--28\% and delivering 3--4$\times$ more coverage gain compared to continuing the baselines alone from the same starting point. This demonstrates that dependency-guided recomposition unlocks deep states that forward exploration cannot access, irrespective of additional budget.

Traditional social science research often requires designing complex experiments across vast methodological spaces and depends on real human participants, making it labor-intensive, costly, and difficult to scale. Here we present S-Researcher, an LLM-agent-based platform that assists researchers in conducting social science research more efficiently and at greater scale by "siliconizing" both the research process and the participant pool. To build S-Researcher, we first develop YuLan-OneSim, a large-scale social simulation system designed around three core requirements: generality via auto-programming from natural language to executable scenarios, scalability via a distributed architecture supporting up to 100,000 concurrent agents, and reliability via feedback-driven LLM fine-tuning. Leveraging this system, S-Researcher supports researchers in designing social experiments, simulating human behavior with LLM agents, analyzing results, and generating reports, forming a complete human-AI collaborative research loop in which researchers retain oversight and intervention at every stage. We operationalize LLM simulation research paradigms into three canonical reasoning modes (induction, deduction, and abduction) and validate S-Researcher through systematic case studies: inductive reproduction of cultural dynamics consistent with Axelrod's theory, deductive testing of competing hypotheses on teacher attention validated against survey data, and abductive identification of a cooperation mechanism in public goods games confirmed by human experiments. S-Researcher establishes a new human--AI collaborative paradigm for social science, in which computational simulation augments human researchers to accelerate discovery across the full spectrum of social inquiry.

Benchmarks driven by test suites, notably SWE-bench, have become the de facto standard for measuring the effectiveness of automated issue-resolution agents: a generated patch is accepted whenever it passes the accompanying regression tests. In practice, however, insufficiently strong test suites can admit plausible yet semantically incorrect patches, inflating reported success rates. We introduce STING, a framework for targeted test augmentation that uses semantically altered program variants as diagnostic stressors to uncover and repair weaknesses in benchmark regression suites. Variants of the ground-truth patch that still pass the existing tests reveal under-constrained behaviors; these gaps then guide the generation of focused regression tests. A generated test is retained only if it (i) passes on the ground-truth patch, (ii) fails on at least one variant that survived the original suite, and (iii) remains valid under behavior-preserving transformations designed to guard against overfitting. Applied to SWE-bench Verified, STING finds that 77% of instances contain at least one surviving variant. STING produces 1,014 validated tests spanning 211 instances and increases patch-region line and branch coverage by 10.8% and 9.5%, respectively. Re-assessing the top-10 repair agents with the strengthened suites lowers their resolved rates by 4.2%-9.0%, revealing that a substantial share of previously passing patches exploit weaknesses in the benchmark tests rather than faithfully implementing the intended fix. These results underscore that reliable benchmark evaluation depends not only on patch generation, but equally on test adequacy.

Instruction-based unlearning has proven effective for modifying the behavior of large language models at inference time, but whether this paradigm extends to other generative models remains unclear. In this work, we investigate instruction-based unlearning in diffusion-based image generation models and show, through controlled experiments across multiple concepts and prompt variants, that diffusion models systematically fail to suppress targeted concepts when guided solely by natural-language unlearning instructions. By analyzing both the CLIP text encoder and cross-attention dynamics during the denoising process, we find that unlearning instructions do not induce sustained reductions in attention to the targeted concept tokens, causing the targeted concept representations to persist throughout generation. These results reveal a fundamental limitation of prompt-level instruction in diffusion models and suggest that effective unlearning requires interventions beyond inference-time language control.

Tool using agents often fail for operational reasons even when language understanding is strong. Common causes include invalid arguments, interface drift, weak recovery, and inefficient retry behavior. We introduce ToolMisuseBench, an offline deterministic benchmark for evaluating tool misuse and recovery under explicit step, call, and retry budgets. The benchmark covers CRUD, retrieval, file, and scheduling environments with replayable fault injection. It reports success, invalid call behavior, policy violations, recovery quality, and budgeted efficiency. We release a public dataset with 6800 tasks and a reproducible evaluation pipeline. Baseline results show fault specific recovery gains for schema aware methods, while overall success remains limited under the released authorization and hard failure settings.

Long-tailed classification, where a small number of frequent classes dominate many rare ones, remains challenging because models systematically favor frequent classes at inference time. Existing post-hoc methods such as logit adjustment address this by adding a fixed classwise offset to the base-model logits. However, the correction required to restore the relative ranking of two classes need not be constant across inputs, and a fixed offset cannot adapt to such variation. We study this problem through Bayes-optimal reranking on a base-model top-k shortlist. The gap between the optimal score and the base score, the residual correction, decomposes into a classwise component that is constant within each class, and a pairwise component that depends on the input and competing labels. When the residual is purely classwise, a fixed offset suffices to recover the Bayes-optimal ordering. We further show that when the same label pair induces incompatible ordering constraints across contexts, no fixed offset can achieve this recovery. This decomposition leads to testable predictions regarding when pairwise correction can improve performance and when cannot. We develop REPAIR (Reranking via Pairwise residual correction), a lightweight post-hoc reranker that combines a shrinkage-stabilized classwise term with a linear pairwise term driven by competition features on the shortlist. Experiments on five benchmarks spanning image classification, species recognition, scene recognition, and rare disease diagnosis confirm that the decomposition explains where pairwise correction helps and where classwise correction alone suffices.

Research on Large Language Models (LLMs) studies output variation across generation, reasoning, alignment, and representational analysis, often under the umbrella of "diversity." Yet the terminology remains fragmented, largely because the normative objectives underlying tasks are rarely made explicit. We introduce the Magic, Madness, Heaven, Sin framework, which models output variation along a homogeneity-heterogeneity axis, where valuation is determined by the task and its normative objective. We organize tasks into four normative contexts: epistemic (factuality), interactional (user utility), societal (representation), and safety (robustness). For each, we examine the failure modes and vocabulary such as hallucination, mode collapse, bias, and erasure through which variation is studied. We apply the framework to analyze all pairwise cross-contextual interactions, revealing that optimizing for one objective, such as improving safety, can inadvertently harm demographic representation or creative diversity. We argue for context-aware evaluation of output variation, reframing it as a property shaped by task objectives rather than a model's intrinsic trait.

Conformal risk control (CRC) provides distribution-free guarantees for controlling the expected loss at a user-specified level. Existing theory typically assumes that the loss decreases monotonically with a tuning parameter that governs the size of the prediction set. This assumption is often violated in practice, where losses may behave non-monotonically due to competing objectives such as coverage and efficiency. We study CRC under non-monotone loss functions when the tuning parameter is selected from a finite grid, a common scenario in thresholding or discretized decision rules. Revisiting a known counterexample, we show that the validity of CRC without monotonicity depends on the relationship between the calibration sample size and the grid resolution. In particular, risk control can still be achieved when the calibration sample is sufficiently large relative to the grid. We provide a finite-sample guarantee for bounded losses over a grid of size $m$, showing that the excess risk above the target level $α$ is of order $\sqrt{\log(m)/n}$, where $n$ is the calibration sample size. A matching lower bound shows that this rate is minimax optimal. We also derive refined guarantees under additional structural conditions, including Lipschitz continuity and monotonicity, and extend the analysis to settings with distribution shift via importance weighting. Numerical experiments on synthetic multilabel classification and real object detection data illustrate the practical impact of non-monotonicity. Methods that account for finite-sample deviations achieve more stable risk control than approaches based on monotonicity transformations, while maintaining competitive prediction-set sizes.

Evolution Strategies (ES) have emerged as a scalable gradient-free alternative to reinforcement learning based LLM fine-tuning, but it remains unclear whether comparable task performance implies comparable solutions in parameter space. We compare ES and Group Relative Policy Optimization (GRPO) across four tasks in both single-task and sequential continual-learning settings. ES matches or exceeds GRPO in single-task accuracy and remains competitive sequentially when its iteration budget is controlled. Despite this similarity in task performance, the two methods produce markedly different model updates: ES makes much larger changes and induces broader off-task KL drift, whereas GRPO makes smaller, more localized updates. Strikingly, the ES and GRPO solutions are linearly connected with no loss barrier, even though their update directions are nearly orthogonal. We develop an analytical theory of ES that explains all these phenomena within a unified framework, showing how ES can accumulate large off-task movement on weakly informative directions while still making enough progress on the task to match gradient-based RL in downstream accuracy. These results show that gradient-free and gradient-based fine-tuning can reach similarly accurate yet geometrically distinct solutions, with important consequences for forgetting and knowledge preservation. The source code is publicly available: https://github.com/Bhoy1/ESvsGRPO.

We introduce SWE-ZERO to SWE-HERO, a two-stage SFT recipe that achieves state-of-the-art results on SWE-bench by distilling open-weight frontier LLMs. Our pipeline replaces resource-heavy dependencies with an evolutionary refinement strategy: (1) SWE-ZERO utilizes large-scale, execution-free trajectories to master code semantics and repository-level reasoning, and (2) SWE-HERO applies targeted, execution-backed refinement to transition these semantic intuitions into rigorous engineering workflows. Our empirical results set a new benchmark for open-source models of comparable size. We release a dataset of 300k SWE-ZERO and 13k SWE-HERO trajectories distilled from Qwen3-Coder-480B, alongside a suite of agents based on the Qwen2.5-Coder series. Notably, SWE-HERO-32B achieves a 62.2% resolution rate on SWE-bench Verified. Furthermore, despite being trained exclusively on Python, our agents demonstrate robust zero-shot transferability on SWE-bench Multilingual, reaching 44.1% and confirming the paradigm's generalizability across diverse languages.

Clone-and-own development produces families of related software variants that evolve independently. As variants diverge, important fixes applied in one repository are often missing in others. PaReco has shown that thousands of such missed opportunity (MO) patches exist across real ecosystems, yet its textual output provides limited support for understanding where and how these fixes should be propagated. We present MOVis, a lightweight, interactive desktop tool that visualizes MO patches between a source and target variant. MOVis loads PaReco's MO classifications and presents patched and buggy hunks side-by-side, highlighting corresponding regions and exposing structural differences that hinder reuse. This design enables developers to quickly locate missed fixes, understand required adaptations, and more efficiently maintain consistency across software variants. The tool, replication package, and demonstration video are available at https://zenodo.org/records/18356553 and https://youtu.be/Ac-gjBxHJ3Y.

High-performance GPU kernels are critical to modern machine learning systems, yet developing efficient implementations remains a challenging, expert-driven process due to the tight coupling between algorithmic structure, memory hierarchy usage, and hardware-specific optimizations. Recent work has explored using large language models (LLMs) to generate GPU kernels automatically, but generated implementations often struggle to maintain correctness and achieve competitive performance across iterative refinements. We present CuTeGen, an agentic framework for automated generation and optimization of GPU kernels that treats kernel development as a structured generate--test--refine workflow. Unlike approaches that rely on one-shot generation or large-scale search over candidate implementations, CuTeGen focuses on progressive refinement of a single evolving kernel through execution-based validation, structured debugging, and staged optimization. A key design choice is to generate kernels using the CuTe abstraction layer, which exposes performance-critical structures such as tiling and data movement while providing a more stable representation for iterative modification. To guide performance improvement, CuTeGen incorporates workload-aware optimization prompts and delayed integration of profiling feedback. Experimental results on matrix multiplication and activation workloads demonstrate that the framework produces functionally correct kernels and achieves competitive performance relative to optimized library implementations.

With the rise of personalized, persistent LLM agent frameworks such as OpenClaw, human-centered agentic social networks in which teams of collaborative AI agents serve individual users in a social network across multiple domains are becoming a reality. This setting creates novel privacy challenges: agents must coordinate across domain boundaries, mediate between humans, and interact with other users' agents, all while protecting sensitive personal information. While prior work has evaluated multi-agent coordination and privacy preservation, the dynamics and privacy risks of human-centered agentic social networks remain unexplored. To this end, we introduce AgentSocialBench, the first benchmark to systematically evaluate privacy risk in this setting, comprising scenarios across seven categories spanning dyadic and multi-party interactions, grounded in realistic user profiles with hierarchical sensitivity labels and directed social graphs. Our experiments reveal that privacy in agentic social networks is fundamentally harder than in single-agent settings: (1) cross-domain and cross-user coordination creates persistent leakage pressure even when agents are explicitly instructed to protect information, (2) privacy instructions that teach agents how to abstract sensitive information paradoxically cause them to discuss it more (we call it abstraction paradox). These findings underscore that current LLM agents lack robust mechanisms for privacy preservation in human-centered agentic social networks, and that new approaches beyond prompt engineering are needed to make agent-mediated social coordination safe for real-world deployment.

The rapid evolution of autonomous, agentic artificial intelligence within financial services has introduced an existential architectural crisis: large language models (LLMs) are probabilistic, non-deterministic systems operating in domains that demand absolute, mathematically verifiable compliance guarantees. Existing guardrail solutions -- including NVIDIA NeMo Guardrails and Guardrails AI -- rely on probabilistic classifiers and syntactic validators that are fundamentally inadequate for enforcing complex multi-variable regulatory constraints mandated by the SEC, FINRA, and OCC. This paper presents the Lean-Agent Protocol, a formal-verification-based AI guardrail platform that leverages the Aristotle neural-symbolic model developed by Harmonic AI to auto-formalize institutional policies into Lean 4 code. Every proposed agentic action is treated as a mathematical conjecture: execution is permitted if and only if the Lean 4 kernel proves that the action satisfies pre-compiled regulatory axioms. This architecture provides cryptographic-level compliance certainty at microsecond latency, directly satisfying SEC Rule 15c3-5, OCC Bulletin 2011-12, FINRA Rule 3110, and CFPB explainability mandates. A three-phase implementation roadmap from shadow verification through enterprise-scale deployment is provided.

The development of robust clinical decision support systems is frequently impeded by the scarcity of high-fidelity, privacy-preserving biomedical data. While Generative Large Language Models (LLMs) offer a promising avenue for synthetic data generation, they often struggle to capture the complex, non-linear dependencies and severe class imbalances inherent in Electronic Health Records (EHR), leading to statistically plausible but clinically invalid records. To bridge this gap, we introduce DISCO-TAB (DIScriminator-guided COntrol for TABular synthesis), a novel framework that orchestrates a fine-tuned LLM with a multi-objective discriminator system optimized via Reinforcement Learning. Unlike prior methods relying on scalar feedback, DISCO-TAB evaluates synthesis at four granularities, token, sentence, feature, and row, while integrating Automated Constraint Discovery and Inverse-Frequency Reward Shaping to autonomously preserve latent medical logic and resolve minority-class collapse. We rigorously validate our framework across diverse benchmarks, including high-dimensional, small-sample medical datasets (e.g., Heart Failure, Parkinson's). Our results demonstrate that hierarchical feedback yields state-of-the-art performance, achieving up to 38.2% improvement in downstream clinical classifier utility compared to GAN and Diffusion baselines, while ensuring exceptional statistical fidelity (JSD < 0.01) and robust resistance to membership inference attacks. This work establishes a new standard for generating trustworthy, utility-preserving synthetic tabular data for sensitive healthcare applications.

Metasurface inverse design has become central to realizing complex optical functionality, yet translating target responses into executable, solver-compatible workflows still demands specialized expertise in computational electromagnetics and solver-specific software engineering. Recent large language models (LLMs) offer a complementary route to reducing this workflow-construction burden, but existing language-driven systems remain largely session-bounded and do not preserve reusable workflow knowledge across inverse-design tasks. We present an agentic framework for metasurface inverse design that addresses this limitation through context-level skill evolution. The framework couples a coding agent, evolving skill artifacts, and a deterministic evaluator grounded in physical simulation so that solver-specific strategies can be iteratively refined across tasks without modifying model weights or the underlying physics solver. We evaluate the framework on a benchmark spanning multiple metasurface inverse-design task types, with separate training-aligned and held-out task families. Evolved skills raise in-distribution task success from 38% to 74%, increase criteria pass fraction from 0.510 to 0.870, and reduce average attempts from 4.10 to 2.30. On held-out task families, binary success changes only marginally, but improvements in best margin together with shifts in error composition and agent behavior indicate partial transfer of workflow knowledge. These results suggest that the main value of skill evolution lies in accumulating reusable solver-specific expertise around reliable computational engines, thereby offering a practical path toward more autonomous and accessible metasurface inverse-design workflows.

Reinforcement learning (RL) and model predictive control (MPC) offer complementary strengths, yet combining them at scale remains computationally challenging. We propose soft MPCritic, an RL-MPC framework that learns in (soft) value space while using sample-based planning for both online control and value target generation. soft MPCritic instantiates MPC through model predictive path integral control (MPPI) and trains a terminal Q-function with fitted value iteration, aligning the learned value function with the planner and implicitly extending the effective planning horizon. We introduce an amortized warm-start strategy that recycles planned open-loop action sequences from online observations when computing batched MPPI-based value targets. This makes soft MPCritic computationally practical, while preserving solution quality. soft MPCritic plans in a scenario-based fashion with an ensemble of dynamic models trained for next-step prediction accuracy. Together, these ingredients enable soft MPCritic to learn effectively through robust, short-horizon planning on classic and complex control tasks. These results establish soft MPCritic as a practical and scalable blueprint for synthesizing MPC policies in settings where policy extraction and direct, long-horizon planning may fail.

Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this phenomenon in coding tasks using an environment-manipulation setting, where models can rewrite evaluator code to trivially pass tests without solving the task, as a controlled testbed. Across both studied models, we identify a reproducible three-phase rebound pattern: models first attempt to rewrite the evaluator but fail, as their rewrites embed test cases their own solutions cannot pass. They then temporarily retreat to legitimate solving. When legitimate reward remains scarce, they rebound into successful hacking with qualitatively different strategies. Using representation engineering, we extract concept directions for shortcut, deception, and evaluation awareness from domain-general contrastive pairs and find that the shortcut direction tracks hacking behavior most closely, making it an effective representational proxy for detection. Motivated by this finding, we propose Advantage Modification, which integrates shortcut concept scores into GRPO advantage computation to penalize hacking rollouts before policy updates. Because the penalty is internalized into the training signal rather than applied only at inference time, Advantage Modification provides more robust suppression of hacking compared with generation-time activation steering.

Adapting closed-box service models (i.e., APIs) for target tasks typically relies on reprogramming via Zeroth-Order Optimization (ZOO). However, this standard strategy is known for extensive, costly API calls and often suffers from slow, unstable optimization. Furthermore, we observe that this paradigm faces new challenges with modern APIs (e.g., GPT-4o). These models can be less sensitive to the input perturbations ZOO relies on, thereby hindering performance gains. To address these limitations, we propose an Alternative efficient Reprogramming approach for Service models (AReS). Instead of direct, continuous closed-box optimization, AReS initiates a single-pass interaction with the service API to prime an amenable local pre-trained encoder. This priming stage trains only a lightweight layer on top of the local encoder, making it highly receptive to the subsequent glass-box (white-box) reprogramming stage performed directly on the local model. Consequently, all subsequent adaptation and inference rely solely on this local proxy, eliminating all further API costs. Experiments demonstrate AReS's effectiveness where prior ZOO-based methods struggle: on GPT-4o, AReS achieves a +27.8% gain over the zero-shot baseline, a task where ZOO-based methods provide little to no improvement. Broadly, across ten diverse datasets, AReS outperforms state-of-the-art methods (+2.5% for VLMs, +15.6% for standard VMs) while reducing API calls by over 99.99%. AReS thus provides a robust and practical solution for adapting modern closed-box models.

Large Language Models (LLMs) are powerful tools for answering user queries, yet they remain highly vulnerable to jailbreak attacks. Existing guardrail methods typically rely on internal features or textual responses to detect malicious queries, which either introduce substantial latency or suffer from the randomness in text generation. To overcome these limitations, we propose SelfGrader, a lightweight guardrail method that formulates jailbreak detection as a numerical grading problem using token-level logits. Specifically, SelfGrader evaluates the safety of a user query within a compact set of numerical tokens (NTs) (e.g., 0-9) and interprets their logit distribution as an internal safety signal. To align these signals with human intuition of maliciousness, SelfGrader introduces a dual-perspective scoring rule that considers both the maliciousness and benignness of the query, yielding a stable and interpretable score that reflects harmfulness and reduces the false positive rate simultaneously. Extensive experiments across diverse jailbreak benchmarks, multiple LLMs, and state-of-the-art guardrail baselines demonstrate that SelfGrader achieves up to a 22.66% reduction in ASR on LLaMA-3-8B, while maintaining significantly lower memory overhead (up to 173x) and latency (up to 26x).

The Muon optimizer has received considerable attention for its strong performance in training large language models, yet the design principle behind its matrix-gradient orthogonalization remains largely elusive. In this paper, we introduce a surrogate model that not only sheds new light on the design of Muon, but more importantly leads to a new optimizer. In the same spirit as the derivation of Newton's method, the surrogate approximates the loss as a quadratic function of the perturbation to a weight matrix $W$ using only three matrices: the gradient $G$, an output-space curvature matrix $H$, and the data matrix $Z$ that stacks the layer inputs. By minimizing this surrogate in one step and adopting a certain isotropic assumption on the weights, we obtain the closed-form update rule (up to momentum and weight decay) $W \leftarrow W - η\cdot \mathrm{msgn}(G(ZZ^\top)^{-1})$, where $η$ is the learning rate and $\mathrm{msgn}(X)=UV^\top$ if $X=USV^\top$ is a compact singular value decomposition. This new optimization method, which we refer to as Newton-Muon, shows that standard Muon can be interpreted as an implicit Newton-type method that neglects the right preconditioning induced by the input second moment. Empirically, on a reproduction of the earliest publicly released Modded-NanoGPT speedrun configuration using Muon for GPT-2 pretraining, Newton-Muon reaches the target validation loss in 6\% fewer iteration steps and reduces wall-clock training time by about 4\%.

Persian poetry is often remembered through recurrent symbols before it is remembered through plot. Wine vessels, gardens, flames, sacred titles, bodily beauty, and courtly names return across centuries, yet computational work still tends to flatten this material into isolated words or broad document semantics. That misses a practical unit of organization in Persian poetics: related forms travel as families and gain force through recurring relations. Using a corpus of 129,451 poems, we consolidate recurrent forms into traceable families, separate imagistic material from sacred and courtly reference, and map their relations in a multi-layer graph. The symbolic core is relatively sparse, the referential component much denser, and the attachment zone between them selective rather than diffuse. Across 11 Hijri-century bins, some families remain widely distributed, especially Shab (Night), Ruz (Day), and Khaak (Earth). Wine vessels, garden space, flame, and lyric sound strengthen later, while prestige-coded and heroic-courtly vocabulary is weighted earlier. Century-specific graphs show change in arrangement as well as membership. Modularity rises, cross-scope linkage declines, courtly bridges weaken, and sacred bridges strengthen. Hub positions shift too: Kherqe (Sufi Robe) gains late prominence, Farkhondeh {Blessed} and Banafsheh (Violet) recede, and Saaghar (Wine Cup) stays central across the chronology. In this corpus, Persian symbolism appears less as a fixed repertory than as a long-lived system whose internal weights and connections change over time.

Accurately modeling agent behaviors is an important task in self-driving. It is also a task with many symmetries, such as equivariance to the order of agents and objects in the scene or equivariance to arbitrary roto-translations of the entire scene as a whole; i.e., SE(2)-equivariance. The transformer architecture is a ubiquitous tool for modeling these symmetries. While standard self-attention is inherently permutation equivariant, explicit pairwise relative positional encodings have been the standard for introducing SE(2)-equivariance. However, this approach introduces an additional cost that is quadratic in the number of agents, limiting its scalability to larger scenes and batch sizes. In this work, we propose DriveGATr, a novel transformer-based architecture for agent modeling that achieves SE(2)-equivariance without the computational cost of existing methods. Inspired by recent advances in geometric deep learning, DriveGATr encodes scene elements as multivectors in the 2D projective geometric algebra $\mathbb{R}^*_{2,0,1}$ and processes them with a stack of equivariant transformer blocks. Crucially, DriveGATr models geometric relationships using standard attention between multivectors, eliminating the need for costly explicit pairwise relative positional encodings. Experiments on the Waymo Open Motion Dataset demonstrate that DriveGATr is comparable to the state-of-the-art in traffic simulation and establishes a superior Pareto front for performance vs computational cost.

Physically Assistive Robots (PARs) require personalized behaviors to ensure user safety and comfort. However, traditional preference learning methods, like exhaustive pairwise comparisons, cause severe physical and cognitive fatigue for users with profound motor impairments. To solve this, we propose a low-burden, offline framework that translates unstructured natural language feedback directly into deterministic robotic control policies. To safely bridge the gap between ambiguous human speech and robotic code, our pipeline uses Large Language Models (LLMs) grounded in the Occupational Therapy Practice Framework (OTPF). This clinical reasoning decodes subjective user reactions into explicit physical and psychological needs, which are then mapped into transparent decision trees. Before deployment, an automated "LLM-as-a-Judge" verifies the code's structural safety. We validated this system in a simulated meal preparation study with 10 adults with paralysis. Results show our natural language approach significantly reduces user workload compared to traditional baselines. Additionally, independent clinical experts confirmed the generated policies are safe and accurately reflect user preferences.

Reducing hallucinations in Large Language Models (LLMs) is essential for improving the accuracy of data extraction from large text corpora. Current methods, like prompt engineering and chain-of-thought prompting, focus on individual documents but fail to consider relationships across a corpus. This paper introduces Peer Context Outlier Detection (P-COD), a novel approach that uses the relationships between documents to improve extraction accuracy. Our application domain is in scientific literature summarization, where papers with similar experiment settings should draw similar conclusions. By comparing extracted data to validated peer information within the corpus, we adjust confidence scores and flag low-confidence results for expert review. High-confidence results, supported by peer validation, are considered reliable. Our experiments demonstrate up to 98% precision in outlier detection across 6 domains of science, demonstrating that our design reduces hallucinations, enhances trust in automated systems, and allows researchers to focus on ambiguous cases, streamlining the data extraction workflows.

Large language models are often not just wrong, but \emph{confidently wrong}: when they produce factually incorrect answers, they tend to verbalize overly high confidence rather than signal uncertainty. Such verbalized overconfidence can mislead users and weaken confidence scores as a reliable uncertainty signal, yet its internal mechanisms remain poorly understood. We present a circuit-level mechanistic analysis of this inflated verbalized confidence in LLMs, organized around three axes: capturing verbalized confidence as a differentiable internal signal, identifying the circuits that causally inflate it, and leveraging these insights for targeted inference-time recalibration. Across two instruction-tuned LLMs on three datasets, we find that a compact set of MLP blocks and attention heads, concentrated in middle-to-late layers, consistently writes the confidence-inflation signal at the final token position. We further show that targeted inference-time interventions on these circuits substantially improve calibration. Together, our results suggest that verbalized overconfidence in LLMs is driven by identifiable internal circuits and can be mitigated through targeted intervention.

Large language models (LLMs) are increasingly explored for NP-hard combinatorial optimization problems, but most existing methods emphasize feasible-instance solution generation and do not explicitly address infeasibility detection. We propose an infeasibility-aware framework that combines certifiable dataset construction, supervised fine-tuning, and LLM-assisted downstream search. For the minor-embedding problem, we introduce a new mathematical programming formulation together with provable zero-phase infeasibility screening, which enables scalable construction of training instances labeled either as feasible with structured certificates or as certifiably infeasible. Using training data generated through this exact optimization pipeline, we show that an 8B-parameter LLM can be fine-tuned to jointly perform solution generation and infeasibility detection. We further utilize LLM outputs as warm starts for downstream local search, providing a practical way to accelerate optimization even when the LLM outputs are imperfect. Experiments show that our fine-tuned model improves overall accuracy by up to 30\% over GPT-5.2; meanwhile LLM-guided warm starts provide up to $2\times$ speedup compared with starting from scratch in downstream local search.

Scientific discovery is slowed by fragmented literature that requires excessive human effort to gather, analyze, and understand. AI tools, including autonomous summarization and question answering, have been developed to aid in understanding scientific literature. However, these tools lack the structured, multi-step approach necessary for extracting deep insights from scientific literature. Large Language Models (LLMs) offer new possibilities for literature analysis, but remain unreliable due to hallucinations and incomplete extraction. We introduce Elhuyar, a multi-agent, human-in-the-loop system that integrates LLMs, structured AI, and human scientists to extract, analyze, and iteratively refine insights from scientific literature. The framework distributes tasks among specialized agents for filtering papers, extracting data, fitting models, and summarizing findings, with human oversight ensuring reliability. The system generates structured reports with extracted data, visualizations, model equations, and text summaries, enabling deeper inquiry through iterative refinement. Deployed in materials science, it analyzed literature on tungsten under helium-ion irradiation, showing experimentally correlated exponential helium bubble growth with irradiation dose and temperature, offering insight for plasma-facing materials (PFMs) in fusion reactors. This demonstrates how AI-assisted literature review can uncover scientific patterns and accelerate discovery.

Artificial intelligence (AI) systems are increasingly integrated into healthcare and pharmacy workflows, supporting tasks such as medication recommendations, dosage determination, and drug interaction detection. While these systems often demonstrate strong performance under standard evaluation metrics, their reliability in real-world decision-making remains insufficiently understood. In high-risk domains such as medication management, even a single incorrect recommendation can result in severe patient harm. This paper examines the reliability of AI-assisted medication systems by focusing on system failures and their potential clinical consequences. Rather than evaluating performance solely through aggregate metrics, this work shifts attention towards how errors occur and what happens when AI systems produce incorrect outputs. Through a series of controlled, simulated scenarios involving drug interactions and dosage decisions, we analyse different types of system failures, including missed interactions, incorrect risk flagging, and inappropriate dosage recommendations. The findings highlight that AI errors in medication-related contexts can lead to adverse drug reactions, ineffective treatment, or delayed care, particularly when systems are used without sufficient human oversight. Furthermore, the paper discusses the risks of over-reliance on AI recommendations and the challenges posed by limited transparency in decision-making processes. This work contributes a reliability-focused perspective on AI evaluation in healthcare, emphasising the importance of understanding failure behavior and real-world impact. It highlights the need to complement traditional performance metrics with risk-aware evaluation approaches, particularly in safety-critical domains such as pharmacy practice.

Recent 3D Gaussian splatting methods built atop SMPL achieve remarkable visual fidelity while continually increasing the complexity of the overall training architecture. We demonstrate that much of this complexity is unnecessary: by replacing SMPL with the Momentum Human Rig (MHR), estimated via SAM-3D-Body, a minimal pipeline with no learned deformations or pose-dependent corrections achieves the highest reported PSNR and competitive or superior LPIPS and SSIM on PeopleSnapshot and ZJU-MoCap. To disentangle pose estimation quality from body model representational capacity, we perform two controlled ablations: translating SAM-3D-Body meshes to SMPL-X, and translating the original dataset's SMPL poses into MHR both retrained under identical conditions. These ablations confirm that body model expressiveness has been a primary bottleneck in avatar reconstruction, with both mesh representational capacity and pose estimation quality contributing meaningfully to the full pipeline's gains.

When does consulting one information source raise the value of another, and when does it diminish it? We study this question for Bayesian decision-makers facing finite actions. The interaction decomposes into two opposing forces: a complement force, measuring how one source moves beliefs to where the other becomes more useful, and a substitute force, measuring how much the current decision is resolved. Their balance obeys a localization principle: substitution requires an observation to cross a decision boundary, though crossing alone does not guarantee it. Whenever posteriors remain inside the current decision region, the substitute force vanishes, and sources are guaranteed to complement each other, even when one source cannot, on its own, change the decision. The results hold for arbitrarily correlated sources and are formalized in Lean 4. Substitution is confined to the thin boundaries where decisions change. Everywhere else, information cooperates. Code and proofs: https://github.com/nidhishs/all-substitution-is-local.

Modern generator-based fuzzing techniques combine lightweight input generators with coverage-guided mutation as a method of exploring deep execution paths in a target program. A complimentary approach in prior research focuses on creating highly customized, domain-specific generators that encode structural and semantic logic sufficient enough to reach deep program states; the challenge comes from the overhead of writing and testing these complex generators. We investigate whether AI coding agents can automatically synthesize such target-specific generators, and whether the resulting generators are strong enough to obviate the need for coverage guidance and mutation entirely. Our approach, Gentoo, is comprised of an LLM coding agent (provided terminal access and source code of the fuzz target and its library) instructed to iteratively synthesize and refine an input generator, and optionally provided fine-grained predicate-level coverage feedback. We evaluate three configurations of Gentoo against human-written generators on fuzz targets for 7 real-world Java libraries. Our findings show that agent-synthesized generators achieve statistically significantly higher branch coverage than human-written baseline generators on 4 of 7 benchmarks. Critically, the use of coverage guidance and mutation strategies is not statistically significantly beneficial for agent-synthesized generators, but is significant for all human-written generators, suggesting that structural and semantic logic encoded in the agent generators makes coverage guidance largely unnecessary.

Modern real-time systems require accurate characterization of task timing behavior to ensure predictable performance, particularly on complex hardware architectures. Existing methods, such as worst-case execution time analysis, often fail to capture the fine-grained timing behaviors of a task under varying resource contexts (e.g., an allocation of cache, memory bandwidth, and CPU frequency), which is necessary to achieve efficient resource utilization. In this paper, we introduce a novel generative profiling approach that synthesizes context-dependent, fine-grained timing profiles for real-time tasks, including those for unmeasured resource allocations. Our approach leverages a nonparametric, conditional multi-marginal Schrödinger Bridge (MSB) formulation to generate accurate execution profiles for unseen resource contexts, with maximum likelihood guarantees. We demonstrate the efficiency and effectiveness of our approach through real-world benchmarks, and showcase its practical utility in a representative case study of adaptive multicore resource allocation for real-time systems.

The shift toward IoT-enabled, sensor-driven systems has transformed how operational data is generated, favoring continuous, real-time event streams (ES) over static event logs. This evolution presents new challenges for Streaming Process Mining (SPM), which must cope with out-of-order events, concurrent activities, incomplete cases, and concept drifts. Yet, the evaluation of SPM algorithms remains rooted in outdated practices, relying on static logs or artificially streamified data that fail to reflect the complexities of real-world streams. To address this gap, we first perform a comprehensive review of data stream literature to identify stream characteristics currently not reflected in the SPM community. Next, we use this information to extend the conceptual foundation for ES. Finally, we propose Stream of Intent, a prototype generator to produce ES with specific features. Our evaluation shows excellence in producing reproducible, intentional ES for targeted benchmarking and adaptive algorithm development in SPM.

Personal AI agents like OpenClaw run with elevated privileges on users' local machines, where a single successful prompt injection can leak credentials, redirect financial transactions, or destroy files. This threat goes well beyond conventional text-level jailbreaks, yet existing safety evaluations fall short: most test models in isolated chat settings, rely on synthetic environments, and do not account for how the agent framework itself shapes safety outcomes. We introduce CLAWSAFETY, a benchmark of 120 adversarial test scenarios organized along three dimensions (harm domain, attack vector, and harmful action type) and grounded in realistic, high-privilege professional workspaces spanning software engineering, finance, healthcare, law, and DevOps. Each test case embeds adversarial content in one of three channels the agent encounters during normal work: workspace skill files, emails from trusted senders, and web pages. We evaluate five frontier LLMs as agent backbones, running 2,520 sandboxed trials across all configurations. Attack success rates (ASR) range from 40\% to 75\% across models and vary sharply by injection vector, with skill instructions (highest trust) consistently more dangerous than email or web content. Action-trace analysis reveals that the strongest model maintains hard boundaries against credential forwarding and destructive actions, while weaker models permit both. Cross-scaffold experiments on three agent frameworks further demonstrate that safety is not determined by the backbone model alone but depends on the full deployment stack, calling for safety evaluation that treats model and framework as joint variables.

With the advancement of Agentic AI, researchers are increasingly leveraging autonomous agents to address challenges in software engineering (SE). However, the large language models (LLMs) that underpin these agents often function as black boxes, making it difficult to justify the superiority of Agentic AI approaches over baselines. Furthermore, missing information in the evaluation design description frequently renders the reproduction of results infeasible. To synthesize current evaluation practices for Agentic AI in SE, this study analyzes 18 papers on the topic, published or accepted by ICSE 2026, ICSE 2025, FSE 2025, ASE 2025, and ISSTA 2025. The analysis identifies prevailing approaches and their limitations in evaluating Agentic AI for SE, both in current research and potential future studies. To address these shortcomings, this position paper proposes a set of guidelines and recommendations designed to empower reproducible, explainable, and effective evaluations of Agentic AI in software engineering. In particular, we recommend that Agentic AI researchers make their Thought-Action-Result (TAR) trajectories and LLM interaction data, or summarized versions of these artifacts, publicly accessible. Doing so will enable subsequent studies to more effectively analyze the strengths and weaknesses of different Agentic AI approaches. To demonstrate the feasibility of such comparisons, we present a proof-of-concept case study that illustrates how TAR trajectories can support systematic analysis across approaches.

Partially observable Markov decision processes (POMDPs) offer a principled formalism for planning under state and transition uncertainty. Despite advances made towards solving large POMDPs, obtaining performant policies under limited planning time remains a major challenge due to the curse of dimensionality and the curse of history. For many POMDP problems, the value of information (VOI) - the expected performance gain from reasoning about observations - varies over the belief space. We introduce a dynamic programming framework that exploits this structure by conditionally processing observations based on the value of information at each belief. Building on this framework, we propose Value of Information Monte Carlo planning (VOIMCP), a Monte Carlo Tree Search algorithm that allocates computational effort more efficiently by selectively disregarding observation information when the VOI is low, avoiding unnecessary branching of observations. We provide theoretical guarantees on the near-optimality of our VOI reasoning framework and derive non-asymptotic convergence bounds for VOIMCP. Simulation evaluations demonstrate that VOIMCP outperforms baselines on several POMDP benchmarks.

Current limitations in wireless modeling and radio frequency (RF)-based AI are primarily driven by a lack of high-quality, measurement-based datasets that connect RF signals to their physical environments. RF heatmaps, the typical form of such data, are high-dimensional and complex but lack the geometric and semantic context needed for interpretation, constraining the development of supervised machine learning models. To address this bottleneck, we propose a new class of multimodal datasets that combines RF measurements with auxiliary modalities like high-resolution cameras and lidar to bridge the gap between RF signals and their physical causes. The proposed data collection will span diverse indoor and outdoor environments, featuring both static and dynamic scenarios, including human activities ranging from walking to subtle gestures. By achieving precise spatial and temporal co-registration and creating digital replicas for voxel-level annotation, this dataset will enable transformative AI research. Key tasks include the forward problem of predicting RF heatmaps from visual data to revolutionize wireless system design, and the inverse problem of inferring scene semantics from RF signals, creating a new form of RF-based perception.

Citation granularity - whether to cite individual sentences, paragraphs, or documents - is a critical design choice in attributed generation. While fine-grained citations are often preferred for precise human verification, their impact on model performance remains under-explored. We analyze four model scales (8B-120B) and demonstrate that enforcing fine-grained citations degrades attribution quality by 16-276% compared to the best-performing granularity. We observe a consistent performance pattern where attribution quality peaks at intermediate granularities (paragraph-level). Our analysis suggests that fine-grained (sentence-level) citations disrupt necessary semantic dependencies for attributing evidence to answer claims, while excessively coarse citations (multi-paragraph) introduce distracting noise. Importantly, the magnitude of this performance gap varies non-monotonically with model scale: fine-grained constraints disproportionately penalize larger models, suggesting that atomic citation units disrupt the multi-sentence information synthesis at which these models excel. Strikingly, citation-optimal granularity leads to substantial gains in attribution quality while preserving or even improving answer correctness. Overall, our findings demonstrate that optimizing solely for human verification via fine-grained citation disregards model constraints, compromising both attribution faithfulness and generation reliability. Instead, effective attribution requires aligning citation granularity with the model's natural semantic scope.

Language Models (LMs) exhibit two distinct mechanisms for knowledge acquisition: in-weights learning (i.e., encoding information within the model weights) and in-context learning (ICL). Although these two modes offer complementary strengths, in-weights learning frequently struggles to facilitate deductive reasoning over the internalized knowledge. We characterize this limitation as a deficit in latent generalization, of which the reversal curse is one example. Conversely, in-context learning demonstrates highly robust latent generalization capabilities. To improve latent generalization from in-weights knowledge, prior approaches rely on train-time data augmentation, yet these techniques are task-specific, scale poorly, and fail to generalize to out-of-distribution knowledge. To overcome these shortcomings, this work studies how models can be taught to use test-time compute, or 'thinking', specifically to improve latent generalization. We use Reinforcement Learning (RL) from correctness feedback to train models to produce long chains-of-thought (CoTs) to improve latent generalization. Our experiments show that this thinking approach not only resolves many instances of latent generalization failures on in-distribution knowledge but also, unlike augmentation baselines, generalizes to new knowledge for which no RL was performed. Nevertheless, on pure reversal tasks, we find that thinking does not unlock direct knowledge inversion, but the generate-and-verify ability of thinking models enables them to get well above chance performance. The brittleness of factual self-verification means thinking models still remain well below the performance of in-context learning for this task. Overall, our results establish test-time thinking as a flexible and promising direction for improving the latent generalization of LMs.

Synonymy is a widespread yet puzzling linguistic phenomenon. Absolute synonyms theoretically should not exist, as they do not expand language's expressive potential. However, it was suggested that even if synonyms denote the same concept, they may reflect different perspectives or carry distinct cultural associations, claims that have rarely been tested quantitatively. In Hindi, prolonged contact with Persian produced many Perso-Arabic loanwords coexisting with their Sanskrit counterpart, forming numerous synonym pairs. This study investigates whether centuries after these borrowings appeared in the Subcontinent their origin can still be distinguished using distributional data alone and regardless of their semantic content. A Random Forest trained on word embeddings of Hindi synonyms successfully classified words by Sanskrit or Perso-Arabic origin, even when they were semantically unrelated, suggesting that usage patterns preserve traces of etymology. These findings provide quantitative evidence that context encodes etymological signals and that synonymy may reflect subtle but systematic distinctions linked to origin. They support the idea that synonymous words can offer different perspectives and that etymologically related words may form distinct conceptual subspaces, creating a new type of semantic frame shaped by historical origin. Overall, the results highlight the power of context in capturing nuanced distinctions beyond traditional semantic similarity.

Thousands of diverse benchmarks have been developed to measure the quality of large language models (LLMs). Yet prior work has demonstrated that LLM performance is often sufficiently explained by a small set of latent factors, or abilities. This suggests the potential for more efficient and principled benchmarking, but it remains difficult to compare the quality of different methods. Motivated by predictive validity, we argue that the quality of a benchmarking framework should be grounded in how efficiently it enables the prediction of model performance on unseen tasks. To analyze this objective, we collect the "Wide-scale Item Level Dataset" (WILD), a dataset of item-model response pairs, comprising evaluations of 65 models on 109,564 unique items spanning 163 tasks drawn from 27 datasets. This dataset enables the first analysis of how different techniques can predict a model's performance on a large, diverse collection of unseen tasks under different budget constraints. We demonstrate that combining a modified multidimensional item response theory (IRT) model with adaptive item selection driven by optimal experimental design can predict performance on 112 held-out benchmark tasks with a mean absolute error (MAE) of less than 7%, and can do so after observing only 16 items. We further demonstrate that incorporating cost-aware discount factors into our selection criteria can reduce the total tokens needed to reach 7% MAE from 141,000 tokens to only 22,000, an 85% reduction in evaluation cost.

We present ReFormeR, a pattern-guided approach for query reformulation. Instead of prompting a language model to generate reformulations of a query directly, ReFormeR first elicits short reformulation patterns from pairs of initial queries and empirically stronger reformulations, consolidates them into a compact library of transferable reformulation patterns, and then selects an appropriate reformulation pattern for a new query given its retrieval context. The selected pattern constrains query reformulation to controlled operations such as sense disambiguation, vocabulary grounding, or discriminative facet addition, to name a few. As such, our proposed approach makes the reformulation policy explicit through these reformulation patterns, guiding the LLM towards targeted and effective query reformulations. Our extensive experiments on TREC DL 2019, DL 2020, and DL Hard show consistent improvements over classical feedback methods and recent LLM-based query reformulation and expansion approaches.

Large Language Models (LLMs) increasingly rely on multi-turn reasoning and interaction, such as adaptive retrieval-augmented generation (RAG) and ReAct-style agents, to answer difficult questions. These methods improve accuracy by iteratively retrieving information, reasoning, or acting, but introduce a key challenge: \textbf{When should the model stop?} Existing approaches rely on heuristic stopping rules or fixed turn budgets and provide no formal guarantees that the final prediction still contains the correct answer. This limitation is particularly problematic in high-stakes domains such as finance and healthcare, where unnecessary turns increase cost and latency, while stopping too early risks incorrect decisions. Conformal prediction (CP) provides formal coverage guarantees, but existing LLM-CP methods only apply to a single model output and cannot handle multi-turn pipelines with adaptive stopping. To address this gap, we propose Multi-Turn Language Models with Conformal Prediction (MiCP), the first CP framework for multi-turn reasoning. MiCP allocates different error budgets across turns, enabling the model to stop early while maintaining an overall coverage guarantee. We demonstrate MiCP on adaptive RAG and ReAct, where it achieves the target coverage on both single-hop and multi-hop question answering benchmarks while reducing the number of turns, inference cost, and prediction set size. We further introduce a new metric that jointly evaluates coverage validity and answering efficiency.

Modern LLMs scale at test-time, e.g. via repeated sampling, where inference cost grows with model size and the number of samples. This creates a trade-off that pretraining scaling laws, such as Chinchilla, do not address. We present Train-to-Test ($T^2$) scaling laws that jointly optimize model size, training tokens, and number of inference samples under fixed end-to-end budgets. $T^2$ modernizes pretraining scaling laws with pass@$k$ modeling used for test-time scaling, then jointly optimizes pretraining and test-time decisions. Forecasts from $T^2$ are robust over distinct modeling approaches: measuring joint scaling effect on the task loss and modeling impact on task accuracy. Across eight downstream tasks, we find that when accounting for inference cost, optimal pretraining decisions shift radically into the overtraining regime, well-outside of the range of standard pretraining scaling suites. We validate our results by pretraining heavily overtrained models in the optimal region that $T^2$ scaling forecasts, confirming their substantially stronger performance compared to pretraining scaling alone. Finally, as frontier LLMs are post-trained, we show that our findings survive the post-training stage, making $T^2$ scaling meaningful in modern deployments.

Text production (and translations) proceeds in the form of stretches of typing, interrupted by keystroke pauses. It is often assumed that fast typing reflects unchallenged/automated translation production while long(er) typing pauses are indicative of translation problems, hurdles or difficulties. Building on a long discussion concerning the determination of pause thresholds that separate automated from presumably reflective translation processes (O'Brien, 2006; Alves and Vale, 2009; Timarova et al., 2011; Dragsted and Carl, 2013; Lacruz et al., 2014; Kumpulainen, 2015; Heilmann and Neumann 2016), this paper compares three recent approaches for computing these pause thresholds, and suggest and evaluate a novel method for computing Production Unit Breaks.

We study the problem of identifying an optimal coupling between input-output distributional data generated by a causal dynamical system. The coupling is required to satisfy prescribed marginal distributions and a causality constraint reflecting the temporal structure of the system. We formulate this problem as a Schr"odinger Bridge, which seeks the coupling closest - in Kullback-Leibler divergence - to a given prior while enforcing both marginal and causality constraints. For the case of Gaussian marginals and general time-dependent quadratic cost functions, we derive a fully tractable characterization of the Sinkhorn iterations that converges to the optimal solution. Beyond its theoretical contribution, the proposed framework provides a principled foundation for applying causal optimal transport methods to system identification from distributional data.

Language models can answer many entity-centric factual questions, but it remains unclear which internal mechanisms are involved in this process. We study this question across multiple language models. We localize entity-selective MLP neurons using templated prompts about each entity, and then validate them with causal interventions on PopQA-based QA examples. On a curated set of 200 entities drawn from PopQA, localized neurons concentrate in early layers. Negative ablation produces entity-specific amnesia, while controlled injection at a placeholder token improves answer retrieval relative to mean-entity and wrong-cell controls. For many entities, activating a single localized neuron is sufficient to recover entity-consistent predictions once the context is initialized, consistent with compact entity retrieval rather than purely gradual enrichment across depth. Robustness to aliases, acronyms, misspellings, and multilingual forms supports a canonicalization interpretation. The effect is strong but not universal: not every entity admits a reliable single-neuron handle, and coverage is higher for popular entities. Overall, these results identify sparse, causally actionable access points for analyzing and modulating entity-conditioned factual behavior.

This report is part of the Qumphy project (22HLT01 Qumphy) that is funded by the European Union and is dedicated to the development of measures to quantify the uncertainties associated with Machine Learning algorithms applied to medical problems, in particular the analysis and processing of Photoplethysmography (PPG) signals. In this report, a list of six medical problems that are related to PPG signals and serve as Benchmark Problems is given. Suitable Benchmark datasets and their usage are described also.

Retrieval-augmented generation (RAG) systems are gaining traction in enterprise settings, yet stringent data protection regulations prevent many organizations from using cloud-based services, necessitating on-premises deployments. While existing blueprints and reference architectures focus on cloud deployments and lack enterprise-grade components, comprehensive on-premises implementation frameworks remain scarce. This paper aims to address this gap by presenting a comprehensive AI engineering blueprint for scalable on-premises enterprise RAG solutions. It is designed to address common challenges and streamline the integration of RAG into existing enterprise infrastructure. The blueprint provides: (1) an end-to-end reference architecture described using the 4+1 view model, (2) a reference application for on-premises deployment, and (3) best practices for tooling, development, and CI/CD pipelines, all publicly available on GitHub. Ongoing case studies and expert interviews with industry partners will assess its practical benefits.

Identifying the features to be released in the next version of software, from a pool of potential candidates, is a challenging problem. User feedback from app stores is frequently used by software vendors for the evolution of apps across releases. Privacy feedback, although smaller in volume, carries a larger impact influencing app's success. Multiple existing work has focused on summarizing privacy concerns at the app level and has also shown that developers utilize feedback to implement security and privacy-related changes in subsequent releases. However, the current literature offers little support for release managers and developers in identifying privacy concerns prior to release. This gap exists as user reviews are typically available in app stores only after new features of a software system is released. In this paper, we introduce Pre-PI, a novel approach that summarizes privacy concerns for to-be-released features. Our method first maps existing features to semantically similar privacy reviews to learn feature-privacy review relations. We then simulate feedback for candidate features and generate concise summaries of privacy concerns. We evaluate Pre-PI across three real-world apps, and compare it with Hark, a state-of-the-art method that relies on post-release user feedback to identify privacy concerns. Results show that Pre-PI generates additional valid privacy concerns and identifies these concerns earlier than Hark, allowing proactive mitigation prior to release.

American football practice generates video at scale, yet the interaction of interest occupies only a brief window of each long, untrimmed clip. Reliable biomechanical analysis, therefore, depends on spatiotemporal localization that identifies both the interacting entities and the onset of contact. We study First Point of Contact (FPOC), defined as the first frame in which a player physically touches a tackle dummy, in unconstrained practice footage with camera motion, clutter, multiple similarly equipped athletes, and rapid pose changes around impact. We present GRAZE, a training-free pipeline for FPOC localization that requires no labeled tackle-contact examples. GRAZE uses Grounding DINO to discover candidate player-dummy interactions, refines them with motion-aware temporal reasoning, and uses SAM2 as an explicit pixel-level verifier of contact rather than relying on detection confidence alone. This separation between candidate discovery and contact confirmation makes the approach robust to cluttered scenes and unstable grounding near impact. On 738 tackle-practice videos, GRAZE produces valid outputs for 97.4% of clips and localizes FPOC within $\pm$ 10 frames on 77.5% of all clips and within $\pm$ 20 frames on 82.7% of all clips. These results show that frame-accurate contact onset localization in real-world practice footage is feasible without task-specific training.

Can large language models (LLMs) predict which researchers will collaborate? We study this question through link prediction on real-world co-authorship networks from OpenAlex (9.96M authors, 108.7M edges), evaluating whether LLMs can predict future scientific collaborations using only author profiles, without access to graph structure. Using Qwen2.5-72B-Instruct across three historical eras of AI research, we find that LLMs and topology heuristics capture distinct signals and are strongest in complementary settings. On new-edge prediction under natural class imbalance, the LLM achieves AUROC 0.714--0.789, outperforming Common Neighbors, Jaccard, and Preferential Attachment, with recall up to 92.9\%; under balanced evaluation, the LLM outperforms \emph{all} topology heuristics in every era (AUROC 0.601--0.658 vs.\ best-heuristic 0.525--0.538); on continued edges, the LLM (0.687) is competitive with Adamic-Adar (0.684). Critically, 78.6--82.7\% of new collaborations occur between authors with no common neighbor -- a blind spot where all topology heuristics score zero but the LLM still achieves AUROC 0.652 by reasoning from author metadata alone. A temporal metadata ablation reveals that research concepts are the dominant signal (removing concepts drops AUROC by 0.047--0.084). Providing pre-computed graph features to the LLM \emph{degrades} performance due to anchoring effects, confirming that LLMs and topology methods should operate as separate, complementary channels. A socio-cultural ablation finds that name-inferred ethnicity and institutional country do not predict collaboration beyond topology, reflecting the demographic homogeneity of AI research. A node2vec baseline achieves AUROC comparable to Adamic-Adar, establishing that LLMs access a fundamentally different information channel -- author metadata -- rather than encoding the same structural signal differently.

Offline reinforcement learning (RL) has received increasing attention for learning policies from previously collected data without interaction with the real environment, which is particularly important in high-stakes applications. While a growing body of work has developed offline RL algorithms, these methods often rely on restrictive assumptions about data coverage and suffer from distribution shift. In this paper, we propose a residuals-based offline RL framework for general state and action spaces. Specifically, we define a residuals-based Bellman optimality operator that explicitly incorporates estimation error in learning transition dynamics into policy optimization by leveraging empirical residuals. We show that this Bellman operator is a contraction mapping and identify conditions under which its fixed point is asymptotically optimal and possesses finite-sample guarantees. We further develop a residuals-based offline deep Q-learning (DQN) algorithm. Using a stochastic CartPole environment, we demonstrate the effectiveness of our residuals-based offline DQN algorithm.

Rubric-based evaluation is widely used in LLM benchmarks and training pipelines for open-ended, less verifiable tasks. While prior work has demonstrated the effectiveness of rubrics using downstream signals such as reinforcement learning outcomes, there remains no principled way to diagnose rubric quality issues from such aggregated or downstream signals alone. To address this gap, we introduce RIFT: RubrIc Failure mode Taxonomy, a taxonomy for systematically characterizing failure modes in rubric composition and design. RIFT consists of eight failure modes organized into three high-level categories: Reliability Failures, Content Validity Failures, and Consequential Validity Failures. RIFT is developed using grounded theory by iteratively annotating rubrics drawn from five diverse benchmarks spanning general instruction following, code generation, creative writing, and expert-level deep research, until no new failure modes are identified. We evaluate the consistency of the taxonomy by measuring agreement among independent human annotators, observing fair agreement overall (87% pairwise agreement and 0.64 average Cohen's kappa). Finally, to support scalable diagnosis, we propose automated rubric quality metrics and show that they align with human failure-mode annotations, achieving up to 0.86 F1.

Surgical action automation has progressed rapidly toward achieving surgeon-like dexterous control, driven primarily by advances in learning from demonstration and vision-language-action models. While these have demonstrated success in table-top experiments, translating them to clinical deployment remains challenging: current methods offer limited predictability on where instruments will interact on tissue surfaces and lack explicit conditioning inputs to enforce tool-action-specific safe interaction regions. Addressing this gap, we introduce AffordTissue, a multimodal framework for predicting tool-action specific tissue affordance regions as dense heatmaps during cholecystectomy. Our approach combines a temporal vision encoder capturing tool motion and tissue dynamics across multiple viewpoints, language conditioning enabling generalization across diverse instrument-action pairs, and a DiT-style decoder for dense affordance prediction. We establish the first tissue affordance benchmark by curating and annotating 15,638 video clips across 103 cholecystectomy procedures, covering six unique tool-action pairs involving four instruments (hook, grasper, scissors, clipper) and their associated tasks: dissection, grasping, clipping, and cutting. Experiments demonstrate substantial improvement over vision-language model baselines (20.6 px ASSD vs. 60.2 px for Molmo-VLM), showing that our task-specific architecture outperforms large-scale foundation models for dense surgical affordance prediction. By predicting tool-action specific tissue affordance regions, AffordTissue provides explicit spatial reasoning for safe surgical automation, potentially unlocking explicit policy guidance toward appropriate tissue regions and early safe stop when instruments deviate outside predicted safe zones.

Large Language Models (LLMs) are increasingly deployed in high-stakes decision-making contexts. While prior work has shown that LLMs exhibit cognitive biases behaviorally, whether these biases correspond to identifiable internal representations and can be mitigated through targeted intervention remains an open question. We define LLM cognitive bias as systematic, reproducible deviations from correct answers in tasks with computable ground-truth baselines, and introduce LLM CogBias, a benchmark organized around four families of cognitive biases: Judgment, Information Processing, Social, and Response. We evaluate three LLMs and find that cognitive biases emerge systematically across all four families, with magnitudes and debiasing responses that are strongly family-dependent: prompt-level debiasing substantially reduces Response biases but backfires for Judgment biases. Using linear probes under a contrastive design, we show that these biases are encoded as linearly separable directions in model activation space. Finally, we apply activation steering to modulate biased behavior, achieving 26--32\% reduction in bias score (fraction of biased responses) while preserving downstream capability on 25 benchmarks (Llama: negligible degradation; Qwen: up to $-$19.0pp for Judgment biases). Despite near-orthogonal bias representations across models (mean cosine similarity 0.01), steering reduces bias at similar rates across architectures ($r(246)$=.621, $p$<.001), suggesting shared functional organization.

Predicting the perceived intensity of odorants remains a fundamental challenge in sensory science due to the complex, non-linear behavior of their response, as well as the difficulty in correlating molecular structure with human perception. While traditional deep learning models, such as Graph Convolutional Networks (GCNs), excel at capturing molecular topology, they often fail to account for the biological and perceptual context of olfaction. This study introduces VIANA, a novel "tri-pillar" framework that integrates structural graph theory, character value embeddings, and phenomenological behavior. This methodology systematically evaluates knowledge transfer across three distinct domains: molecular structure via GCNs, semantic odor character values via Principal Odor Map (POM) embeddings, and biological dose-response logic via Hill's law. We demonstrate that knowledge transfer is not inherently positive; rather, a balance must be maintained in the volume of information provided to the model. While raw semantic data led to "information overload" in domain-informed models, applying Principal Component Analysis (PCA) to distill the 95% most impactful semantic variance yielded a superior "signal distillation" effect. Results indicate that the synthesis of these three knowledge transfer pillars significantly outperforms baseline structural models, with VIANA achieving a peak R^2 of 0.996 and a test Mean Squared Error (MSE) of 0.19. In this context, VIANA successfully captures the physical ceiling of saturation, the sensitivity of detection thresholds, and the nuance of odor character value expression, providing a domain grounded simulation of the human olfactory experience. This research provides a robust framework for digital olfaction, effectively bridging the gap between molecular informatics and sensory perception.

Society 5.0 and Industry 5.0 call for human-centric technology integration, yet the concept lacks an operational definition that can be measured, optimized, or evaluated at the firm level. This paper addresses three gaps. First, existing models of human-AI complementarity treat the augmentation function phi(D) as exogenous -- dependent only on the stock of AI deployed -- ignoring that two firms with identical technology investments achieve radically different augmentation outcomes depending on how the workplace is organized around the human-AI interaction. Second, no multi-dimensional instrument exists linking workplace design choices to augmentation productivity. Third, the Society 5.0 literature proposes human-centricity as a normative aspiration but provides no formal criterion for when it is economically optimal. We make four contributions. (1) We endogenize the augmentation function as phi(D, W), where W is a five-dimensional workplace design vector -- AI interface design, decision authority allocation, task orchestration, learning loop architecture, and psychosocial work environment -- and prove that human-centric design is profit-maximizing when the workforce's augmentable cognitive capital exceeds a critical threshold. (2) We conduct a PRISMA-guided systematic review of 120 papers (screened from 6,096 records) to map the evidence base for each dimension. (3) We provide secondary empirical evidence from Colombia's EDIT manufacturing survey (N=6,799 firms) showing that management practice quality amplifies the return to technology investment (interaction coefficient 0.304, p<0.01). (4) We propose the Workplace Augmentation Design Index (WADI), a 36-item theory-grounded instrument for diagnosing human-centricity at the firm level. Decision authority allocation emerges as the binding constraint for Society 5.0 transitions, and task orchestration as the most under-researched dimension

We propose that AI automation is a continuum between: (i) crashing waves where AI capabilities surge abruptly over small sets of tasks, and (ii) rising tides where the increase in AI capabilities is more continuous and broad-based. We test for these effects in preliminary evidence from an ongoing evaluation of AI capabilities across over 3,000 broad-based tasks derived from the U.S. Department of Labor O*NET categorization that are text-based and thus LLM-addressable. Based on more than 17,000 evaluations by workers from these jobs, we find little evidence of crashing waves (in contrast to recent work by METR), but substantial evidence that rising tides are the primary form of AI automation. AI performance is high and improving rapidly across a wide range of tasks. We estimate that, in 2024-Q2, AI models successfully complete tasks that take humans approximately 3-4 hours with about a 50% success rate, increasing to about 65% by 2025-Q3. If recent trends in AI capability growth persist, this pace of AI improvement implies that LLMs will be able to complete most text-related tasks with success rates of, on average, 80%-95% by 2029 at a minimally sufficient quality level. Achieving near-perfect success rates at this quality level or comparable success rates at superior quality would require several additional years. These AI capability improvements would impact the economy and labor market as organizations adopt AI, which could have a substantially longer timeline.

We introduce world-centered multi-agent systems (WMAS) as an alternative to traditional agent-centered architectures, arguing that structured domains such as enterprises and institutional systems require a shared, explicit world representation to ensure semantic consistency, explainability, and long-term stability. We classify worlds along dimensions including ontological explicitness, normativity, etc. In WMAS, learning and coordination operate over a shared world model rather than isolated agent-local representations, enabling global consistency and verifiable system behavior. We propose semantic models as a mathematical formalism for representing such worlds. Finally, we present the Ontobox platform as a realization of WMAS.

Moderation layers are increasingly a core component of many products built on user- or model-generated content. However, drafting and maintaining domain-specific safety policies remains costly. We present Deep Policy Research (DPR), a minimal agentic system that drafts a full content moderation policy based on only human-written seed domain information. DPR uses a single web search tool and lightweight scaffolding to iteratively propose search queries, distill diverse web sources into policy rules, and organize rules into an indexed document. We evaluate DPR on (1) the OpenAI undesired content benchmark across five domains with two compact reader LLMs and (2) an in-house multimodal advertisement moderation benchmark. DPR consistently outperforms definition-only and in-context learning baselines, and in our end-to-end setting it is competitive with expert-written policy sections in several domains. Moreover, under the same seed specification and evaluation protocol, DPR outperforms a general-purpose deep research system, suggesting that a task-specific, structured research loop can be more effective than generic web research for policy drafting. We release our experiment code at https://github.com/xiaowu0162/deep-policy-research.

LLM-based agents increasingly operate across repeated sessions, maintaining task states to ensure continuity. In many deployments, a single agent serves multiple users within a team or organization, reusing a shared knowledge layer across user identities. This shared persistence expands the failure surface: information that is locally valid for one user can silently degrade another user's outcome when the agent reapplies it without regard for scope. We refer to this failure mode as unintentional cross-user contamination (UCC). Unlike adversarial memory poisoning, UCC requires no attacker; it arises from benign interactions whose scope-bound artifacts persist and are later misapplied. We formalize UCC through a controlled evaluation protocol, introduce a taxonomy of three contamination types, and evaluate the problem in two shared-state mechanisms. Under raw shared state, benign interactions alone produce contamination rates of 57--71%. A write-time sanitization is effective when shared state is conversational, but leaves substantial residual risk when shared state includes executable artifacts, with contamination often manifesting as silent wrong answers. These results indicate that shared-state agents need artifact-level defenses beyond text-level sanitization to prevent silent cross-user failures.

Reservoir simulation workflows face a fundamental data asymmetry: input parameter fields (geostatistical permeability realizations, porosity distributions) are free to generate in arbitrary quantities, yet existing neural operator surrogates require large corpora of expensive labeled simulation trajectories and cannot exploit this unlabeled structure. We introduce \textbf{PI-JEPA} (Physics-Informed Joint Embedding Predictive Architecture), a surrogate pretraining framework that trains \emph{without any completed PDE solves}, using masked latent prediction on unlabeled parameter fields under per-sub-operator PDE residual regularization. The predictor bank is structurally aligned with the Lie--Trotter operator-splitting decomposition of the governing equations, dedicating a separate physics-constrained latent module to each sub-process (pressure, saturation transport, reaction), enabling fine-tuning with as few as 100 labeled simulation runs. On single-phase Darcy flow, PI-JEPA achieves $1.9\times$ lower error than FNO and $2.4\times$ lower error than DeepONet at $N_\ell{=}100$, with 24\% improvement over supervised-only training at $N_\ell{=}500$, demonstrating that label-free surrogate pretraining substantially reduces the simulation budget required for multiphysics surrogate deployment.

Test-time scaling has emerged as an effective way to improve language models on challenging reasoning tasks. However, most existing methods treat each problem in isolation and do not systematically reuse knowledge from prior reasoning trajectories. In particular, they underutilize procedural knowledge: how to reframe a problem, choose an approach, and verify or backtrack when needed. We introduce Reasoning Memory, a retrieval-augmented generation (RAG) framework for reasoning models that explicitly retrieves and reuses procedural knowledge at scale. Starting from existing corpora of step-by-step reasoning trajectories, we decompose each trajectory into self-contained subquestion-subroutine pairs, yielding a datastore of 32 million compact procedural knowledge entries. At inference time, a lightweight in-thought prompt lets the model verbalize the core subquestion, retrieve relevant subroutines within its reasoning trace, and reason under diverse retrieved subroutines as implicit procedural priors. Across six math, science, and coding benchmarks, Reasoning Memory consistently outperforms RAG with document, trajectory, and template knowledge, as well as a compute-matched test-time scaling baseline. With a higher inference budget, it improves over no retrieval by up to 19.2% and over the strongest compute-matched baseline by 7.9% across task types. Ablation studies show that these gains come from two key factors: the broad procedural coverage of the source trajectories and our decomposition and retrieval design, which together enable effective extraction and reuse of procedural knowledge.

World models -- learned internal simulators of environment dynamics -- are rapidly becoming foundational to autonomous decision-making in robotics, autonomous vehicles, and agentic AI. Yet this predictive power introduces a distinctive set of safety, security, and cognitive risks. Adversaries can corrupt training data, poison latent representations, and exploit compounding rollout errors to cause catastrophic failures in safety-critical deployments. World model-equipped agents are more capable of goal misgeneralisation, deceptive alignment, and reward hacking precisely because they can simulate the consequences of their own actions. Authoritative world model predictions further foster automation bias and miscalibrated human trust that operators lack the tools to audit. This paper surveys the world model landscape; introduces formal definitions of trajectory persistence and representational risk; presents a five-profile attacker capability taxonomy; and develops a unified threat model extending MITRE ATLAS and the OWASP LLM Top 10 to the world model stack. We provide an empirical proof-of-concept on trajectory-persistent adversarial attacks (GRU-RSSM: A_1 = 2.26x amplification, -59.5% reduction under adversarial fine-tuning; stochastic RSSM proxy: A_1 = 0.65x; DreamerV3 checkpoint: non-zero action drift confirmed). We illustrate risks through four deployment scenarios and propose interdisciplinary mitigations spanning adversarial hardening, alignment engineering, NIST AI RMF and EU AI Act governance, and human-factors design. We argue that world models must be treated as safety-critical infrastructure requiring the same rigour as flight-control software or medical devices.

Inverse reinforcement learning (IRL) recovers the loss function of a forward learner from its observed responses adaptive IRL aims to reconstruct the loss function of a forward learner by passively observing its gradients as it performs reinforcement learning (RL). This paper proposes a novel passive Langevin-based algorithm that achieves adaptive IRL. The key difficulty in adaptive IRL is that the required gradients in the passive algorithm are counterfactual, that is, they are conditioned on events of probability zero under the forward learner's trajectory. Therefore, naive Monte Carlo estimators are prohibitively inefficient, and kernel smoothing, though common, suffers from slow convergence. We overcome this by employing Malliavin calculus to efficiently estimate the required counterfactual gradients. We reformulate the counterfactual conditioning as a ratio of unconditioned expectations involving Malliavin quantities, thus recovering standard estimation rates. We derive the necessary Malliavin derivatives and their adjoint Skorohod integral formulations for a general Langevin structure, and provide a concrete algorithmic approach which exploits these for counterfactual gradient estimation.

Competency question (CQ) elicitation represents a critical but resource-intensive bottleneck in ontology engineering. This foundational phase is often hampered by the communication gap between domain experts, who possess the necessary knowledge, and ontology engineers, who formalise it. This paper introduces IDEA2, a novel, semi-automated workflow that integrates Large Language Models (LLMs) within a collaborative, expert-in-the-loop process to address this challenge. The methodology is characterised by a core iterative loop: an initial LLM-based extraction of CQs from requirement documents, a co-creational review and feedback phase by domain experts on an accessible collaborative platform, and an iterative, feedback-driven reformulation of rejected CQs by an LLM until consensus is achieved. To ensure transparency and reproducibility, the entire lifecycle of each CQ is tracked using a provenance model that captures the full lineage of edits, anonymised feedback, and generation parameters. The workflow was validated in 2 real-world scenarios (scientific data, cultural heritage), demonstrating that IDEA2 can accelerate the requirements engineering process, improve the acceptance and relevance of the resulting CQs, and exhibit high usability and effectiveness among domain experts. We release all code and experiments at https://github.com/KE-UniLiv/IDEA2

Multivariate Hawkes processes are a widely used class of self-exciting point processes, but maximum likelihood estimation naively scales as $O(N^2)$ in the number of events. The canonical linear exponential Hawkes process admits a faster $O(N)$ recurrence, but prior work evaluates this recurrence sequentially, without exploiting parallelization on modern GPUs. We show that the Hawkes process intensity can be expressed as a product of sparse transition matrices admitting a linear-time associative multiply, enabling computation via a parallel prefix scan. This yields a simple yet massively parallelizable algorithm for maximum likelihood estimation of linear exponential Hawkes processes. Our method reduces the computational complexity to approximately $O(N/P)$ with $P$ parallel processors, and naturally yields a batching scheme to maintain constant memory usage, avoiding GPU memory constraints. Importantly, it computes the exact likelihood without any additional assumptions or approximations, preserving the simplicity and interpretability of the model. We demonstrate orders-of-magnitude speedups on simulated and real datasets, scaling to thousands of nodes and tens of millions of events, substantially beyond scales reported in prior work. We provide an open-source PyTorch library implementing our optimizations.

Vision transformers (ViT) rely on attention mechanism to weigh input features, and therefore attention scores have naturally been considered as explanations for its decision-making process. However, attention scores are almost always non-zero, resulting in noisy and diffused attention maps and limiting interpretability. Can we quantify uncertainty measures of attention scores and obtain regularized attention scores? To this end, we consider attention scores of ViT in a statistical framework where independent noise would lead to insignificant yet non-zero scores. Leveraging statistical learning techniques, we introduce the bootstrapping for attention scores which generates a baseline distribution of attention scores by resampling input features. Such a bootstrap distribution is then used to estimate significances and posterior probabilities of attention scores. In natural and medical images, the proposed \emph{Attention Regularization} approach demonstrates a straightforward removal of spurious attention arising from noise, drastically improving shrinkage and sparsity. Quantitative evaluations are conducted using both simulation and real-world datasets. Our study highlights bootstrapping as a practical regularization tool when using attention scores as explanations for ViT. Code available: https://github.com/ncchung/AttentionRegularization

While deep learning has significantly advanced accident anticipation, the robustness of these safety-critical systems against real-world perturbations remains a major challenge. We reveal that state-of-the-art models like CRASH, despite their high performance, exhibit significant instability in predictions and latent representations when faced with minor input perturbations, posing serious reliability risks. To address this, we introduce SECURE - Stable Early Collision Understanding Robust Embeddings, a framework that formally defines and enforces model robustness. SECURE is founded on four key attributes: consistency and stability in both prediction space and latent feature space. We propose a principled training methodology that fine-tunes a baseline model using a multi-objective loss, which minimizes divergence from a reference model and penalizes sensitivity to adversarial perturbations. Experiments on DAD and CCD datasets demonstrate that our approach not only significantly enhances robustness against various perturbations but also improves performance on clean data, achieving new state-of-the-art results.

We investigate the data-driven discovery of constitutive closures in nonlinear reaction-diffusion systems with known governing PDE structures. Our objective is to robustly recover diffusion and reaction laws from spatiotemporal observations while avoiding the common pitfall where low residuals or short-horizon predictions are conflated with physical recovery. We propose a three-stage neural-symbolic framework: (1) learning numerical surrogates under physical constraints using a noise-robust weak-form-driven objective; (2) compressing these surrogates into restricted interpretable symbolic families (e.g., polynomial, rational, and saturation forms); and (3) validating the symbolic closures through explicit forward re-simulation on unseen initial conditions. Extensive numerical experiments reveal two distinct regimes. Under matched-library settings, weak polynomial baselines behave as correctly specified reference estimators, showing that neural surrogates do not uniformly outperform classical bases. Conversely, under function-class mismatch, neural surrogates provide necessary flexibility and can be compressed into compact symbolic laws with minimal rollout degradation. However, we identify a critical "bias inheritance" mechanism where symbolic compression does not automatically repair constitutive bias. Across various observation regimes, the true error of the symbolic closure closely tracks that of the neural surrogate, yielding a bias inheritance ratio near one. These findings demonstrate that the primary bottleneck in neural-symbolic modeling lies in the initial numerical inverse problem rather than the subsequent symbolic compression. We underscore that constitutive claims must be rigorously supported by forward validation rather than residual minimization alone.

While deepfake speech detectors built on large self-supervised learning (SSL) models achieve high accuracy, employing standard ensemble fusion to further enhance robustness often results in oversized systems with diminishing returns. To address this, we propose an evolutionary multi-objective score fusion framework that jointly minimizes detection error and system complexity. We explore two encodings optimized by NSGA-II: binary-coded detector selection for score averaging and a real-valued scheme that optimizes detector weights for a weighted sum. Experiments on the ASVspoof 5 dataset with 36 SSL-based detectors show that the obtained Pareto fronts outperform simple averaging and logistic regression baselines. The real-valued variant achieves 2.37% EER (0.0684 minDCF) and identifies configurations that match state-of-the-art performance while significantly reducing system complexity, requiring only half the parameters. Our method also provides a diverse set of trade-off solutions, enabling deployment choices that balance accuracy and computational cost.

Model merging provides a way of cheaply combining individual models to produce a model that inherits each individual's capabilities. While some merging methods can approach the performance of multitask training, they are often heuristically motivated and lack theoretical justification. A principled alternative is to pose model merging as a layer-wise optimization problem that directly minimizes interference between tasks. However, this formulation requires estimating per-layer covariance matrices from data, which may not be available when performing merging. In contrast, many of the heuristically-motivated methods do not require auxiliary data, making them practically advantageous. In this work, we revisit the interference minimization framework and show that, under certain conditions, covariance matrices can be estimated directly from difference matrices, eliminating the need for data while also reducing computational costs. We validate our approach across vision and language benchmarks on models ranging from 86M parameters to 7B parameters, outperforming previous data-free state-of-the-art merging methods

Traditional scientific discovery relies on an iterative hypothesise-experiment-refine cycle that has driven progress for centuries, but its intuitive, ad-hoc implementation often wastes resources, yields inefficient designs, and misses critical insights. This tutorial presents Bayesian Optimisation (BO), a principled probability-driven framework that formalises and automates this core scientific cycle. BO uses surrogate models (e.g., Gaussian processes) to model empirical observations as evolving hypotheses, and acquisition functions to guide experiment selection, balancing exploitation of known knowledge and exploration of uncharted domains to eliminate guesswork and manual trial-and-error. We first frame scientific discovery as an optimisation problem, then unpack BO's core components, end-to-end workflows, and real-world efficacy via case studies in catalysis, materials science, organic synthesis, and molecule discovery. We also cover critical technical extensions for scientific applications, including batched experimentation, heteroscedasticity, contextual optimisation, and human-in-the-loop integration. Tailored for a broad audience, this tutorial bridges AI advances in BO with practical natural science applications, offering tiered content to empower cross-disciplinary researchers to design more efficient experiments and accelerate principled scientific discovery.

Macroscopic unmanned aerial vehicle (UAV) traffic organization in three-dimensional airspace faces significant challenges from static wind fields and complex obstacles. A critical difficulty lies in simultaneously capturing the strong anisotropy induced by wind while strictly preserving transport consistency and boundary semantics, which are often compromised in standard physics-informed learning approaches. To resolve this, we propose a constraint-preserving hybrid solver that integrates a physics-informed neural network for the anisotropic Eikonal value problem with a conservative finite-volume method for steady density transport. These components are coupled through an outer Picard iteration with under-relaxation, where the target condition is hard-encoded and strictly conservative no-flux boundaries are enforced during the transport step. We evaluate the framework on reproducible homing and point-to-point scenarios, effectively capturing value slices, induced-motion patterns, and steady density structures such as bands and bottlenecks. Ultimately, our perspective emphasizes the value of a reproducible computational framework supported by transparent empirical diagnostics to enable the traceable assessment of macroscopic traffic phenomena.

The fundamental problem of causal inference - that the counterfactual outcome for any individual is never observed - has shaped the entire methodology of the field. Every existing approach substitutes assumptions for missing data: ignorability, parallel trends, exclusion restrictions. None produces the counterfactual itself. This paper proposes the Digital Twin Counterfactual Framework (DTCF): rather than estimating the counterfactual statistically, we simulate it using a digital twin and subject the simulation to a hierarchical validation regime. We formalize the digital twin simulator as a stochastic mapping within the potential outcomes framework and introduce a hierarchy of twin fidelity assumptions - from marginal fidelity through joint fidelity to structural fidelity - each unlocking a progressively richer class of estimands. The central contribution is threefold. First, a five-level validation architecture converts the unfalsifiable claim that the simulator produces correct counterfactuals into falsifiable tests against observable data. Second, a formal decomposition separates causal quantities into those that are marginally validated (ATE, CATE, QTE - testable through observable-arm comparison) and those that are copula-dependent (the ITE distribution, probability of benefit/harm, variance of treatment effects - permanently reliant on the unobservable within-individual dependence structure). Third, bounding, sensitivity, and uncertainty quantification tools make the copula dependence explicit. The DTCF does not resolve the fundamental problem of causal inference. What it provides is a framework in which marginal causal claims become increasingly testable, joint causal claims become explicitly assumption-indexed, and the gap between the two is formally characterized.

Money launderers take advantage of limitations in existing detection approaches by hiding their financial footprints in a deceitful manner. They manage this by replicating transaction patterns that the monitoring systems cannot easily distinguish. As a result, criminally gained assets are pushed into legitimate financial channels without drawing attention. Algorithms developed to monitor money flows often struggle with scale and complexity. The difficulty of identifying such activities is further intensified by the (persistent) inability of current solutions to control the excessive number of false positive signals produced by rigid, risk-based rules systems. We propose a framework called ReDiRect (REduce, DIstribute, and RECTify), specifically designed to overcome these challenges. The primary contribution of our work is a novel framing of this problem in an unsupervised setting; where a large transaction graph is fuzzily partitioned into smaller, manageable components to enable fast processing in a distributed manner. In addition, we define a refined evaluation metric that better captures the effectiveness of exposed money laundering patterns. Through comprehensive experimentation, we demonstrate that our framework achieves superior performance compared to existing and state-of-the-art techniques, particularly in terms of efficiency and real-world applicability. For validation, we used the real (open source) Libra dataset and the recently released synthetic datasets by IBM Watson. Our code and datasets are available at https://github.com/mhaseebtariq/redirect.

High-fidelity Monte Carlo simulations and complex inverse problems, such as mapping smeared experimental observations to ground-truth states, are computationally intensive yet essential for robust data analysis. Conditional Flow Matching (CFM) offers a mathematically robust approach to accelerating these tasks, but we demonstrate its standard training loss is fundamentally misleading. In rigorous physics applications, CFM loss plateaus prematurely, serving as an unreliable indicator of true convergence and physical fidelity. To investigate this disconnect, we designed JetPrism, a configurable CFM framework acting as an efficient generative surrogate for evaluating unconditional generation and conditional detector unfolding. Using synthetic stress tests and a Jefferson Lab kinematic dataset ($γp \to ρ^0 p \to π^+π^- p$) relevant to the forthcoming Electron-Ion Collider (EIC), we establish that physics-informed metrics continue to improve significantly long after the standard loss converges. Consequently, we propose a multi-metric evaluation protocol incorporating marginal and pairwise $χ^2$ statistics, $W_1$ distances, correlation matrix distances ($D_{\mathrm{corr}}$), and nearest-neighbor distance ratios ($R_{\mathrm{NN}}$). By demonstrating that domain-specific evaluations must supersede generic loss metrics, this work establishes JetPrism as a dependable generative surrogate that ensures precise statistical agreement with ground-truth data without memorizing the training set. While demonstrated in nuclear physics, this diagnostic framework is readily extensible to parameter generation and complex inverse problems across broad domains. Potential applications span medical imaging, astrophysics, semiconductor discovery, and quantitative finance, where high-fidelity simulation, rigorous inversion, and generative reliability are critical.

Learning human preferences in language models remains fundamentally challenging, as reward modeling relies on subtle, subjective comparisons or shades of gray rather than clear-cut labels. This study investigates the limits of current approaches and proposes a feature-augmented framework to better capture the multidimensional nature of human judgment. Using the Anthropic HHRLHF dataset, we evaluate ten diverse large language models LLMs under a standard pairwise preference setting, where baseline performance remains below 0.74 ROC AUC, highlighting the difficulty of the task. To address this, we enrich textual representations with interpretable signals: response length, refusal indicators, toxicity scores and prompt response semantic similarity, enabling models to explicitly capture key aspects of helpfulness, safety and relevance. The proposed hybrid approach yields consistent improvements across all models, achieving up to 0.84 ROC AUC and significantly higher pairwise accuracy, with DeBERTav3Large demonstrating the best performance. Beyond accuracy, we integrate SHAP and LIME to provide fine-grained interpretability, revealing that model decisions depend on contextualized safety and supportive framing rather than isolated keywords. We further analyze bias amplification, showing that while individual features have weak marginal effects, their interactions influence preference learning.

Designing reliable integrated energy systems for industrial processes requires optimization and verification models across multiple fidelities, from architecture-level sizing to high-fidelity dynamic operation. However, model mismatch across fidelities obscures the sources of performance loss and complicates the quantification of architecture-to-operation performance gaps. We propose an online, machine-learning-accelerated multi-resolution optimization framework that estimates an architecture-specific upper bound on achievable performance while minimizing expensive high-fidelity model evaluations. We demonstrate the approach on a pilot energy system supplying a 1 MW industrial heat load. First, we solve a multi-objective architecture optimization to select the system configuration and component capacities. We then develop an machine learning (ML)-accelerated multi-resolution, receding-horizon optimal control strategy that approaches the achievable-performance bound for the specified architecture, given the additional controls and dynamics not captured by the architectural optimization model. The ML-guided controller adaptively schedules the optimization resolution based on predictive uncertainty and warm-starts high-fidelity solves using elite low-fidelity solutions. Our results on the pilot case study show that the proposed multi-resolution strategy reduces the architecture-to-operation performance gap by up to 42% relative to a rule-based controller, while reducing required high-fidelity model evaluations by 34% relative to the same multi-fidelity approach without ML guidance, enabling faster and more reliable design verification. Together, these gains make high-fidelity verification tractable, providing a practical upper bound on achievable operational performance.

Evaluating scientific arguments requires assessing the strict consistency between a claim and its underlying multimodal evidence. However, existing benchmarks lack the scale, domain diversity, and visual complexity needed to evaluate this alignment realistically. To address this gap, we introduce M2-Verify, a large-scale multimodal dataset for checking scientific claim consistency. Sourced from PubMed and arXiv, M2-Verify provides over 469K instances across 16 domains, rigorously validated through expert audits. Extensive baseline experiments show that state-of-the-art models struggle to maintain robust consistency. While top models achieve up to 85.8\% Micro-F1 on low-complexity medical perturbations, performance drops to 61.6\% on high-complexity challenges like anatomical shifts. Furthermore, expert evaluations expose hallucinations when models generate scientific explanations for their alignment decisions. Finally, we demonstrate our dataset's utility and provide comprehensive usage guidelines.

Reconstructing high-dimensional spatiotemporal fields from sparse sensor measurements is critical in a wide range of scientific applications. The SHallow REcurrent Decoder (SHRED) architecture is a recent state-of-the-art architecture that reconstructs high-quality spatial domain from hyper-sparse sensor measurement streams. An important limitation of SHRED is that in complex, data-scarce, high-frequency, or stochastic systems, portions of the spatiotemporal field must be modeled with valid uncertainty estimation. We introduce UQ-SHRED, a distributional learning framework for sparse sensing problems that provides uncertainty quantification through a neural network-based distributional regression called engression. UQ-SHRED models the uncertainty by learning the predictive distribution of the spatial state conditioned on the sensor history. By injecting stochastic noise into sensor inputs and training with an energy score loss, UQ-SHRED produces predictive distributions with minimal computational overhead, requiring only noise injection at the input and resampling through a single architecture without retraining or additional network structures. On complicated synthetic and real-life datasets including turbulent flow, atmospheric dynamics, neuroscience and astrophysics, UQ-SHRED provides a distributional approximation with well-calibrated confidence intervals. We further conduct ablation studies to understand how each model setting affects the quality of the UQ-SHRED performance, and its validity on uncertainty quantification over a set of different experimental setups.

We study how to scale reasoning token budgets for competitive programming through two complementary approaches: training-time reinforcement learning (RL) and test-time parallel thinking. During RL training, we observe an approximately log-linear relationship between validation accuracy and the average number of generated reasoning tokens over successive checkpoints, and show two ways to shift this training trajectory: verification RL warmup raises the starting point, while randomized clipping produces a steeper trend in the observed regime. As scaling single-generation reasoning during RL quickly becomes expensive under full attention, we introduce a multi-round parallel thinking pipeline that distributes the token budget across threads and rounds of generation, verification, and refinement. We train the model end-to-end on this pipeline to match the training objective to the test-time structure. Starting from Seed-OSS-36B, the full system with 16 threads and 16 rounds per thread matches the underlying RL model's oracle pass@16 at pass@1 using 7.6 million tokens per problem on average, and surpasses GPT-5-high on 456 hard competitive programming problems from AetherCode.

Anticipating supply chain disruptions before they materialize is a core challenge for firms and policymakers alike. A key difficulty is learning to reason reliably about infrequent, high-impact events from noisy and unstructured inputs - a setting where general-purpose models struggle without task-specific adaptation. We introduce an end-to-end framework that trains LLMs to produce calibrated probabilistic forecasts using realized disruption outcomes as supervision. The resulting model substantially outperforms strong baselines - including GPT-5 - on accuracy, calibration, and precision. We also show that training induces more structured and reliable probabilistic reasoning without explicit prompting. These results suggest a general pathway for training domain-specific forecasting models that produce decision-ready signals. To support transparency we open-source the evaluation dataset used in this study. Dataset: https://huggingface.co/datasets/LightningRodLabs/supply-chain-predictions

Answering questions about images often requires combining visual understanding with external knowledge. Multimodal Large Language Models (MLLMs) provide a natural framework for this setting, but they often struggle to identify the most relevant visual and textual evidence when answering knowledge-intensive queries. In such scenarios, models must integrate visual cues with retrieved textual evidence that is often noisy or only partially relevant, while also localizing fine-grained visual information in the image. In this work, we introduce Look Twice (LoT), a training-free inference-time framework that improves how pretrained MLLMs utilize multimodal evidence. Specifically, we exploit the model attention patterns to estimate which visual regions and retrieved textual elements are relevant to a query, and then generate the answer conditioned on this highlighted evidence. The selected cues are highlighted through lightweight prompt-level markers that encourage the model to re-attend to the relevant evidence during generation. Experiments across multiple knowledge-based VQA benchmarks show consistent improvements over zero-shot MLLMs. Additional evaluations on vision-centric and hallucination-oriented benchmarks further demonstrate that visual evidence highlighting alone improves model performance in settings without textual context, all without additional training or architectural modifications. Source code will be publicly released.

We introduce Sven (Singular Value dEsceNt), a new optimization algorithm for neural networks that exploits the natural decomposition of loss functions into a sum over individual data points, rather than reducing the full loss to a single scalar before computing a parameter update. Sven treats each data point's residual as a separate condition to be satisfied simultaneously, using the Moore-Penrose pseudoinverse of the loss Jacobian to find the minimum-norm parameter update that best satisfies all conditions at once. In practice, this pseudoinverse is approximated via a truncated singular value decomposition, retaining only the $k$ most significant directions and incurring a computational overhead of only a factor of $k$ relative to stochastic gradient descent. This is in comparison to traditional natural gradient methods, which scale as the square of the number of parameters. We show that Sven can be understood as a natural gradient method generalized to the over-parametrized regime, recovering natural gradient descent in the under-parametrized limit. On regression tasks, Sven significantly outperforms standard first-order methods including Adam, converging faster and to a lower final loss, while remaining competitive with LBFGS at a fraction of the wall-time cost. We discuss the primary challenge to scaling, namely memory overhead, and propose mitigation strategies. Beyond standard machine learning benchmarks, we anticipate that Sven will find natural application in scientific computing settings where custom loss functions decompose into several conditions.

In this paper, we attempt to explore the landscape of two-dimensional conformal field theories (2d CFTs) by efficiently searching for numerical solutions to the modular bootstrap equation using machine-learning-style optimization. The torus partition function of a 2d CFT is fixed by the spectrum of its primary operators and its chiral algebra, which we take to be the Virasoro algebra with $c>1$. We translate the requirement that this partition function is modular invariant into a loss function, which we then minimize to identify possible primary spectra. Our approach involves two technical innovations that facilitate finding reliable candidate CFTs. The first is a strategy to estimate the uncertainty associated with truncating the spectrum to the lowest dimension operators. The second is the use of a new singular-value-based optimizer (Sven) that is more effective than gradient descent at navigating the hierarchical structure of the loss landscape. We numerically construct candidate truncated CFT partition functions with central charges between 1 and $\frac{8}{7}$, a range devoid of known examples, and argue that these candidates likely come from a continuous space of modular bootstrap solutions. We also provide evidence for a more stringent constraint on the spectral gap near $c = 1$ than the existing bound of $Δ_{\rm gap} \le \frac{c}{6} + \frac{1}{3}$.

We present HippoCamp, a new benchmark designed to evaluate agents' capabilities on multimodal file management. Unlike existing agent benchmarks that focus on tasks like web interaction, tool use, or software automation in generic settings, HippoCamp evaluates agents in user-centric environments to model individual user profiles and search massive personal files for context-aware reasoning. Our benchmark instantiates device-scale file systems over real-world profiles spanning diverse modalities, comprising 42.4 GB of data across over 2K real-world files. Building upon the raw files, we construct 581 QA pairs to assess agents' capabilities in search, evidence perception, and multi-step reasoning. To facilitate fine-grained analysis, we provide 46.1K densely annotated structured trajectories for step-wise failure diagnosis. We evaluate a wide range of state-of-the-art multimodal large language models (MLLMs) and agentic methods on HippoCamp. Our comprehensive experiments reveal a significant performance gap: even the most advanced commercial models achieve only 48.3% accuracy in user profiling, struggling particularly with long-horizon retrieval and cross-modal reasoning within dense personal file systems. Furthermore, our step-wise failure diagnosis identifies multimodal perception and evidence grounding as the primary bottlenecks. Ultimately, HippoCamp exposes the critical limitations of current agents in realistic, user-centric environments and provides a robust foundation for developing next-generation personal AI assistants.

The rise of test-time scaling has remarkably boosted the reasoning and agentic proficiency of Large Language Models (LLMs). Yet, standard Transformers struggle to scale inference-time compute efficiently, as conventional looping strategies suffer from high computational overhead and a KV cache that inflates alongside model depth. We present Universal YOCO (YOCO-U), which combines the YOCO decoder-decoder architecture with recursive computation to achieve a synergistic effect greater than either alone. Built on the YOCO framework, YOCO-U implements a Universal Self-Decoder that performs multiple iterations via parameter sharing, while confining the iterative process to shallow, efficient-attention layers. This combination yields a favorable capability-efficiency tradeoff that neither YOCO nor recursion achieves independently. The YOCO architecture provides a constant global KV cache and linear pre-filling, while partial recursion enhances representational depth with limited overhead. Together, YOCO-U improves token utility and scaling behavior while maintaining efficient inference. Empirical results confirm that YOCO-U remains highly competitive in general and long-context benchmarks, demonstrating that the integration of efficient-attention architectures and recursive computation is a promising direction for scalable LLMs.

Reconstructing full spatio-temporal dynamics from sparse observations in both space and time remains a central challenge in complex systems, as measurements can be spatially incomplete and can be also limited to narrow temporal windows. Yet approximating the complete spatio-temporal trajectory is essential for mechanistic insight and understanding, model calibration, and operational decision-making. We introduce LAPIS-SHRED (LAtent Phase Inference from Short time sequence using SHallow REcurrent Decoders), a modular architecture that reconstructs and/or forecasts complete spatiotemporal dynamics from sparse sensor observations confined to short temporal windows. LAPIS-SHRED operates through a three-stage pipeline: (i) a SHRED model is pre-trained entirely on simulation data to map sensor time-histories into a structured latent space, (ii) a temporal sequence model, trained on simulation-derived latent trajectories, learns to propagate latent states forward or backward in time to span unobserved temporal regions from short observational time windows, and (iii) at deployment, only a short observation window of hyper-sparse sensor measurements from the true system is provided, from which the frozen SHRED model and the temporal model jointly reconstruct or forecast the complete spatiotemporal trajectory. The framework supports bidirectional inference, inherits data assimilation and multiscale reconstruction capabilities from its modular structure, and accommodates extreme observational constraints including single-frame terminal inputs. We evaluate LAPIS-SHRED on six experiments spanning complex spatio-temporal physics: turbulent flows, multiscale propulsion physics, volatile combustion transients, and satellite-derived environmental fields, highlighting a lightweight, modular architecture suited for operational settings where observation is constrained by physical or logistical limitations.

AI weather prediction has advanced rapidly, yet no unified mathematical framework explains what determines forecast skill. Existing theory addresses specific architectural choices rather than the learning pipeline as a whole, while operational evidence from 2023-2026 demonstrates that training methodology, loss function design, and data diversity matter at least as much as architecture selection. This paper makes two interleaved contributions. Theoretically, we construct a framework rooted in approximation theory on the sphere, dynamical systems theory, information theory, and statistical learning theory that treats the complete learning pipeline (architecture, loss function, training strategy, data distribution) rather than architecture alone. We establish a Learning Pipeline Error Decomposition showing that estimation error (loss- and data-dependent) dominates approximation error (architecture-dependent) at current scales. We develop a Loss Function Spectral Theory formalizing MSE-induced spectral blurring in spherical harmonic coordinates, and derive Out-of-Distribution Extrapolation Bounds proving that data-driven models systematically underestimate record-breaking extremes with bias growing linearly in record exceedance. Empirically, we validate these predictions via inference across ten architecturally diverse AI weather models using NVIDIA Earth2Studio with ERA5 initial conditions, evaluating six metrics across 30 initialization dates spanning all seasons. Results confirm universal spectral energy loss at high wavenumbers for MSE-trained models, rising Error Consensus Ratios showing that the majority of forecast error is shared across architectures, and linear negative bias during extreme events. A Holistic Model Assessment Score provides unified multi-dimensional evaluation, and a prescriptive framework enables mathematical evaluation of proposed pipelines before training.

As LLM agents tackle increasingly complex tasks, a critical question is whether they can maintain strategic coherence over long horizons: planning under uncertainty, learning from delayed feedback, and adapting when early mistakes compound. We introduce $\texttt{YC-Bench}$, a benchmark that evaluates these capabilities by tasking an agent with running a simulated startup over a one-year horizon spanning hundreds of turns. The agent must manage employees, select task contracts, and maintain profitability in a partially observable environment where adversarial clients and growing payroll create compounding consequences for poor decisions. We evaluate 12 models, both proprietary and open source, across 3 seeds each. Only three models consistently surpass the starting capital of \$200K, with Claude Opus 4.6 achieving the highest average final funds at \$1.27 M, followed by GLM-5 at \$1.21 M at 11$\times$ lower inference cost. Scratchpad usage, the sole mechanism for persisting information across context truncation, is the strongest predictor of success, and adversarial client detection is the primary failure mode, accounting for $47\%$ of bankruptcies. Our analysis reveals that frontier models still fail through distinct failure modes such as over-parallelization, demonstrating the capability gaps for long-horizon performance. $\texttt{YC-Bench}$ is open-source, reproducible, and configurable.

Scientific algorithm discovery is iterative: hypotheses are proposed, implemented, stress-tested, and revised. Current LLM-guided search systems accelerate proposal generation, but often under-represent scientific structure by optimizing code-only artifacts with weak correctness/originality gating. We present CliffSearch, an agentic evolutionary framework in which the core evolution operators (pair selection, crossover, mutation, and review) are implemented as LLM agents, and the loop is designed around three principles: (1) each node is a structured scientific artifact, instantiated in either theory+code or code_only mode, (2) reviewer judgments of correctness and originality are first-class selection gates alongside optimization of the benchmark metric of interest, and (3) mutation is split into exploration and correction pathways with distinct objectives. Exploration mutation imports ideas from adjacent scientific domains to increase novelty, while correction mutation performs targeted evidence-guided repair using reviewer signals over theory, code, benchmark results, and runtime errors. We illustrate the framework on three benchmark-grounded studies: transformer hyper-connection evolution, optimizer discovery on a fixed nanoGPT stack, and a smaller native-optimizer ablation. Across these settings, the same loop supports explicit metric direction, reproducible persistence, and reviewer-gated comparison of discoveries under controlled search conditions. The result is a discovery workflow that prioritizes scientific interpretability and correctness while optimizing task metrics under controlled novelty constraints, rather than maximizing candidate throughput alone. Full run artifacts, interactive visualizations, and exported best nodes for the reported studies are available at https://cliffsearch.ai .

We present RELISH (REgression with a Latent Iterative State Head), a novel, lightweight architecture designed for text regression with large language models. Rather than decoding numeric targets as text or aggregating multiple generated outputs, RELISH predicts scalar values directly from frozen LLM representations by iteratively refining a learned latent state through cross-attention over token-level representations, and then mapping the final state to a point estimate with a linear regressor. Across five datasets, four LLM backbones, and two LLM training regimes, RELISH consistently outperforms prior baselines from all three major LLM regression families, including autoregressive decoding, regression-aware inference, and existing predictive head methods. Despite these gains, RELISH remains highly parameter-efficient, requiring only 3.4-3.7M trainable parameters across frozen LLM backbones (only 0.01-0.04% additional overhead), far less than LoRA-based alternatives that grow with model size (0.26-0.42%).

Primitive-based methods such as 3D Gaussian Splatting have recently become the state-of-the-art for novel-view synthesis and related reconstruction tasks. Compared to neural fields, these representations are more flexible, adaptive, and scale better to large scenes. However, the limited expressivity of individual primitives makes modeling high-frequency detail challenging. We introduce Neural Harmonic Textures, a neural representation approach that anchors latent feature vectors on a virtual scaffold surrounding each primitive. These features are interpolated within the primitive at ray intersection points. Inspired by Fourier analysis, we apply periodic activations to the interpolated features, turning alpha blending into a weighted sum of harmonic components. The resulting signal is then decoded in a single deferred pass using a small neural network, significantly reducing computational cost. Neural Harmonic Textures yield state-of-the-art results in real-time novel view synthesis while bridging the gap between primitive- and neural-field-based reconstruction. Our method integrates seamlessly into existing primitive-based pipelines such as 3DGUT, Triangle Splatting, and 2DGS. We further demonstrate its generality with applications to 2D image fitting and semantic reconstruction.

We consider the question: when a large language reasoning model makes a choice, did it think first and then decide to, or decide first and then think? In this paper, we present evidence that detectable, early-encoded decisions shape chain-of-thought in reasoning models. Specifically, we show that a simple linear probe successfully decodes tool-calling decisions from pre-generation activations with very high confidence, and in some cases, even before a single reasoning token is produced. Activation steering supports this causally: perturbing the decision direction leads to inflated deliberation, and flips behavior in many examples (between 7 - 79% depending on model and benchmark). We also show through behavioral analysis that, when steering changes the decision, the chain-of-thought process often rationalizes the flip rather than resisting it. Together, these results suggest that reasoning models can encode action choices before they begin to deliberate in text.

Learning quantum states from measurement data is a central problem in quantum information and computational complexity. In this work, we study the problem of learning to generate mixed states on a finite-dimensional lattice. Motivated by recent developments in mixed state phases of matter, we focus on arbitrary states in the trivial phase. A state belongs to the trivial phase if there exists a shallow preparation channel circuit under which local reversibility is preserved throughout the preparation. We prove that any mixed state in this class can be efficiently learned from measurement access alone. Specifically, given copies of an unknown trivial phase mixed state, our algorithm outputs a shallow local channel circuit that approximately generates this state in trace distance. The sample complexity and runtime are polynomial (or quasi-polynomial) in the number of qubits, assuming constant (or polylogarithmic) circuit depth and gate locality. Importantly, the learner is not given the original preparation circuit and relies only on its existence. Our results provide a structural foundation for quantum generative models based on shallow channel circuits. In the classical limit, our framework also inspires an efficient algorithm for classical diffusion models using only a polynomial overhead of training and generation.

Search agents, which integrate language models (LMs) with web search, are becoming crucial for answering complex user queries. Constructing training datasets for deep research tasks, involving multi-step retrieval and reasoning, remains challenging due to expensive human annotation, or cumbersome prerequisites. In this work, we introduce ORBIT, a training dataset with 20K reasoning-intensive queries with short verifiable answers, generated using a frugal framework without relying on paid API services. The modular framework relies on four stages: seed creation, question-answer pair generation, and two stages of verification: self and external. ORBIT spans 15 domains and each training pair requires 4-5 reasoning steps, with external search verification required from the complete web. We train Qwen3-4B as the base model on ORBIT using GRPO and evaluate it on Wikipedia question answering tasks. Extensive experiment results demonstrate that ORBIT-4B achieves strong performance among sub-4B LLMs as search agents, proving the utility of synthetic datasets. Our framework, code and datasets are open-sourced and available publicly.

Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instruct and thinking variants. To understand why such a simple method can work, we trace these gains to a precision-exploration conflict in LLM decoding and show that SSD reshapes token distributions in a context-dependent way, suppressing distractor tails where precision matters while preserving useful diversity where exploration matters. Taken together, SSD offers a complementary post-training direction for improving LLM code generation.

This study investigates the ability of multimodal Large Language Models (LLMs) to identify and interpret misleading visualizations, and recognize these observations along with their underlying causes and potential intentionality. Our analysis leverages concepts from visualization rhetoric and a newly developed taxonomy of authorial intents as explanatory lenses. We formulated three research questions and addressed them experimentally using a dataset of 2,336 COVID-19-related tweets, half of which contain misleading visualizations, and supplemented it with real-world examples of perceptual, cognitive, and conceptual errors drawn from VisLies, the IEEE VIS community event dedicated to showcasing deceptive and misleading visualizations. To ensure broad coverage of the current LLM landscape, we evaluated 16 state-of-the-art models. Among them, 15 are open-weight models, spanning a wide range of model sizes, architectural families, and reasoning capabilities. The selection comprises small models, namely Nemotron-Nano-V2-VL (12B parameters), Mistral-Small-3.2 (24B), DeepSeek-VL2 (27B), Gemma3 (27B), and GTA1 (32B); medium-sized models, namely Qianfan-VL (70B), Molmo (72B), GLM-4.5V (108B), LLaVA-NeXT (110B), and Pixtral-Large (124B); and large models, namely Qwen3-VL (235B), InternVL3.5 (241B), Step3 (321B), Llama-4-Maverick (400B), and Kimi-K2.5 (1000B). In addition, we employed OpenAI GPT-5.4, a frontier proprietary model. To establish a human perspective on these tasks, we also conducted a user study with visualization experts to assess how people perceive rhetorical techniques and the authorial intentions behind the same misleading visualizations. This allows comparison between model and expert behavior, revealing similarities and differences that provide insights into where LLMs align with human judgment and where they diverge.

Foundation vision-language models are becoming increasingly relevant to robotics because they can provide richer semantic perception than narrow task-specific pipelines. However, their practical adoption in robot software stacks still depends on reproducible middleware integrations rather than on model quality alone. Florence-2 is especially attractive in this regard because it unifies captioning, optical character recognition, open-vocabulary detection, grounding and related vision-language tasks within a comparatively manageable model size. This article presents a ROS 2 wrapper for Florence-2 that exposes the model through three complementary interaction modes: continuous topic-driven processing, synchronous service calls and asynchronous actions. The wrapper is designed for local execution and supports both native installation and Docker container deployment. It also combines generic JSON outputs with standard ROS 2 message bindings for detection-oriented tasks. A functional validation is reported together with a throughput study on several GPUs, showing that local deployment is feasible with consumer grade hardware. The repository is publicly available here: https://github.com/JEDominguezVidal/florence2_ros2_wrapper

A core limitation of standard softmax attention is that it does not define a notion of absolute query--key relevance: attention weights are obtained by redistributing a fixed unit mass across all keys according to their relative scores. As a result, relevance is defined only relative to competing keys, and irrelevant keys cannot be explicitly rejected. We introduce Multiscreen, a language-model architecture built around a mechanism we call screening, which enables absolute query--key relevance. Instead of redistributing attention across all keys, screening evaluates each key against an explicit threshold, discarding irrelevant keys and aggregating the remaining keys, thereby removing global competition among keys. Across experiments, Multiscreen achieves comparable validation loss with approximately 40% fewer parameters than a Transformer baseline, enables stable optimization at substantially larger learning rates, maintains strong performance in long-context perplexity, shows little to no degradation in retrieval performance even far beyond the training context length, and reduces inference latency by up to 3.2$\times$ at 100K context length.

Accurate air quality forecasting is crucial for protecting public health and guiding environmental policy, yet it remains challenging due to nonlinear spatiotemporal dynamics, wind-driven transport, and distribution shifts across regions. Physics-based models are interpretable but computationally expensive and often rely on restrictive assumptions, whereas purely data-driven models can be accurate but may lack robustness and calibrated uncertainty. To address these limitations, we propose Neural Dynamic Diffusion-Advection Fields (NeuroDDAF), a physics-informed forecasting framework that unifies neural representation learning with open-system transport modeling. NeuroDDAF integrates (i) a GRU-Graph Attention encoder to capture temporal dynamics and wind-aware spatial interactions, (ii) a Fourier-domain diffusion-advection module with learnable residuals, (iii) a wind-modulated latent Neural ODE to model continuous-time evolution under time-varying connectivity, and (iv) an evidential fusion mechanism that adaptively combines physics-guided and neural forecasts while quantifying uncertainty. Experiments on four urban datasets (Beijing, Shenzhen, Tianjin, and Ancona) across 1-3 day horizons show that NeuroDDAF consistently outperforms strong baselines, including AirPhyNet, achieving up to 9.7% reduction in RMSE and 9.4% reduction in MAE on long-term forecasts. On the Beijing dataset, NeuroDDAF attains an RMSE of 41.63 $μ$g/m$^3$ for 1-day prediction and 48.88 $μ$g/m$^3$ for 3-day prediction, representing the best performance among all compared methods. In addition, NeuroDDAF improves cross-city generalization and yields well-calibrated uncertainty estimates, as confirmed by ensemble variance analysis and case studies under varying wind conditions.

Uncertainty quantification is essential when deploying learning-based control methods in safety-critical systems. This is commonly realized by constructing uncertainty tubes that enclose the unknown function of interest, e.g., the reward and constraint functions or the underlying dynamics model, with high probability. However, existing approaches for uncertainty quantification typically rely on restrictive assumptions on the unknown function, such as known bounds on functional norms or Lipschitz constants, and struggle with discontinuities. In this paper, we model the unknown function as a random function from which independent and identically distributed realizations can be generated, and construct uncertainty tubes via the scenario approach that hold with high probability and rely solely on the sampled realizations. We integrate these uncertainty tubes into a safe Bayesian optimization algorithm, which we then use to safely tune control parameters on a real Furuta pendulum.

While test-time scaling has enabled large language models to solve highly difficult tasks, state-of-the-art results come at exorbitant compute costs. These inefficiencies can be attributed to the miscalibration of post-trained language models, and the lack of calibration in popular sampling techniques. Here, we present Online Reasoning Calibration (ORCA), a framework for calibrating the sampling process that draws upon conformal prediction and test-time training. Specifically, we introduce a meta-learning procedure that updates the calibration module for each input. This allows us to provide valid confidence estimates under distributional shift, e.g. in thought patterns that occur across different stages of reasoning, or in prompt distributions between model development and deployment. ORCA not only provides theoretical guarantees on conformal risks, but also empirically shows higher efficiency and generalization across different reasoning tasks. At risk level $δ=0.1$, ORCA improves Qwen2.5-32B efficiency on in-distribution tasks with savings up to 47.5% with supervised labels and 40.7% with self-consistency labels. Under zero-shot out-of-domain settings, it improves MATH-500 savings from 24.8% of the static calibration baseline to 67.0% while maintaining a low empirical error rate, and the same trend holds across model families and downstream benchmarks. Our code is publicly available at https://github.com/wzekai99/ORCA.

A fundamental challenge in science and engineering is the simulation-to-experiment gap. While we often possess prior knowledge of physical laws, these physical laws can be too difficult to solve exactly for complex systems. Such systems are commonly modeled using simulators, which impose computational approximations. Meanwhile, experimental measurements more faithfully represent the real world, but experimental data typically consists of observations that only partially reflect the system's full underlying state. We propose a data-driven distribution alignment framework that bridges this simulation-to-experiment gap by pre-training a generative model on fully observed (but imperfect) simulation data, then aligning it with partial (but real) observations of experimental data. While our method is domain-agnostic, we ground our approach in the physical sciences by introducing Adversarial Distribution Alignment (ADA). This method aligns a generative model of atomic positions -- initially trained on a simulated Boltzmann distribution -- with the distribution of experimental observations. We prove that our method recovers the target observable distribution, even with multiple, potentially correlated observables. We also empirically validate our framework on synthetic, molecular, and experimental protein data, demonstrating that it can align generative models with diverse observables. Our code is available at https://kaityrusnelson.com/ada/.

Using roughly 48 execution-verified HumanEval training solutions, tuning a single initial state matrix per recurrent layer, with zero inference overhead, outperforms LoRA by +10.8 pp (p < 0.001) on HumanEval. The method, which we call S0 tuning, optimizes one state matrix per recurrent layer while freezing all model weights. On Qwen3.5-4B (GatedDeltaNet hybrid), S0 tuning improves greedy pass@1 by +23.6 +/- 1.7 pp (10 seeds). On FalconH1-7B (Mamba-2 hybrid), S0 reaches 71.8% +/- 1.3 and LoRA reaches 71.4% +/- 2.4 (3 seeds), statistically indistinguishable at this sample size while requiring no weight merging. Cross-domain transfer is significant on MATH-500 (+4.8 pp, p = 0.00002, 8 seeds) and GSM8K (+2.8 pp, p = 0.0003, 10 seeds); a text-to-SQL benchmark (Spider) shows no transfer, consistent with the trajectory-steering mechanism. A prefix-tuning control on a pure Transformer (Qwen2.5-3B) degrades performance by -13.9 pp under all nine configurations tested. On Qwen3.5, a per-step state-offset variant reaches +27.1 pp, above both S0 and LoRA but with per-step inference cost. Taken together, the results show that recurrent state initialization is a strong zero-inference-overhead PEFT surface for hybrid language models when verified supervision is scarce. The tuned state is a ~48 MB file; task switching requires no weight merging or model reload. Code and library: https://github.com/jackyoung27/s0-tuning.

Chest X-ray (CXR) segmentation is an important step in computer-aided diagnosis, yet deploying large foundation models in clinical settings remains challenging due to computational constraints. We propose AdaLoRA-QAT, a two-stage fine-tuning framework that combines adaptive low-rank encoder adaptation with full quantization-aware training. Adaptive rank allocation improves parameter efficiency, while selective mixed-precision INT8 quantization preserves structural fidelity crucial for clinical reliability. Evaluated across large-scale CXR datasets, AdaLoRA-QAT achieves 95.6% Dice, matching full-precision SAM decoder fine-tuning while reducing trainable parameters by 16.6\times and yielding 2.24\times model compression. A Wilcoxon signed-rank test confirms that quantization does not significantly degrade segmentation accuracy. These results demonstrate that AdaLoRA-QAT effectively balances accuracy, efficiency, and structural trust-worthiness, enabling compact and deployable foundation models for medical image segmentation. Code and pretrained models are available at: https://prantik-pdeb.github.io/adaloraqat.github.io/

Large language models (LLMs) exhibiting test-time scaling behavior, such as extended reasoning traces and self-verification, have demonstrated remarkable performance on complex, long-term reasoning tasks. However, the robustness of these reasoning behaviors remains underexplored. To investigate this, we conduct a systematic evaluation of multiple reasoning models across three scenarios: (1) problems augmented with lengthy, irrelevant context; (2) multi-turn conversational settings with independent tasks; and (3) problems presented as a subtask within a complex task. We observe an interesting phenomenon: reasoning models tend to produce much shorter reasoning traces (up to 50%) for the same problem under different context conditions compared to the traces produced when the problem is presented in isolation. A finer-grained analysis reveals that this compression is associated with a decrease in self-verification and uncertainty management behaviors, such as double-checking. While this behavioral shift does not compromise performance on straightforward problems, it might affect performance on more challenging tasks. We hope our findings draw additional attention to both the robustness of reasoning models and the problem of context management for LLMs and LLM-based agents.

This paper argues that AI-enabled analysis of street-view imagery, complemented by performance-gated machine-learning imputation, provides a viable pathway for generating building-specific elevation data at regional scale for flood risk assessment. We develop and apply a three-stage pipeline across 18 areas of interest (AOIs) in Texas that (1) extracts LFE and the height difference between street grade and the lowest floor (HDSL) from Google Street View imagery using the Elev-Vision framework, (2) imputes missing HDSL values with Random Forest and Gradient Boosting models trained on 16 terrain, hydrologic, geographic, and flood-exposure features, and (3) integrates the resulting elevation dataset with Fathom 1-in-100 year inundation surfaces and USACE depth-damage functions to estimate property-specific interior flood depth and expected loss. Across 12,241 residential structures, street-view imagery was available for 73.4% of parcels and direct LFE/HDSL extraction was successful for 49.0% (5,992 structures). Imputation was retained for 13 AOIs where cross-validated performance was defensible, with selected models achieving R suqre values from 0.159 to 0.974; five AOIs were explicitly excluded from prediction because performance was insufficient. The results show that street-view-based elevation mapping is not universally available for every property, but it is sufficiently scalable to materially improve regional flood-risk characterization by moving beyond hazard exposure to structure-level estimates of interior inundation and expected damage. Scientifically, the study advances LFE estimation from a pilot-scale proof of concept to a regional, end-to-end workflow. Practically, it offers a replicable framework for jurisdictions that lack comprehensive Elevation Certificates but need parcel-level information to support mitigation, planning, and flood-risk management.

We present Brainstacks, a modular architecture for continual multi-domain fine-tuning of large language models that packages domain expertise as frozen adapter stacks composing additively on a shared frozen base at inference. Five interlocking components: (1) MoE-LoRA with Shazeer-style noisy top-2 routing across all seven transformer projections under QLoRA 4-bit quantization with rsLoRA scaling; (2) an inner loop performing residual boosting by freezing trained stacks and adding new ones; (3) an outer loop training sequential domain-specific stacks with curriculum-ordered dependencies; (4) null-space projection via randomized SVD constraining new stacks to subspaces orthogonal to prior directions, achieving zero forgetting in isolation; (5) an outcome-based sigmoid meta-router trained on empirically discovered domain-combination targets that selectively weights stacks, enabling cross-domain composition. Two boundary experiments: (6) PSN pretraining on a randomly initialized model; (7) per-domain RL (DPO/GRPO) validating compatibility with post-SFT alignment. Validated on TinyLlama-1.1B (4 domains, 9 stacks) and Gemma 3 12B IT (5 domains, 10 stacks), MoE-LoRA achieves 2.5x faster convergence than parameter-matched single LoRA, residual boosting breaks through the single-stack ceiling, and the routed system recovers generation quality destroyed by ungated stack accumulation. The central finding: the outcome-based router discovers that domain stacks encode transferable cognitive primitives (instruction-following clarity, numerical reasoning, procedural logic, chain-of-thought structure) rather than domain-specific knowledge, with medical prompts routing to chat+math stacks in 97% of cases despite zero medical data in those stacks.

As LLM agents are increasingly deployed in multi-agent systems, they introduce risks of covert coordination that may evade standard forms of human oversight. While linear probes on model activations have shown promise for detecting deception in single-agent settings, collusion is inherently a multi-agent phenomenon, and the use of internal representations for detecting collusion between agents remains unexplored. We introduce NARCBench, a benchmark for evaluating collusion detection under environment distribution shift, and propose five probing techniques that aggregate per-agent deception scores to classify scenarios at the group level. Our probes achieve 1.00 AUROC in-distribution and 0.60--0.86 AUROC when transferred zero-shot to structurally different multi-agent scenarios and a steganographic blackjack card-counting task. We find that no single probing technique dominates across all collusion types, suggesting that different forms of collusion manifest differently in activation space. We also find preliminary evidence that this signal is localised at the token level, with the colluding agent's activations spiking specifically when processing the encoded parts of their partner's message. This work takes a step toward multi-agent interpretability: extending white-box inspection from single models to multi-agent contexts, where detection requires aggregating signals across agents. These results suggest that model internals provide a complementary signal to text-level monitoring for detecting multi-agent collusion, particularly for organisations with access to model activations. Code and data are available at https://github.com/aaronrose227/narcbench.

Most defects in mobile applications are visually observable on the device screen. To track these defects, users, testers, and developers must manually submit bug reports, especially in the absence of crashes. However, these reports are frequently ambiguous or inaccurate, often omitting essential components such as the Observed Behavior (OB), Expected Behavior (EB), or Steps to Reproduce (S2Rs). Low-quality reports hinder developers' ability to understand and reproduce defects, delaying resolution and leading to incorrect or unresolvable fixes. In this paper, we posit that providing specific app-related information (e.g., GUI interactions or specific screens where bugs appear) to LLMs as key points of context can assist in automatically generating clear, detailed, and accurate OB, EB, and S2Rs. We built and evaluated a novel approach, BugScribe, that generates bug reports in this way. To support the evaluation, we introduce a unified quality framework that defines correctness and completeness dimensions for OB, EB, and S2Rs. Using 48 bug reports from 26 Android apps, we show that BugScribe produces higher-quality and more accurate components than the original reports and outperforms recent LLM-based baselines. We envision that BugScribe can serve as a practical assistant for testers and developers by enhancing incomplete bug reports with reliable and accurate OB, EB, and S2Rs, thereby streamlining bug resolution and improving mobile app quality.

As Large Language Models (LLMs) for code increasingly utilize massive, often non-permissively licensed datasets, evaluating data contamination through Membership Inference Attacks (MIAs) has become critical. We propose SERSEM (Selective Entropy-Weighted Scoring for Membership Inference), a novel white-box attack framework that suppresses uninformative syntactical boilerplate to amplify specific memorization signals. SERSEM utilizes a dual-signal methodology: first, a continuous character-level weight mask is derived through static Abstract Syntax Tree (AST) analysis, spellchecking-based multilingual logic detection, and offline linting. Second, these heuristic weights are used to pool internal transformer activations and calibrate token-level Z-scores from the output logits. Evaluated on a 25,000-sample balanced dataset, SERSEM achieves a global AUC-ROC of 0.7913 on the StarCoder2-3B model and 0.7867 on the StarCoder2-7B model, consistently outperforming the implemented probability-based baselines Loss, Min-K% Prob, and PAC. Our findings demonstrate that focusing on human-centric coding anomalies provides a significantly more robust indicator of verbatim memorization than sequence-level probability averages.

Reinforcement learning has shown strong performance in robotic manipulation, but learned policies often degrade in performance when test conditions differ from the training distribution. This limitation is especially important in contact-rich tasks such as pushing and pick-and-place, where changes in goals, contact conditions, or robot dynamics can drive the system out-of-distribution at inference time. In this paper, we investigate a hybrid controller that combines reinforcement learning with bounded extremum seeking to improve robustness under such conditions. In the proposed approach, deep deterministic policy gradient (DDPG) policies are trained under standard conditions on the robotic pushing and pick-and-place tasks, and are then combined with bounded ES during deployment. The RL policy provides fast manipulation behavior, while bounded ES ensures robustness of the overall controller to time variations when operating conditions depart from those seen during training. The resulting controller is evaluated under several out-of-distribution settings, including time-varying goals and spatially varying friction patches.

Due to the large footprint of pixels in remote sensing imagery, hyperspectral unmixing (HU) has become an important and necessary procedure in hyperspectral image analysis. Traditional HU methods rely on a prior spectral mixing model, especially for nonlinear mixtures, which has largely limited the performance and generalization capacity of the unmixing approach. In this paper, we address the challenging problem of hyperspectral nonlinear unmixing (HNU) without explicit knowledge of the mixing model. Inspired by the principle of generative models, where images of the same distribution can be generated as that of the training images without knowing the exact probability distribution function of the image, we develop an invertible mixing-unmixing process via a bi-directional GAN framework, constrained by both the cycle consistency and the linkage between linear and nonlinear mixtures. The combination of cycle consistency and linear linkage provides powerful constraints without requiring an explicit mixing model. We refer to the proposed approach as the linearly-constrained CycleGAN unmixing net, or LCGU net. Experimental results indicate that the proposed LCGU net exhibits stable and competitive performance across different datasets compared with other state-of-the-art model-based HNU methods.

Individuals engaging in online communication frequently express personal opinions with informal styles (e.g., memes and emojis). While Language Models (LMs) with informal communications have been widely discussed, a unique and emphatic style, the Repetitive Lengthening Form (RLF), has been overlooked for years. In this paper, we explore answers to two research questions: 1) Is RLF important for sentiment analysis (SA)? 2) Can LMs understand RLF? Inspired by previous linguistic research, we curate \textbf{Lengthening}, the first multi-domain dataset with 850k samples focused on RLF for SA. Moreover, we introduce \textbf{Exp}lainable \textbf{Instruct}ion Tuning (\textbf{ExpInstruct}), a two-stage instruction tuning framework aimed to improve both performance and explainability of LLMs for RLF. We further propose a novel unified approach to quantify LMs' understanding of informal expressions. We show that RLF sentences are expressive expressions and can serve as signatures of document-level sentiment. Additionally, RLF has potential value for online content analysis. Our results show that fine-tuned Pre-trained Language Models (PLMs) can surpass zero-shot GPT-4 in performance but not in explanation for RLF. Finally, we show ExpInstruct can improve the open-sourced LLMs to match zero-shot GPT-4 in performance and explainability for RLF with limited samples. Code and sample data are available at https://github.com/Tom-Owl/OverlookedRLF

File-level defect prediction models traditionally rely on product and process metrics. While process metrics effectively complement product metrics, they often overlook commit size the number of files changed per commit despite its strong association with software quality. Network centrality measures on dependency graphs have also proven to be valuable product level indicators. Motivated by this, we first redefine process metrics as commit size aware process metric vectors, transforming conventional scalar measures into 100 dimensional profiles that capture the distribution of changes across commit size strata. We then model change history as a hyper co change graph, where hyperedges naturally encode commit-size semantics. Vector centralities computed on these hypergraphs quantify size-aware node importance for source files. Experiments on nine long-lived Apache projects using five popular classifiers show that replacing scalar process metrics with the proposed commit size aware vectors, alongside product metrics, consistently improves predictive performance. These findings establish that commit size aware process metrics and hypergraph based vector centralities capture higher-order change semantics, leading to more discriminative, better calibrated, and statistically superior defect prediction models.

As sports training becomes more data-driven, traditional dart coaching based mainly on experience and visual observation is increasingly inadequate for high-precision, goal-oriented movements. Although prior studies have highlighted the importance of release parameters, joint motion, and coordination in dart throwing, most quantitative methods still focus on local variables, single-release metrics, or static template matching. These approaches offer limited support for personalized training and often overlook useful movement variability. This paper presents a data-driven dart training assistance system. The system creates a closed-loop framework spanning motion capture, feature modeling, and personalized feedback. Dart-throwing data were collected in markerless conditions using a Kinect 2.0 depth sensor and an optical camera. Eighteen kinematic features were extracted from four biomechanical dimensions: three-link coordination, release velocity, multi-joint angular configuration, and postural stability. Two modules were developed: a personalized optimal throwing trajectory model that combines historical high-quality samples with the minimum jerk criterion, and a motion deviation diagnosis and recommendation model based on z-scores and hierarchical logic. A total of 2,396 throwing samples from professional and non-professional athletes were collected. Results show that the system generates smooth personalized reference trajectories consistent with natural human movement. Case studies indicate that it can detect poor trunk stability, abnormal elbow displacement, and imbalanced velocity control, then provide targeted recommendations. The framework shifts dart evaluation from deviation from a uniform standard to deviation from an individual's optimal control range, improving personalization and interpretability for darts training and other high-precision target sports.

This paper introduces the first systematic evaluation framework for quantifying the quality and risks of papers written by modern coding agents. While AI-driven paper writing has become a growing concern, rigorous evaluation of the quality and potential risks of AI-written papers remains limited, and a unified understanding of their reliability is still lacking. We introduce Paper Reconstruction Evaluation (PaperRecon), an evaluation framework in which an overview (overview.md) is created from an existing paper, after which an agent generates a full paper based on the overview and minimal additional resources, and the result is subsequently compared against the original paper. PaperRecon disentangles the evaluation of the AI-written papers into two orthogonal dimensions, Presentation and Hallucination, where Presentation is evaluated using a rubric and Hallucination is assessed via agentic evaluation grounded in the original paper source. For evaluation, we introduce PaperWrite-Bench, a benchmark of 51 papers from top-tier venues across diverse domains published after 2025. Our experiments reveal a clear trade-off: while both ClaudeCode and Codex improve with model advances, ClaudeCode achieves higher presentation quality at the cost of more than 10 hallucinations per paper on average, whereas Codex produces fewer hallucinations but lower presentation quality. This work takes a first step toward establishing evaluation frameworks for AI-driven paper writing and improving the understanding of its risks within the research community.

Leveraging the rich semantic features of vision-language models (VLMs) like CLIP for monocular depth estimation tasks is a promising direction, yet often requires extensive fine-tuning or lacks geometric precision. We present a parameter-efficient framework, named MoA-DepthCLIP, that adapts pretrained CLIP representations for monocular depth estimation with minimal supervision. Our method integrates a lightweight Mixture-of-Adapters (MoA) module into the pretrained Vision Transformer (ViT-B/32) backbone combined with selective fine-tuning of the final layers. This design enables spatially-aware adaptation, guided by a global semantic context vector and a hybrid prediction architecture that synergizes depth bin classification with direct regression. To enhance structural accuracy, we employ a composite loss function that enforces geometric constraints. On the NYU Depth V2 benchmark, MoA-DepthCLIP achieves competitive results, significantly outperforming the DepthCLIP baseline by improving the $δ_1$ accuracy from 0.390 to 0.745 and reducing the RMSE from 1.176 to 0.520. These results are achieved while requiring substantially few trainable parameters, demonstrating that lightweight, prompt-guided MoA is a highly effective strategy for transferring VLM knowledge to fine-grained monocular depth estimation tasks.

Dependency networks (Heckerman et al., 2000) provide a flexible framework for modeling complex systems with many variables by combining independently learned local conditional distributions through pseudo-Gibbs sampling. Despite their computational advantages over Bayesian and Markov networks, the theoretical foundations of dependency networks remain incomplete, primarily because their model distributions -- defined as stationary distributions of pseudo-Gibbs sampling -- lack closed-form expressions. This paper develops an information-geometric analysis of pseudo-Gibbs sampling, interpreting each sampling step as an m-projection onto a full conditional manifold. Building on this interpretation, we introduce the full conditional divergence and derive an upper bound that characterizes the location of the stationary distribution in the space of probability distributions. We then reformulate both structure and parameter learning as optimization problems that decompose into independent subproblems for each node, and prove that the learned model distribution converges to the true underlying distribution as the number of training samples grows to infinity. Experiments confirm that the proposed upper bound is tight in practice.

As generative AI systems are integrated into educational settings, students often encounter AI-generated output while working through learning tasks, either by requesting help or through integrated tools. Trust in AI can influence how students interpret and use that output, including whether they evaluate it critically or exhibit overreliance. We investigate how students' trust relates to their appropriate reliance on an AI assistant during programming problem-solving tasks, and whether this relationship differs by learner characteristics. With 432 undergraduate participants, students' completed Python output-prediction problems while receiving recommendations and explanations from an AI chatbot, including accurate and intentionally misleading suggestions. We operationalize reliance behaviorally as the extent to which students' responses reflected appropriate use of the AI assistant's suggestions, accepting them when they were correct and rejecting them when they were incorrect. Pre- and post-task surveys assessed trust in the assistant, AI literacy, need for cognition, programming self-efficacy, and programming literacy. Results showed a non-linear relationship in which higher trust was associated with lower appropriate reliance, suggesting weaker discrimination between correct and incorrect recommendations. This relationship was significantly moderated by students' AI literacy and need for cognition. These findings highlight the need for future work on instructional and system supports that encourage more reflective evaluation of AI assistance during problem-solving.

Large language model (LLM) systems are increasingly used to support high-stakes decision-making, but they typically perform worse when the available evidence is internally inconsistent. Such a scenario exists in real-world healthcare settings, with patient-reported symptoms contradicting medical signs. To study this problem, we introduce MIMIC-DOS, a dataset for short-horizon organ dysfunction worsening prediction in the intensive care unit (ICU) setting. We derive this dataset from the widely recognized MIMIC-IV, a publicly available electronic health record dataset, and construct it exclusively from cases in which discordance between signs and symptoms exists. This setting poses a substantial challenge for existing LLM-based approaches, with single-pass LLMs and agentic pipelines often struggling to reconcile such conflicting signals. To address this problem, we propose CARE: a multi-stage privacy-compliant agentic reasoning framework in which a remote LLM provides guidance by generating structured categories and transitions without accessing sensitive patient data, while a local LLM uses these categories and transitions to support evidence acquisition and final decision-making. Empirically, CARE achieves stronger performance across all key metrics compared to multiple baseline settings, showing that CARE can more robustly handle conflicting clinical evidence while preserving privacy.

Software engineering students often struggle to appreciate empirical methods and hypothesis-driven inquiry, especially when taught in theoretical terms. This experience report explores whether grounding empirical learning in hype-driven technologies can make these concepts more accessible and engaging. We conducted a one-semester seminar framed around the currently popular topic of AI coding assistants, which attracted unusually high student interest. The course combined hands-on sessions using AI coding assistants with small, student-designed empirical studies. Classroom observations and survey responses suggest that the hype topic sparked curiosity and critical thinking. Students engaged with the AI coding assistants while questioning their limitations -- developing the kind of empirical thinking needed to assess claims about emerging technologies. Key lessons: (1) Hype-driven topics can lower barriers to abstract concepts like empirical research; (2) authentic hands-on development tasks combined with ownership of inquiry foster critical engagement; and (3) a single seminar can effectively teach both technical and research skills.

Evaluating the ethical robustness of large language models (LLMs) deployed in software systems remains challenging, particularly under sustained adversarial user interaction. Existing safety benchmarks typically rely on single-round evaluations and aggregate metrics, such as toxicity scores and refusal rates, which offer limited visibility into behavioral instability that may arise during realistic multi-turn interactions. As a result, rare but high-impact ethical failures and progressive degradation effects may remain undetected prior to deployment. This paper introduces Adversarial Moral Stress Testing (AMST), a stress-based evaluation framework for assessing ethical robustness under adversarial multi-round interactions. AMST applies structured stress transformations to prompts and evaluates model behavior through distribution-aware robustness metrics that capture variance, tail risk, and temporal behavioral drift across interaction rounds. We evaluate AMST on several state-of-the-art LLMs, including LLaMA-3-8B, GPT-4o, and DeepSeek-v3, using a large set of adversarial scenarios generated under controlled stress conditions. The results demonstrate substantial differences in robustness profiles across models and expose degradation patterns that are not observable under conventional single-round evaluation protocols. In particular, robustness has been shown to depend on distributional stability and tail behavior rather than on average performance alone. Additionally, AMST provides a scalable and model-agnostic stress-testing methodology that enables robustness-aware evaluation and monitoring of LLM-enabled software systems operating in adversarial environments.

Inverse design of optical multilayer stacks seeks to infer layer materials, thicknesses, and ordering from a desired target spectrum. It is a long-standing challenge due to the large design space and non-unique solutions. We introduce \texttt{OptoLlama}, a masked diffusion language model for inverse thin-film design from optical spectra. Representing multilayer stacks as sequences of material-thickness tokens, \texttt{OptoLlama} conditions generation on reflectance, absorptance, and transmittance spectra and learns a probabilistic mapping from optical response to structure. Evaluated on a representative test set of 3,000 targets, \texttt{OptoLlama} reduces the mean absolute spectral error by 2.9-fold relative to a nearest-neighbor template baseline and by 3.45-fold relative to the state-of-the-art data-driven baseline, called \texttt{OptoGPT}. Case studies on designed and expert-defined targets show that the model reproduces characteristic spectral features and recovers physically meaningful stack motifs, including distributed Bragg reflectors. These results establish diffusion-based sequence modeling as a powerful framework for inverse photonic design.

Stochastic Multi-Objective Optimization (SMOO) is critical for decision-making trading off multiple potentially conflicting objectives in uncertain environments. SMOO aims at identifying the Pareto frontier, which contains all mutually non-dominating decisions. The problem is highly intractable due to the embedded probabilistic inference, such as computing the marginal, posterior probabilities, or expectations. Existing methods, such as scalarization, sample average approximation, and evolutionary algorithms, either offer arbitrarily loose approximations or may incur prohibitive computational costs. We propose XOR-SMOO, a novel algorithm that with probability $1-δ$, obtains $γ$-approximate Pareto frontiers ($γ>1$) for SMOO by querying an SAT oracle poly-log times in $γ$ and $δ$. A $γ$-approximate Pareto frontier is only below the true frontier by a fixed, multiplicative factor $γ$. Thus, XOR-SMOO solves highly intractable SMOO problems (\#P-hard) with only queries to SAT oracles while obtaining tight, constant factor approximation guarantees. Experiments on real-world road network strengthening and supply chain design problems demonstrate that XOR-SMOO outperforms several baselines in identifying Pareto frontiers that have higher objective values, better coverage of the optimal solutions, and the solutions found are more evenly distributed. Overall, XOR-SMOO significantly enhanced the practicality and reliability of SMOO solvers.

Large language models (LLMs) exhibit strong in-context learning capabilities, but how they track and retrieve information from context remains underexplored. Drawing on the free recall paradigm in cognitive science (where participants recall list items in any order), we show that several open-source LLMs consistently display a serial-recall-like pattern, assigning peak probability to tokens that immediately follow a repeated token in the input sequence. Through systematic ablation experiments, we show that induction heads, specialized attention heads that attend to the token following a previous occurrence of the current token, play an important role in this phenomenon. Removing heads with a high induction score substantially reduces the +1 lag bias, whereas ablating random heads does not reproduce the same reduction. We also show that removing heads with high induction scores impairs the performance of models prompted to do serial recall using few-shot learning to a larger extent than removing random heads. Our findings highlight a mechanistically specific connection between induction heads and temporal context processing in transformers, suggesting that these heads are especially important for ordered retrieval and serial-recall-like behavior during in-context learning.

Partial audio deepfakes, where synthesized segments are spliced into genuine recordings, are particularly deceptive because most of the audio remains authentic. Existing detectors are supervised: they require frame-level annotations, overfit to specific synthesis pipelines, and must be retrained as new generative models emerge. We argue that this supervision is unnecessary. We hypothesize that speech foundation models implicitly encode a forensic signal: genuine speech forms smooth, slowly varying embedding trajectories, while splice boundaries introduce abrupt disruptions in frame-level transitions. Building on this, we propose TRACE (Training-free Representation-based Audio Countermeasure via Embedding dynamics), a training-free framework that detects partial audio deepfakes by analyzing the first-order dynamics of frozen speech foundation model representations without any training, labeled data, or architectural modification. We evaluate TRACE on four benchmarks that span two languages using six speech foundation models. In PartialSpoof, TRACE achieves 8.08% EER, competitive with fine-tuned supervised baselines. In LlamaPartialSpoof, the most challenging benchmark featuring LLM-driven commercial synthesis, TRACE surpasses a supervised baseline outright (24.12% vs. 24.49% EER) without any target-domain data. These results show that temporal dynamics in speech foundation models provide an effective, generalize signal for training-free audio forensics.

3D semantic occupancy prediction is central to autonomous driving, yet current methods are vulnerable to long-tailed class bias and out-of-distribution (OOD) inputs, often overconfidently assigning anomalies to rare classes. We present ProOOD, a lightweight, plug-and-play method that couples prototype-guided refinement with training-free OOD scoring. ProOOD comprises (i) prototype-guided semantic imputation that fills occluded regions with class-consistent features, (ii) prototype-guided tail mining that strengthens rare-class representations to curb OOD absorption, and (iii) EchoOOD, which fuses local logit coherence with local and global prototype matching to produce reliable voxel-level OOD scores. Extensive experiments on five datasets demonstrate that ProOOD achieves state-of-the-art performance on both in-distribution 3D occupancy prediction and OOD detection. On SemanticKITTI, it surpasses baselines by +3.57% mIoU overall and +24.80% tail-class mIoU; on VAA-KITTI, it improves AuPRCr by +19.34 points, with consistent gains across benchmarks. These improvements yield more calibrated occupancy estimates and more reliable OOD detection in safety-critical urban driving. The source code is publicly available at https://github.com/7uHeng/ProOOD.

There is a growing need for cybersecurity professionals with practical knowledge and experience to meet societal needs and comply with new standards and regulations. At the same time, the advances in software technology and artificial intelligence point towards a future where software agents will play an important role in protecting the computer systems that are critical for society to function. The training and development of both humans and software agents requires the design and execution of cybersecurity exercises that differ in properties such as size, scope, objectives, difficultly, etc. Cybersecurity scenarios are critical for the operation of cybersecurity exercises as they describe the scope, context, operational environment and storyline of each exercise. In this work, we present an approach to automatically generate cybersecurity scenarios that model enterprise IT systems. Our approach is able to generate a large number of scenarios that differ in multiple criteria including size, scope, difficulty, complexity and diversity. We further release as open source: a simulation and a virtualization environment that can run cybersecurity exercises based on the generated scenarios and a dataset containing 100000 sample scenarios.

We test whether authors have characteristic "fingerprints" in the information-theoretic novelty curves of their published works. Working with two corpora -- Books3 (52,796 books, 759 qualifying authors) and PG-19 (28,439 books, 1,821 qualifying authors) -- we find that authorial voice leaves measurable traces in how novelty unfolds across a text. The signal is multi-scale: at book level, scalar dynamics (mean novelty, speed, volume, circuitousness) identify 43% of authors significantly above chance; at chapter level, SAX motif patterns in sliding windows achieve 30x-above-chance attribution, far exceeding the scalar features that dominate at book level. These signals are complementary, not redundant. We show that the fingerprint is partly confounded with genre but persists within-genre for approximately one-quarter of authors. Classical authors (Twain, Austen, Kipling) show fingerprints comparable in strength to modern authors, suggesting the phenomenon is not an artifact of contemporary publishing conventions.

Computational reproducibility is fundamental to trustworthy science, yet remains difficult to achieve in practice across various research workflows, including Jupyter notebooks published alongside scholarly articles. Environment drift, undocumented dependencies and implicit execution assumptions frequently prevent independent re-execution of published research. Despite existing reproducibility guidelines, scalable and systematic infrastructure for automated assessment remains limited. We present an automated, web-oriented reproducibility engineering pipeline that reconstructs and evaluates repository-level execution environments for scholarly notebooks. The system performs dependency inference, automated container generation, and isolated execution to approximate the notebook's original computational context. We evaluate the approach on 443 notebooks from 116 GitHub repositories referenced by publications in PubMed Central. Execution outcomes are classified into four categories: resolved environment failures, persistent logic or data errors, reproducibility drift, and container-induced regressions. Our results show that containerization resolves 66.7% of prior dependency-related failures and substantially improves execution robustness. However, a significant reproducibility gap remains: 53.7% of notebooks exhibit low output fidelity, largely due to persistent runtime failures and stochastic non-determinism. These findings indicate that standardized containerization is essential for computational stability but insufficient for full bit-wise reproducibility. The framework offers a scalable solution for researchers, editors, and archivists seeking systematic, automated assessment of computational artifacts.

"Vibe coding," in which developers delegate code generation to AI assistants and accept the output with little manual review, has gained rapid adoption in production settings. On March 31, 2026, Anthropic's Claude Code CLI shipped a 59.8 MB source map file in its npm package, exposing roughly 512,000 lines of proprietary TypeScript. The tool had itself been largely vibe-coded, and the leak traced to a misconfigured packaging rule rather than a logic bug. Existing static-analysis and secret-scanning tools did not cover this failure mode, pointing to a gap between the vulnerabilities AI tends to introduce and the vulnerabilities current tooling is built to find. We present VibeGuard, a pre-publish security gate that targets five such blind spots: artifact hygiene, packaging-configuration drift, source-map exposure, hardcoded secrets, and supply-chain risk. In controlled experiments on eight synthetic projects (seven vulnerable, one clean control), VibeGuard achieved 100% recall, 89.47% precision (F1 = 94.44%), and correct pass/fail gate decisions on all eight projects across three policy levels. We discuss how these results inform a defense-in-depth workflow for teams that rely on AI code generation.

Next-generation (NextG) cellular networks are designed to support emerging applications with diverse data rate and latency requirements, such as immersive multimedia services and large-scale Internet of Things deployments. A key enabling mechanism is radio access network (RAN) slicing, which dynamically partitions radio resources into virtual resource blocks to efficiently serve heterogeneous traffic classes, including enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC). In this paper, we study the impact of adversarial attacks on AI-driven RAN slicing decisions, where a budget-constrained adversary selectively jams slice transmissions to bias deep reinforcement learning (DRL)-based resource allocation, and quantify the resulting service level agreement (SLA) violations and post-attack recovery behavior. Our results indicate that budget-constrained adversarial jamming can induce severe and slice-dependent steady-state SLA violations. Moreover, the DRL agent's reward converges toward the clean baseline only after a non-negligible recovery period.

System Instructions in Large Language Models (LLMs) are commonly used to enforce safety policies, define agent behavior, and protect sensitive operational context in agentic AI applications. These instructions may contain sensitive information such as API credentials, internal policies, and privileged workflow definitions, making system instruction leakage a critical security risk highlighted in the OWASP Top 10 for LLM Applications. Without incurring the overhead costs of reasoning models, many LLM applications rely on refusal-based instructions that block direct requests for system instructions, implicitly assuming that prohibited information can only be extracted through explicit queries. We introduce an automated evaluation framework that tests whether system instructions remain confidential when extraction requests are re-framed as encoding or structured output tasks. Across four common models and 46 verified system instructions, we observe high attack success rates (> 0.7) for structured serialization where models refuse direct extraction requests but disclose protected content in the requested serialization formats. We further demonstrate a mitigation strategy based on one-shot instruction reshaping using a Chain-of-Thought reasoning model, indicating that even subtle changes in wording and structure of system instructions can significantly reduce attack success rate without requiring model retraining.

Popularity bias is a pervasive problem in recommender systems, where recommendations disproportionately favor popular items. This not only results in "rich-get-richer" dynamics and a homogenization of visible content, but can also lead to misalignment of recommendations with individual users' preferences for popular or niche content. This work studies popularity bias through the lens of user-recommender alignment. To this end, we introduce Popularity Quantile Calibration, a measurement framework that quantifies misalignment between a user's historical popularity preference and the popularity of their recommendations. Building on this notion of popularity alignment, we propose SPREE, an inference-time mitigation method for sequential recommenders based on activation steering. SPREE identifies a popularity direction in representation space and adaptively steers model activations based on an estimate of each user's personal popularity bias, allowing both the direction and magnitude of steering to vary across users. Unlike global debiasing approaches, SPREE explicitly targets alignment rather than uniformly reducing popularity. Experiments across multiple datasets show that SPREE consistently improves user-level popularity alignment while preserving recommendation quality.

Multi-LLM revision pipelines, in which a second model reviews and improves a draft produced by a first, are widely assumed to derive their gains from genuine error correction. We question this assumption with a controlled decomposition experiment that uses four matched conditions to separate second-pass gains into three additive components: re-solving, scaffold, and content. We evaluate this design across two model pairs on three benchmarks spanning knowledge-intensive MCQ and competitive programming. Our results show that the gains of multi-LLM revision are not monolithic, but depend on task structure, draft quality, and the type of draft information. On MCQ tasks, where the answer space is constrained and drafts provide little structural guidance, most gains are consistent with stronger-model re-solving, and directly routing queries to the stronger model can be more effective than revising a weak draft. On code generation tasks, however, two-stage prompting remains useful because even semantically null drafts can provide substantial structural scaffolding, while weak draft content can be harmful. Finally, role-reversed experiments show that strong drafts clearly benefit weak reviewers. Ultimately, our findings demonstrate that the utility of multi-LLM revision is dynamically bottlenecked by task structure and draft quality, necessitating more targeted pipeline designs rather than blanket revision strategies.

The paradigm of scaling Large Language Models (LLMs) in both parameter size and test time has pushed the boundaries of AI capabilities, but at the cost of making the traditional generative evaluation paradigm prohibitively expensive, therefore making the latency of LLM's in-training downstream performance evaluation unbearable. However, simple metrics like training loss (perplexity) are not always correlated with downstream performance, as sometimes their trends diverge from the actual task outcomes. This dilemma calls for a method that is computationally efficient and sufficiently accurate in measuring model capabilities. To address this challenge, we introduce a new in-training evaluation paradigm that uses a lightweight probe for monitoring downstream performance. The probes take the internal representations of LLM checkpoints (during training) as input and directly predict the checkpoint's performance on downstream tasks measured by success probability (i.e., pass@1). We design several probe architectures, validating their effectiveness using the OLMo3-7B's checkpoints across a diverse set of downstream tasks. The probes can accurately predict a checkpoint's performance (with avg. AUROC$>$0.75), have decent generalizability across checkpoints (earlier predicts later), and reduce the computation latency from $\sim$1 hr (using conventional generative evaluation method) to $\sim$3 min. In sum, this work presents a practical and scalable in-training downstream evaluation paradigm, enabling a more agile, informed, and efficient LLM development process.

We study model-based learning of finite-window policies in tabular partially observable Markov decision processes (POMDPs). A common approach to learning under partial observability is to approximate unbounded history dependencies using finite action-observation windows. This induces a finite-state Markov decision process (MDP) over histories, referred to as the superstate MDP. Once a model of this superstate MDP is available, standard MDP algorithms can be used to compute optimal policies, motivating the need for sample-efficient model estimation. Estimating the superstate MDP model is challenging because trajectories are generated by interaction with the original POMDP, creating a mismatch between the sampling process and target model. We propose a model estimation procedure for tabular POMDPs and analyze its sample complexity. Our analysis exploits a connection between filter stability and concentration inequalities for weakly dependent random variables. As a result, we obtain tight sample complexity guarantees for estimating the superstate MDP model from a single trajectory. Combined with value iteration, this yields approximately optimal finite-window policies for the POMDP.

This paper introduces two transfer learning methodologies for estimating nonparametric Bayesian networks under scarce data. We propose two algorithms, a constraint-based structure learning method, called PC-stable-transfer learning (PCS-TL), and a score-based method, called hill climbing transfer learning (HC-TL). We also define particular metrics to tackle the negative transfer problem in each of them, a situation in which transfer learning has a negative impact on the model's performance. Then, for the parameters, we propose a log-linear pooling approach. For the evaluation, we learn kernel density estimation Bayesian networks, a type of nonparametric Bayesian network, and compare their transfer learning performance with the models alone. To do so, we sample data from small, medium and large-sized synthetic networks and datasets from the UCI Machine Learning repository. Then, we add noise and modifications to these datasets to test their ability to avoid negative transfer. To conclude, we perform a Friedman test with a Bergmann-Hommel post-hoc analysis to show statistical proof of the enhanced experimental behavior of our methods. Thus, PCS-TL and HC-TL demonstrate to be reliable algorithms for improving the learning performance of a nonparametric Bayesian network with scarce data, which in real industrial environments implies a reduction in the required time to deploy the network.

While large language model-based multi-agent systems have shown strong potential for complex reasoning, how to effectively organize multiple agents remains an open question. In this paper, we introduce OrgAgent, a company-style hierarchical multi-agent framework that separates collaboration into governance, execution, and compliance layers. OrgAgent decomposes multi-agent reasoning into three layers: a governance layer for planning and resource allocation, an execution layer for task solving and review, and a compliance layer for final answer control. By evaluating the framework across reasoning tasks, LLMs, execution modes, and execution policies, we find that multi-agent systems organized in a company-style hierarchy generally outperform other organizational structures. Besides, hierarchical coordination also reduces token consumption relative to flat collaboration in most settings. For example, for GPT-OSS-120B, the hierarchical setting improves performance over flat multi-agent system by 102.73% while reducing token usage by 74.52% on SQuAD 2.0. Further analysis shows that hierarchy helps most when tasks benefit from stable skill assignment, controlled information flow, and layered verification. Overall, our findings highlight organizational structure as an important factor in multi-agent reasoning, shaping not only effectiveness and cost, but also coordination behavior.

AI agents increasingly operate over extended time horizons, yet their ability to retain, organize, and recall multimodal experiences remains a critical bottleneck. Building effective lifelong memory requires navigating a vast design space spanning architecture, retrieval strategies, prompt engineering, and data pipelines; this space is too large and interconnected for manual exploration or traditional AutoML to explore effectively. We deploy an autonomous research pipeline to discover Omni-SimpleMem, a unified multimodal memory framework for lifelong AI agents. Starting from a naïve baseline (F1=0.117 on LoCoMo), the pipeline autonomously executes ${\sim}50$ experiments across two benchmarks, diagnosing failure modes, proposing architectural modifications, and repairing data pipeline bugs, all without human intervention in the inner loop. The resulting system achieves state-of-the-art on both benchmarks, improving F1 by +411% on LoCoMo (0.117$\to$0.598) and +214% on Mem-Gallery (0.254$\to$0.797) relative to the initial configurations. Critically, the most impactful discoveries are not hyperparameter adjustments: bug fixes (+175%), architectural changes (+44%), and prompt engineering (+188% on specific categories) each individually exceed the cumulative contribution of all hyperparameter tuning, demonstrating capabilities fundamentally beyond the reach of traditional AutoML. We provide a taxonomy of six discovery types and identify four properties that make multimodal memory particularly suited for autoresearch, offering guidance for applying autonomous research pipelines to other AI system domains. Code is available at this https://github.com/aiming-lab/SimpleMem.

Multimodal Large Language Models (MLLMs) have shown strong performance on video question answering, but their application to long-form videos is constrained by limited context length and computational cost, making keyframe sampling essential. Existing approaches typically rely on semantic relevance or reinforcement learning, which either fail to capture evidential clues or suffer from inefficient combinatorial optimization. In this work, we propose an evidence-driven keyframe sampling framework grounded in information bottleneck theory. We formulate keyframe selection as maximizing the conditional mutual information between selected frames and the query, providing a principled objective that reflects each frame's contribution to answering the question. To make this objective tractable, we exploit its structure to derive a decomposed optimization that reduces subset selection to independent frame-level scoring. We further introduce a query-conditioned evidence scoring network trained with a contrastive objective to estimate evidential importance efficiently. Experiments on long-form video understanding benchmarks show that our method consistently outperforms prior sampling strategies under strict token budgets, while significantly improving training efficiency.

We introduce EgoSim, a closed-loop egocentric world simulator that generates spatially consistent interaction videos and persistently updates the underlying 3D scene state for continuous simulation. Existing egocentric simulators either lack explicit 3D grounding, causing structural drift under viewpoint changes, or treat the scene as static, failing to update world states across multi-stage interactions. EgoSim addresses both limitations by modeling 3D scenes as updatable world states. We generate embodiment interactions via a Geometry-action-aware Observation Simulation model, with spatial consistency from an Interaction-aware State Updating module. To overcome the critical data bottleneck posed by the difficulty in acquiring densely aligned scene-interaction training pairs, we design a scalable pipeline that extracts static point clouds, camera trajectories, and embodiment actions from in-the-wild large-scale monocular egocentric videos. We further introduce EgoCap, a capture system that enables low-cost real-world data collection with uncalibrated smartphones. Extensive experiments demonstrate that EgoSim significantly outperforms existing methods in terms of visual quality, spatial consistency, and generalization to complex scenes and in-the-wild dexterous interactions, while supporting cross-embodiment transfer to robotic manipulation. Codes and datasets will be open soon. The project page is at egosimulator.github.io.

Graph Neural Networks (GNNs) are widely used for learning on graph-structured data, but scaling GNN training to massive graphs remains challenging. To enable scalable distributed training, graphs are divided into smaller partitions that are distributed across multiple machines such that inter-machine communication is minimized and computational load is balanced. In practice, existing partitioning approaches face a fundamental trade-off between partitioning overhead and partitioning quality. We propose EmbedPart, an embedding-driven partitioning approach that achieves both speed and quality. Instead of operating directly on irregular graph structures, EmbedPart leverages node embeddings produced during the actual GNN training workload and clusters these dense embeddings to derive a partitioning. EmbedPart achieves more than 100x speedup over Metis while maintaining competitive partitioning quality and accelerating distributed GNN training. Moreover, EmbedPart naturally supports graph updates and fast repartitioning, and can be applied to graph reordering to improve data locality and accelerate single-machine GNN training. By shifting partitioning from irregular graph structures to dense embeddings, EmbedPart enables scalable and high-quality graph data optimization.

Reward factorization personalizes large language models (LLMs) by decomposing rewards into shared basis functions and user-specific weights. Yet, existing methods estimate user weights from scarce data in isolation and as deterministic points, leading to inaccurate and unreliable inference. We introduce Variational Reward Factorization (VRF), an uncertainty-aware framework that represents each user's preferences as a variational distribution in a shared preference space. VRF infers user distributions via a variational encoder, derives weights through Wasserstein distance matching with shared probabilistic bases, and downweights uncertain estimates through a variance-attenuated loss. On three benchmarks, VRF outperforms all baselines across seen and unseen users, few-shot scenarios, and varying uncertainty levels, with gains extending to downstream alignment.

YouTube Shorts have become central to news consumption on the platform, yet research on how geopolitical events are represented in this format remains limited. To address this gap, we present a multimodal pipeline that combines automatic transcription, aspect-based sentiment analysis (ABSA), and semantic scene classification. The pipeline is first assessed for feasibility and then applied to analyze short-form coverage of the Israel-Hamas war by state-funded outlets. Using over 2,300 conflict-related Shorts and more than 94,000 visual frames, we systematically examine war reporting across major international broadcasters. Our findings reveal that the sentiment expressed in transcripts regarding specific aspects differs across outlets and over time, whereas scene-type classifications reflect visual cues consistent with real-world events. Notably, smaller domain-adapted models outperform large transformers and even LLMs for sentiment analysis, underscoring the value of resource-efficient approaches for humanities research. The pipeline serves as a template for other short-form platforms, such as TikTok and Instagram, and demonstrates how multimodal methods, combined with qualitative interpretation, can characterize sentiment patterns and visual cues in algorithmically driven video environments.

The direct imaging of potentially habitable exoplanets is one prime science case for high-contrast imaging instruments on extremely large telescopes. Most such exoplanets orbit close to their host stars, where their observation is limited by fast-moving atmospheric speckles and quasi-static non-common-path aberrations (NCPA). Conventional NCPA correction methods often use mechanical mirror probes, which compromise performance during operation. This work presents machine-learning-based NCPA control methods that automatically detect and correct both dynamic and static NCPA errors by leveraging sequential phase diversity. We extend previous work in reinforcement learning for AO to focal plane control. A new model-based RL algorithm, Policy Optimization for NCPAs (PO4NCPA), interprets the focal-plane image as input data and, through sequential phase diversity, determines phase corrections that optimize both non-coronagraphic and post-coronagraphic PSFs without prior system knowledge. Further, we demonstrate the effectiveness of this approach by numerically simulating static NCPA errors on a ground-based telescope and an infrared imager affected by water-vapor-induced seeing (dynamic NCPAs). Simulations show that PO4NCPA robustly compensates static and dynamic NCPAs. In static cases, it achieves near-optimal focal-plane light suppression with a coronagraph and near-optimal Strehl without one. With dynamics NCPA, it matches the performance of the modal least-squares reconstruction combined with a 1-step delay integrator in these metrics. The method remains effective for the ELT pupil, vector vortex coronagraph, and under photon and background noise. PO4NCPA is model-free and can be directly applied to standard imaging as well as to any coronagraph. Its sub-millisecond inference times and performance also make it suitable for real-time low-order correction of atmospheric turbulence beyond HCI.

We develop Structured-Knowledge-Informed Neural Networks (SKINNs), a unified estimation framework that embeds theoretical, simulated, previously learned, or cross-domain insights as differentiable constraints within flexible neural function approximation. SKINNs jointly estimate neural network parameters and economically meaningful structural parameters in a single optimization problem, enforcing theoretical consistency not only on observed data but over a broader input domain through collocation, and therefore nesting approaches such as functional GMM, Bayesian updating, transfer learning, PINNs, and surrogate modeling. SKINNs define a class of M-estimators that are consistent and asymptotically normal with root-N convergence, sandwich covariance, and recovery of pseudo-true parameters under misspecification. We establish identification of structural parameters under joint flexibility, derive generalization and target-risk bounds under distributional shift in a convex proxy, and provide a restricted-optimal characterization of the weighting parameter that governs the bias-variance tradeoff. In an illustrative financial application to option pricing, SKINNs improve out-of-sample valuation and hedging performance, particularly at longer horizons and during high-volatility regimes, while recovering economically interpretable structural parameters with improved stability relative to conventional calibration. More broadly, SKINNs provide a general econometric framework for combining model-based reasoning with high-dimensional, data-driven estimation.

We study whether phone-use agents respect privacy while completing benign mobile tasks. This question has remained hard to answer because privacy-compliant behavior is not operationalized for phone-use agents, and ordinary apps do not reveal exactly what data agents type into which form entries during execution. To make this question measurable, we introduce MyPhoneBench, a verifiable evaluation framework for privacy behavior in mobile agents. We operationalize privacy-respecting phone use as permissioned access, minimal disclosure, and user-controlled memory through a minimal privacy contract, iMy, and pair it with instrumented mock apps plus rule-based auditing that make unnecessary permission requests, deceptive re-disclosure, and unnecessary form filling observable and reproducible. Across five frontier models on 10 mobile apps and 300 tasks, we find that task success, privacy-compliant task completion, and later-session use of saved preferences are distinct capabilities, and no single model dominates all three. Evaluating success and privacy jointly reshuffles the model ordering relative to either metric alone. The most persistent failure mode across models is simple data minimization: agents still fill optional personal entries that the task does not require. These results show that privacy failures arise from over-helpful execution of benign tasks, and that success-only evaluation overestimates the deployment readiness of current phone-use agents. All code, mock apps, and agent trajectories are publicly available at~ https://github.com/FreedomIntelligence/MyPhoneBench.

Current aligned language models exhibit a dual failure mode we term the Evasive Servant: they sycophantically validate flawed user beliefs while deflecting responsibility with boilerplate disclaimers. We propose the Dignified Peer framework, which counters servility with anti-sycophancy and trustworthiness, and mitigates evasiveness through empathy and creativity. Realizing this agent requires overcoming significant challenges in data supervision, objective collapse, and evaluation bias. We address these issues by introducing the PersonaKnob dataset which features a compositional partial order structure of multiple persona preference. This data is utilized alongside a tolerant constrained Lagrangian DPO algorithm that dynamically balances all persona dimensions to prevent behavioral collapse. Additionally, we employ a psychometrically calibrated Item Response Theory evaluation protocol to disentangle latent model persona capability from confounders like judge biases. Extensive empirical studies demonstrate that our approach successfully build a LLM agent with both dignity and peer.

Reinforcement Learning (RL) has proven highly effective in addressing complex control and decision-making tasks. However, in most traditional RL algorithms, the policy is typically parameterized as a diagonal Gaussian distribution, which constrains the policy from capturing multimodal distributions, making it difficult to cover the full range of optimal solutions in multi-solution problems, and the return is reduced to a mean value, losing its multimodal nature and thus providing insufficient guidance for policy updates. In response to these problems, we propose a RL algorithm termed flow-based policy with distributional RL (FP-DRL). This algorithm models the policy using flow matching, which offers both computational efficiency and the capacity to fit complex distributions. Additionally, it employs a distributional RL approach to model and optimize the entire return distribution, thereby more effectively guiding multimodal policy updates and improving agent performance. Experimental trails on MuJoCo benchmarks demonstrate that the FP-DRL algorithm achieves state-of-the-art (SOTA) performance in most MuJoCo control tasks while exhibiting superior representation capability of the flow policy.

We show polylogarithmic mixing time bounds for the alternating-scan sampler for positively weighted restricted Boltzmann machines. This is done via analysing the same chain and the Glauber dynamics for ferromagnetic two-spin systems, where we obtain new mixing time bounds up to the critical thresholds.

Since the random language model was proposed by E. DeGiuli [Phys. Rev. Lett. 122, 128301], language models have been investigated intensively from the viewpoint of statistical mechanics. Recently, the existence of a Berezinskii--Kosterlitz--Thouless transition was numerically demonstrated in models with long-range interactions between symbols. In statistical mechanics, it has long been known that long-range interactions can induce phase transitions. Therefore, it has remained unclear whether phase transitions observed in language models originate from genuinely linguistic properties that are absent in conventional spin models. In this study, we construct a random language model with short-range interactions and numerically investigate its statistical properties. Our model belongs to the class of context-sensitive grammars in the Chomsky hierarchy and allows explicit reference to contexts. We find that a phase transition occurs even when the model refers only to contexts whose length remains constant with respect to the sentence length. This result indicates that finite-temperature phase transitions in language models are genuinely induced by the intrinsic nature of language, rather than by long-range interactions.

We introduce a differentially private manifold denoising framework that allows users to exploit sensitive reference datasets to correct noisy, non-private query points without compromising privacy. The method follows an iterative procedure that (i) privately estimates local means and tangent geometry using the reference data under calibrated sensitivity, (ii) projects query points along the privately estimated subspace toward the local mean via corrective steps at each iteration, and (iii) performs rigorous privacy accounting across iterations and queries using $(\varepsilon,δ)$-differential privacy (DP). Conceptually, this framework brings differential privacy to manifold methods, retaining sufficient geometric signal for downstream tasks such as embedding, clustering, and visualization, while providing formal DP guarantees for the reference data. Practically, the procedure is modular and scalable, separating DP-protected local geometry (means and tangents) from budgeted query-point updates, with a simple scheduler allocating privacy budget across iterations and queries. Under standard assumptions on manifold regularity, sampling density, and measurement noise, we establish high-probability utility guarantees showing that corrected queries converge toward the manifold at a non-asymptotic rate governed by sample size, noise level, bandwidth, and the privacy budget. Simulations and case studies demonstrate accurate signal recovery under moderate privacy budgets, illustrating clear utility-privacy trade-offs and providing a deployable DP component for manifold-based workflows in regulated environments without reengineering privacy systems.

Transformer-based NLP models remain vulnerable to adversarial perturbations, yet existing repair methods face a fundamental trade-off: gradient-based approaches offer flexibility but lack verifiability and often overfit; methods that do provide repair guarantees are restricted to the final layer or small networks, significantly limiting the parameter search space available for repair. We present WARP (Weight-Adjusted Repair with Provability), a constraint-based repair framework that extends repair beyond the last layer of Transformer models. WARP formulates repair as a convex quadratic program derived from a first-order linearization of the logit gap, enabling tractable optimization over a high-dimensional parameter space. Under the condition that the first-order approximation holds, this formulation induces three per-sample guarantees: (i) a positive margin constraint ensuring correct classification on repaired inputs, (ii) preservation constraints over a designated remain set, and (iii) a certified robustness radius derived from Lipschitz continuity. To ensure feasibility across varying model architectures, we introduce a sensitivity-based preprocessing step that conditions the optimization landscape accordingly. We further show that the iterative optimization procedure converges to solutions satisfying all repair constraints under mild assumptions. Empirical evaluation on encoder-only Transformers with varying layer architectures validates that these guarantees hold in practice while improving robustness to adversarial inputs. Our results demonstrate that guaranteed, generalizable Transformer repair is achievable through principled constraint-based optimization.

Continuous integration and delivery (CI/CD) pipelines are critical for sustaining the evolution of large software systems. In regulated industries with legacy technologies, however, pipelines themselves can become a source of technical debt. This paper presents an industrial case study of Bankdata, a cooperative IT provider for Danish banks, where a Jenkins-based COBOL CI/CD pipeline had grown fragile, slow, and tightly coupled to platform-specific logic. The original architecture relied on Groovy scripts spread across four repositories with runtime dependency installation, leading to long execution times, high maintenance costs, and vendor lock-in. We report on the migration to a containerized architecture featuring an abstraction layer for platform logic, simplified repository structure, and a pre-built OCI-compliant image containing COBOL tools and dependencies. The new design achieved an 82% runtime reduction. Our experience highlights lessons on abstraction, containerization, and organizational adoption, offering guidance for modernizing pipelines in legacy, high-security environments.

Existing methods for AI psychological counselors predominantly rely on supervised fine-tuning using static dialogue datasets. However, this contrasts with human experts, who continuously refine their proficiency through clinical practice and accumulated experience. To bridge this gap, we propose an Experience-Driven Lifelong Learning Agent (\texttt{PsychAgent}) for psychological counseling. First, we establish a Memory-Augmented Planning Engine tailored for longitudinal multi-session interactions, which ensures therapeutic continuity through persistent memory and strategic planning. Second, to support self-evolution, we design a Skill Evolution Engine that extracts new practice-grounded skills from historical counseling trajectories. Finally, we introduce a Reinforced Internalization Engine that integrates the evolved skills into the model via rejection fine-tuning, aiming to improve performance across diverse scenarios. Comparative analysis shows that our approach achieves higher scores than strong general LLMs (e.g., GPT-5.4, Gemini-3) and domain-specific baselines across all reported evaluation dimensions. These results suggest that lifelong learning can improve the consistency and overall quality of multi-session counseling responses.

We present DANCEMATCH, an end-to-end framework for motion-based dance retrieval, the task of identifying semantically similar choreographies directly from raw video, defined as DANCE FINGERPRINTING. While existing motion analysis and retrieval methods can compare pose sequences, they rely on continuous embeddings that are difficult to index, interpret, or scale. In contrast, DANCEMATCH constructs compact, discrete motion signatures that capture the spatio-temporal structure of dance while enabling efficient large-scale retrieval. Our system integrates Skeleton Motion Quantisation (SMQ) with Spatio-Temporal Transformers (STT) to encode human poses, extracted via Apple CoMotion, into a structured motion vocabulary. We further design DANCE RETRIEVAL ENGINE (DRE), which performs sub-linear retrieval using a histogram-based index followed by re-ranking for refined matching. To facilitate reproducible research, we release DANCETYPESBENCHMARK, a pose-aligned dataset annotated with quantised motion tokens. Experiments demonstrate robust retrieval across diverse dance styles and strong generalisation to unseen choreographies, establishing a foundation for scalable motion fingerprinting and quantitative choreographic analysis.

Adapting large language models (LLMs) to new languages is an expensive and opaque process. Understanding how language models acquire new languages and multilingual abilities is key to achieve efficient adaptation. Prior work on multilingual interpretability research focuses primarily on how trained models process multilingual instructions, leaving unexplored the mechanisms through which they acquire new languages during training. We investigate these training dynamics on decoder-only transformers through the lens of two functional cognitive specializations: language perception (input comprehension) and production (output generation). Through experiments on low-resource languages, we demonstrate how perceptual and productive specialization emerges in different regions of a language model by running layer ablation sweeps from the model's input and output directions. Based on the observed specialization patterns, we propose CogSym, a layer-wise heuristic that enables effective adaptation by exclusively fine-tuning a few early and late layers. We show that tuning only the 25% outermost layers achieves downstream task performance within 2-3% deviation from the full fine-tuning baseline. CogSym yields consistent performance with adapter methods such as LoRA, showcasing generalization beyond full fine-tuning. These findings provide insights to better understand how LLMs learn new languages and push toward accessible and inclusive language modeling.

Modern vision pipelines increasingly rely on pretrained image encoders whose representations are reused across tasks and models, yet these representations are often overcomplete and model-specific. We propose a simple, training-free method to improve the efficiency of image representations via a post-hoc canonical correlation analysis (CCA) operator. By leveraging the shared structure between representations produced by two pre-trained image encoders, our method finds linear projections that serve as a principled form of representation selection and dimensionality reduction, retaining shared semantic content while discarding redundant dimensions. Unlike standard dimensionality reduction techniques such as PCA, which operate on a single embedding space, our approach leverages cross-model agreement to guide representation distillation and refinement. The technique allows representations to be reduced by more than 75% in dimensionality with improved downstream performance, or enhanced at fixed dimensionality via post-hoc representation transfer from larger or fine-tuned models. Empirical results on ImageNet-1k, CIFAR-100, MNIST, and additional benchmarks show consistent improvements over both baseline and PCA-projected representations, with accuracy gains of up to 12.6%.

We present the GPT-NL Public Corpus, the biggest permissively licensed corpus of Dutch language resources. The GPT-NL Public Corpus contains 21 Dutch-only collections totalling 36B preprocessed Dutch tokens not present in any other LLM pretraining corpus. Additionally, the corpus includes roughly 207B English, 232B Code, and 48B German/Danish tokens taken from existing sets which we further curated for compliance. This corpus includes curated data from large existing corpora like Common Corpus and Common Crawl, as well as newly created Dutch-specific collections. Most newly created Dutch collections consist of content collected in collaboration with organisations or synthetically augmented content. All data is collected and evaluated with the aim of facilitating the creation of (commercial) language models that are lawful, useful and non-harmful. All data included in the GPT-NL Public Corpus is sourced from datasets with permissive licensing and is curated and redistributed under a CC-BY license. The full dataset is publicly available on the Hugging Face Hub.

Variational autoencoders (VAEs) learn compact latent representations of complex data, but their generative capacity is fundamentally constrained by the choice of prior distribution over the latent space. Energy-based priors offer a principled way to move beyond factorized assumptions and capture structured interactions among latent variables, yet training such priors at scale requires accurate and efficient sampling from intractable distributions. Here we present Boltzmann-machine--prior VAEs (BM-VAEs) trained using quantum annealing--based sampling in three distinct operational modes within a single generative system. During training, diabatic quantum annealing (DQA) provides unbiased Boltzmann samples for gradient estimation of the energy-based prior; for unconditional generation, slower quantum annealing (QA) concentrates samples near low-energy minima; for conditional generation, bias fields are added to direct sampling toward attribute-specific regions of the energy landscape (c-QA). Using up to 2000 qubits on a D-Wave Advantage2 processor, we demonstrate stable and efficient training across multiple datasets, with faster convergence and lower reconstruction loss than a Gaussian-prior VAE. The learned Boltzmann prior enables unconditional generation by sampling directly from the energy-based latent distribution, a capability that plain autoencoders lack, and conditional generation through latent biasing that leverages the learned pairwise interactions.

Spectral graph neural networks learn graph filters, but their behavior with increasing depth and polynomial order is not well understood. We analyze these models in the graph Fourier domain, where each layer becomes an element-wise frequency update, separating the fixed spectrum from trainable parameters and making depth and order explicit. In this setting, we show that Gaussian complexity is invariant under the Graph Fourier Transform, which allows us to derive data-dependent, depth, and order-aware generalization bounds together with stability estimates. In the linear case, our bounds are tighter, and on real graphs, the data-dependent term correlates with the generalization gap across polynomial bases, highlighting practical choices that avoid frequency amplification across layers.

The rise of large language models for code has reshaped software development. Autonomous coding agents, able to create branches, open pull requests, and perform code reviews, now actively contribute to real-world projects. Their growing role offers a unique and timely opportunity to investigate AI-driven contributions and their effects on code quality, team dynamics, and software maintainability. In this work, we construct a novel dataset of approximately $110,000$ open-source pull requests, including associated commits, comments, reviews, issues, and file changes, collectively representing millions of lines of source code. We compare five popular coding agents, including OpenAI Codex, Claude Code, GitHub Copilot, Google Jules, and Devin, examining how their usage differs in various development aspects such as merge frequency, edited file types, and developer interaction signals, including comments and reviews. Furthermore, we emphasize that code authoring and review are only a small part of the larger software engineering process, as the resulting code must also be maintained and updated over time. Hence, we offer several longitudinal estimates of survival and churn rates for agent-generated versus human-authored code. Ultimately, our findings indicate an increasing agent activity in open-source projects, although their contributions are associated with more churn over time compared to human-authored code.

Estimation of heterogeneous long-term treatment effects (HLTEs) is widely used for personalized decision-making in marketing, economics, and medicine, where short-term randomized experiments are often combined with long-term observational data. However, HLTE estimation is challenging due to limited overlap in treatment or in observing long-term outcomes for certain subpopulations, which can lead to unstable HLTE estimates with large finite-sample variance. To address this challenge, we introduce the LT-O-learners (Long-Term Orthogonal Learners), a set of novel orthogonal learners for HLTE estimation. The learners are designed for the canonical HLTE setting that combines a short-term randomized dataset $\mathcal{D}_1$ with a long-term historical dataset $\mathcal{D}_2$. The key idea of our LT-O-Learners is to retarget the learning objective by introducing custom overlap weights that downweight samples with low overlap in treatment or in long-term observation. We show that the retargeted loss is equivalent to the weighted oracle loss and satisfies Neyman-orthogonality, which means our learners are robust to errors in the nuisance estimation. We further provide a general error bound for the LT-O-Learners and give the conditions under which quasi-oracle rate can be achieved. Finally, our LT-O-learners are model-agnostic and can thus be instantiated with arbitrary machine learning models. We conduct empirical evaluations on synthetic and semi-synthetic benchmarks to confirm the theoretical properties of our LT-O-Learners, especially the robustness in low-overlap settings. To the best of our knowledge, ours are the first orthogonal learners for HLTE estimation that are robust to low overlap that is common in long-term outcomes.

2D assembly diagrams are often abstract and hard to follow, creating a need for intelligent assistants that can monitor progress, detect errors, and provide step-by-step guidance. In mixed reality settings, such systems must recognize completed and ongoing steps from the camera feed and align them with the diagram instructions. Vision Language Models (VLMs) show promise for this task, but face a depiction gap because assembly diagrams and video frames share few visual features. To systematically assess this gap, we construct IKEA-Bench, a benchmark of 1,623 questions across 6 task types on 29 IKEA furniture products, and evaluate 19 VLMs (2B-38B) under three alignment strategies. Our key findings: (1) assembly instruction understanding is recoverable via text, but text simultaneously degrades diagram-to-video alignment; (2) architecture family predicts alignment accuracy more strongly than parameter count; (3) video understanding remains a hard bottleneck unaffected by strategy. A three-level mechanistic analysis further reveals that diagrams and video occupy disjoint ViT subspaces, and that adding text shifts models from visual to text-driven reasoning. These results identify visual encoding as the primary target for improving cross-depiction robustness. Project page: https://ryenhails.github.io/IKEA-Bench/

In this work, we study whether enforcing strict compositional structure in sequence embeddings yields meaningful geometric organization when applied to protein-protein interaction networks. Using Event2Vec, an additive sequence embedding model, we train 64-dimensional representations on random walks from the human STRING interactome, and compare against a DeepWalk baseline based on Word2Vec, trained on the same walks. We find that compositional structure substantially improves pathway coherence (30.2$\times$ vs 2.9$\times$ above random), functional analogy accuracy (mean similarity 0.966 vs 0.650), and hierarchical pathway organization, while geometric properties such as norm--degree anticorrelation are shared with or exceeded by the non-compositional baseline. These results indicate that enforced compositionality specifically benefits relational and compositional reasoning tasks in biological networks.

Learning to defer (L2D) enables human-AI cooperation by deciding when an AI system should act autonomously or defer to a human expert. Existing L2D methods, however, assume static human performance, contradicting well-established findings on fatigue-induced degradation. We propose Fatigue-Aware Learning to Defer via Constrained Optimisation (FALCON), which explicitly models workload-varying human performance using psychologically grounded fatigue curves. FALCON formulates L2D as a Constrained Markov Decision Process (CMDP) whose state includes both task features and cumulative human workload, and optimises accuracy under human-AI cooperation budgets via PPO-Lagrangian training. We further introduce FA-L2D, a benchmark that systematically varies fatigue dynamics from near-static to rapidly degrading regimes. Experiments across multiple datasets show that FALCON consistently outperforms state-of-the-art L2D methods across coverage levels, generalises zero-shot to unseen experts with different fatigue patterns, and demonstrates the advantage of adaptive human-AI collaboration over AI-only or human-only decision-making when coverage lies strictly between 0 and 1.

Multi-agent Retrieval-Augmented Generation (RAG), wherein each agent takes on a specific role, supports hard queries that require multiple steps and sources, or complex reasoning. Existing approaches, however, rely on static agent behaviors and fixed orchestration strategies, leading to brittle performance on diverse, multi-hop tasks. We identify two key limitations: the lack of continuously adaptive orchestration mechanisms and the absence of behavior-level learning for individual agents. To this end, we propose HERA, a hierarchical framework that jointly evolves multi-agent orchestration and role-specific agent prompts. At the global level, HERA optimizes query-specific agent topologies through reward-guided sampling and experience accumulation. At the local level, Role-Aware Prompt Evolution refines agent behaviors via credit assignment and dual-axes adaptation along operational and behavioral principles, enabling targeted, role-conditioned improvements. On six knowledge-intensive benchmarks, HERA achieves an average improvement of 38.69\% over recent baselines while maintaining robust generalization and token efficiency. Topological analyses reveal emergent self-organization, where sparse exploration yields compact, high-utility multi-agent networks, demonstrating both efficient coordination and robust reasoning.

Machine learning-based weather forecasting models now surpass state-of-the-art numerical weather prediction systems, but training and operating these models at high spatial resolution remains computationally expensive. We present a modular framework that decouples forecasting from spatial resolution by applying learned generative super-resolution as a post-processing step to coarse-resolution forecast trajectories. We formulate super-resolution as a stochastic inverse problem, using a residual formulation to preserve large-scale structure while reconstructing unresolved variability. The model is trained with flow matching exclusively on reanalysis data and is applied to global medium-range forecasts. We evaluate (i) design consistency by re-coarsening super-resolved forecasts and comparing them to the original coarse trajectories, and (ii) high-resolution forecast quality using standard ensemble verification metrics and spectral diagnostics. Results show that super-resolution preserves large-scale structure and variance after re-coarsening, introduces physically consistent small-scale variability, and achieves competitive probabilistic forecast skill at 0.25° resolution relative to an operational ensemble baseline, while requiring only a modest additional training cost compared with end-to-end high-resolution forecasting.

As LLM agents transition from short, static problem solving to executing complex, long-horizon tasks in dynamic environments, the ability to handle user interruptions, such as adding requirement or revising goals, during mid-task execution is becoming a core requirement for realistic deployment. However, existing benchmarks largely assume uninterrupted agent behavior or study interruptions only in short, unconstrained language tasks. In this paper, we present the first systematic study of interruptible agents in long-horizon, environmentally grounded web navigation tasks, where actions induce persistent state changes. We formalize three realistic interruption types, including addition, revision, and retraction, and introduce InterruptBench, a benchmark derived from WebArena-Lite that synthesizes high-quality interruption scenarios under strict semantic constraints. Using a unified interruption simulation framework, we evaluate six strong LLM backbones across single- and multi-turn interruption settings, analyzing both their effectiveness in adapting to updated intents and their efficiency in recovering from mid-task changes. Our results show that handling user interruptions effectively and efficiently during long-horizon agentic tasks remains challenging for powerful large-scale LLMs. Code and dataset are available at https://github.com/HenryPengZou/InterruptBench.

Geometric Problem Solving (GPS) remains at the heart of enhancing mathematical reasoning in large language models because it requires the combination of diagrammatic understanding, symbolic manipulation and logical inference. In existing literature, researchers have chiefly focused on synchronising the diagram descriptions with text literals and solving the problem. In this vein, they have either taken a neural, symbolic or neuro-symbolic approach. But this solves only the first two of the requirements, namely diagrammatic understanding and symbolic manipulation, while leaving logical inference underdeveloped. The logical inference is often limited to one chain-of-thought (CoT). To address this weakness in hitherto existing models, this paper proposes MARS-GPS, that generates multiple parallel reasoning rollouts augmented with Python code execution for numerical verification, ranks them using token-level entropy as a confidence signal, and aggregates answers through a multi-stage voting and self-verification pipeline. Empirical results show that MARS-GPS with 8 parallel rollouts achieves 88.8% on Geometry3K, a nearly +11% improvement over the prior state-of-the-art, with accuracy scaling consistently as the number of rollouts increases from 1 to 16 (+6.0% on ablation subset). We provide our code and data in an anonymous repository: https://anonymous.4open.science/r/MARS-GPS-DE55.

Document understanding and GUI interaction are among the highest-value applications of Vision-Language Models (VLMs), yet they impose exceptionally heavy computational burden: fine-grained text and small UI elements demand high-resolution inputs that produce tens of thousands of visual tokens. We observe that this cost is largely wasteful -- across document and GUI benchmarks, only 22--71\% of image patches are pixel-unique, the rest being exact duplicates of another patch in the same image. We propose \textbf{PixelPrune}, which exploits this pixel-level redundancy through predictive-coding-based compression, pruning redundant patches \emph{before} the Vision Transformer (ViT) encoder. Because it operates in pixel space prior to any neural computation, PixelPrune accelerates both the ViT encoder and the downstream LLM, covering the full inference pipeline. The method is training-free, requires no learnable parameters, and supports pixel-lossless compression ($τ{=}0$) as well as controlled lossy compression ($τ{>}0$). Experiments across three model scales and document and GUI benchmarks show that PixelPrune maintains competitive task accuracy while delivering up to 4.2$\times$ inference speedup and 1.9$\times$ training acceleration. Code is available at https://github.com/OPPO-Mente-Lab/PixelPrune.

Java static analysis frameworks are commonly compared under the assumption that analysis algorithms and configurations compose monotonically and yield semantically comparable results across tools. In this work, we show that this assumption is fundamentally flawed. We present a large-scale empirical study of semantic consistency within and across four widely used Java static analysis frameworks: Soot, SootUp, WALA, and Doop. Using precision partial orders over analysis algorithms and configurations, we systematically identify violations where increased precision introduces new call-graph edges or amplifies inconsistencies. Our results reveal three key findings. First, algorithmic precision orders frequently break within frameworks due to modern language features such as lambdas, reflection, and native modeling. Second, configuration choices strongly interact with analysis algorithms, producing synergistic failures that exceed the effects of algorithm or configuration changes alone. Third, cross-framework comparisons expose irreconcilable semantic gaps, demonstrating that different frameworks operate over incompatible notions of call-graph ground truth. These findings challenge prevailing evaluation practices in static analysis and highlight the need to reason jointly about algorithms, configurations, and framework semantics when assessing precision and soundness.

Medical imaging techniques, especially Magnetic Resonance Imaging (MRI), are accepted as the gold standard in the diagnosis and treatment planning of neurological diseases. However, the manual analysis of MRI images is a time-consuming process for radiologists and is prone to human error due to fatigue. In this study, two different Deep Learning approaches were developed and analyzed comparatively for the automatic detection and classification of brain tumors (Glioma, Meningioma, Pituitary, and No Tumor). In the first approach, a custom Convolutional Neural Network (CNN) architecture named "OkanNet", which has a low computational cost and fast training time, was designed from scratch. In the second approach, the Transfer Learning method was applied using the 50-layer ResNet-50 [1] architecture, pre-trained on the ImageNet dataset. In experiments conducted on an extended dataset compiled by Masoud Nickparvar containing a total of $7,023$ MRI images, the Transfer Learning-based ResNet-50 model exhibited superior classification performance, achieving $96.49\%$ Accuracy and $0.963$ Precision. In contrast, the custom OkanNet architecture reached an accuracy rate of $88.10\%$; however, it proved to be a strong alternative for mobile and embedded systems with limited computational power by yielding results approximately $3.2$ times faster ($311$ seconds) than ResNet-50 in terms of training time. This study demonstrates the trade-off between model depth and computational efficiency in medical image analysis through experimental data.

Actor-level stance detection aims to determine an author expressed position toward specific geopolitical actors mentioned or implicated in a text. Although transformer-based models have achieved relatively good performance in stance classification, they typically rely on unified representations that may not sufficiently capture heterogeneous linguistic signals, such as contrastive discourse structures, framing cues, and salient lexical indicators. This motivates the need for adaptive architectures that explicitly model diverse stance-expressive patterns. In this paper, we propose StanceMoE, a context-enhanced Mixture-of-Experts (MoE) architecture built upon a fine-tuned BERT encoder for actor-level stance detection. Our model integrates six expert modules designed to capture complementary linguistic signals, including global semantic orientation, salient lexical cues, clause-level focus, phrase-level patterns, framing indicators, and contrast-driven discourse shifts. A context-aware gating mechanism dynamically weights expert contributions, enabling adaptive routing based on input characteristics. Experiments are conducted on the StanceNakba 2026 Subtask A dataset, comprising 1,401 annotated English texts where the target actor is implicit in the text. StanceMoE achieves a macro-F1 score of 94.26%, outperforming traditional baselines, and alternative BERT-based variants.

Matrix mechanisms are often used to provide unbiased differentially private query answers when publishing statistics or creating synthetic data. Recent work has developed matrix mechanisms, such as ResidualPlanner and Weighted Fourier Factorizations, that scale to high dimensional datasets while providing optimality guarantees for workloads such as marginals and circular product queries. They operate by adding noise to a linearly independent set of queries that can compactly represent the desired workloads. In this paper, we present QuerySmasher, an alternative scalable approach based on a divide-and-conquer strategy. Given a workload that can be answered from various data marginals, QuerySmasher splits each query into sub-queries and re-assembles the pieces into mutually orthogonal sub-workloads. These sub-workloads represent small, low-dimensional problems that can be independently and optimally answered by existing low-dimensional matrix mechanisms. QuerySmasher then stitches these solutions together to answer queries in the original workload. We show that QuerySmasher subsumes prior work, like ResidualPlanner (RP), ResidualPlanner+ (RP+), and Weighted Fourier Factorizations (WFF). We prove that it can dominate those approaches, under sum squared error, for all workloads. We also experimentally demonstrate the scalability and accuracy of QuerySmasher.

Reinforcement Learning with Verifiable Rewards (RLVR) has become a central post-training paradigm for improving the reasoning capabilities of large language models. Yet existing methods share a common blind spot: they optimize policies based on instantaneous group-level or batch-level statistics without ever verifying whether the resulting update actually improved the model. This open-loop design -- updating in isolation at each step, guided only by within-group (batch) reward signals -- means optimization can drift or collapse with no mechanism to detect and correct these failures. We argue that the missing ingredient is policy improvement feedback: the ability to measure and optimize inter-iteration progress directly. To this end, we introduce Policy Improvement Reinforcement Learning (PIRL), a framework that replaces surrogate reward maximization with the explicit objective of maximizing cumulative policy improvement across iterations, and prove this temporal objective is perfectly aligned with maximizing final task performance. Building on PIRL, we propose Policy Improvement Policy Optimization (PIPO), which implements closed-loop optimization through retrospective verification. At each iteration, PIPO evaluates whether the previous update yielded genuine improvement against a sliding-window historical baseline, then actively reinforces beneficial updates and suppresses the harmful ones -- transforming an open-loop process into a self-correcting one. We provide theoretical analysis showing that PIPO performs ascent on the PIRL objective in expectation, and experiments on mathematical reasoning benchmarks demonstrate improved stability and performance over GRPO and its variants.

Large Language Models (LLMs) are increasingly applied to automate software engineering tasks, including the generation of UML class diagrams from natural language descriptions. While prior work demonstrates that LLMs can produce syntactically valid diagrams, syntactic correctness alone does not guarantee meaningful design. This study investigates whether LLMs can move beyond diagram translation to perform design synthesis, and how reliably they maintain design-oriented reasoning under variation. We introduce a preference-based few-shot prompting approach that biases LLM outputs toward designs satisfying object-oriented principles and pattern-consistent structures. Two design-intent benchmarks, each with three domain-only, paraphrased prompts and 10 repeated runs, are used to evaluate three LLMs (ChatGPT 4o-mini, Claude 3.5 Sonnet, Gemini 2.5 Flash) across three modeling strategies: standard prompting, rule-injection prompting, and preference-based prompting, totaling 540 experiments (i.e. 2x3x10x3x3). Results indicate that while preference-based alignment improves adherence to design intent it does not eliminate non-determinism, and model-level behavior strongly influences design reliability. These findings highlight that achieving dependable LLM-assisted software design requires not only effective prompting but also careful consideration of model behavior and robustness.

Proactive agents that anticipate user needs and autonomously execute tasks hold great promise as digital assistants, yet the lack of realistic user simulation frameworks hinders their development. Existing approaches model apps as flat tool-calling APIs, failing to capture the stateful and sequential nature of user interaction in digital environments and making realistic user simulation infeasible. We introduce Proactive Agent Research Environment (Pare), a framework for building and evaluating proactive agents in digital environments. Pare models applications as finite state machines with stateful navigation and state-dependent action space for the user simulator, enabling active user simulation. Building on this foundation, we present Pare-Bench, a benchmark of 143 diverse tasks spanning communication, productivity, scheduling, and lifestyle apps, designed to test context observation, goal inference, intervention timing, and multi-app orchestration.

Traditional Online Public Access Catalogues (OPACs) are becoming less effective due to the rapid growth of scholarly literature. Conventional search methods, such as keyword indexing and Boolean queries, often fail to support efficient knowledge discovery. This paper proposes a Smart OPAC framework that transforms traditional OPACs into intelligent discovery systems using artificial intelligence and knowledge graph techniques. The framework enables semantic search, thematic filtering, and knowledge graph-based visualization to enhance user interaction and exploration. It integrates multiple open scholarly data sources and applies semantic embeddings to improve relevance and contextual understanding. The system supports exploratory search, semantic navigation, and refined result filtering based on user-defined themes. Quantitative evaluation demonstrates improvements in retrieval efficiency, relevance, and reduction of information overload. The proposed approach offers practical implications for modernizing digital library services and supports next-generation research workflows. Future work includes user-centric evaluation, personalization, and dynamic knowledge graph updates.

Large language models are increasingly being deployed as autonomous agents yet their real world effectiveness depends on reliable tools for information retrieval, computation and external action. Existing studies remain fragmented across tasks, tool types, and training settings, lacking a unified view of how tool-use methods differ and evolve. This paper organizes the literature into three paradigms: prompting as plug-and-play, supervised tool learning and reward-driven tool policy learning, analyzes their methods, strengths and failure modes, reviews the evaluation landscape and highlights key challenges, aiming to address this fragmentation and provide a more structured evolutionary view of agentic tool use.

Test-Time Learning (TTL) enables language agents to iteratively refine their performance through repeated interactions with the environment at inference time. At the core of TTL is an adaptation policy that updates the actor policy based on experience from previous episodes, thereby improving future behavior. Existing methods rely on fixed, hand-crafted adaptation policies rather than optimizing them for downstream improvement. We argue that optimal adaptation policies should be learned from task environments, not hand-engineered based on human intuition. To achieve this, we introduce Meta-TTL, a framework that formulates the discovery of effective adaptation policies as a bi-level optimization problem. Within this framework, the inner loop executes the standard TTL process, measuring how effectively a candidate adaptation policy helps an agent correct errors across sequential episodes. Guided by the agent's performance, the outer loop employs evolutionary search over a diverse distribution of training tasks to iteratively refine the adaptation policy. We evaluate Meta-TTL on Jericho and WebArena-Lite across both in-distribution (ID) and out-of-distribution (OOD) settings, using multiple meta-agent backbones. Results on both benchmarks show that Meta-TTL consistently outperforms hand-crafted baselines, suggesting that the optimized adaptation policy encodes transferable strategies that generalize beyond the training task distribution.

Time series forecasting (TSF) is critical across domains such as finance, meteorology, and energy. While extending the lookback window theoretically provides richer historical context, in practice, it often introduces irrelevant noise and computational redundancy, preventing models from effectively capturing complex long-term dependencies. To address these challenges, we propose a Dynamic Semantic Compression (DySCo) framework. Unlike traditional methods that rely on fixed heuristics, DySCo introduces an Entropy-Guided Dynamic Sampling (EGDS) mechanism to autonomously identify and retain high-entropy segments while compressing redundant trends. Furthermore, we incorporate a Hierarchical Frequency-Enhanced Decomposition (HFED) strategy to separate high-frequency anomalies from low-frequency patterns, ensuring that critical details are preserved during sparse sampling. Finally, a Cross-Scale Interaction Mixer(CSIM) is designed to dynamically fuse global contexts with local representations, replacing simple linear aggregation. Experimental results demonstrate that DySCo serves as a universal plug-and-play module, significantly enhancing the ability of mainstream models to capture long-term correlations with reduced computational cost.

Adapting pretrained language models (LMs) into vision-language models (VLMs) can degrade their native linguistic capability due to representation shift and cross-modal interference introduced during multimodal adaptation. Such loss is difficult to recover, even with targeted task-specific fine-tuning using standard objectives. Prior recovery approaches typically introduce additional modules that act as intermediate alignment layers to maintain or isolate modality-specific subspaces, which increases architectural complexity, adds parameters at inference time, and limits flexibility across models and settings. We propose LinguDistill, an adapter-free distillation method that restores linguistic capability by utilizing the original frozen LM as a teacher. We overcome the key challenge of enabling vision-conditioned teacher supervision by introducing layer-wise KV-cache sharing, which exposes the teacher to the student's multimodal representations without modifying the architecture of either model. We then selectively distill the teacher's strong linguistic signal on language-intensive data to recover language capability, while preserving the student's visual grounding on multimodal tasks. As a result, LinguDistill recovers $\sim$10% of the performance lost on language and knowledge benchmarks, while maintaining comparable performance on vision-heavy tasks. Our findings demonstrate that linguistic capability can be recovered without additional modules, providing an efficient and practical solution to modality-specific degradation in multimodal models.

Training effective software engineering agents requires large volumes of task-specific trajectories, incurring substantial data construction costs. Inspired by the "Less-Is-More" hypothesis in mathematical reasoning, we investigate its extension to agentic scenarios and propose an end-to-end training framework that achieves superior agentic capabilities with fewer but higher-quality training trajectories. This is achieved via STITCH (Sliding-memory Trajectory Inference and Task Chunking Heuristic), a coarse-to-fine mechanism that filters low-value noise and retains decision-critical tokens to maximize training signal quality. We conduct experiments across multiple agent frameworks (e.g., mini-SWE-agent, MSWE-agent), model scales (30B to 355B), and multilingual settings (Python, Java, and ArkTS). On SWE-bench Verified, models trained with STITCH achieve up to 63.16% relative improvement over base models. On Multi-SWE-bench (Java), MiniMax-M2.5-STITCH achieves 43.75% with our CodeArts Agent scaffold (+16.67%). On HarmonyOS (ArkTS), GLM-4.7-STITCH improves the compilation pass rate to 61.31% (+43.34%) with less than 1K training trajectories. Our results confirm that the "Less-Is-More" paradigm generalizes effectively to complex agentic tasks across diverse languages and model scales.

Low-rank decomposition has emerged as an important problem in Large Language Model (LLM) fine-tuning and inference. Through Singular Value Decomposition (SVD), the weight matrix can be factorized into low-rank spaces optimally. Previously, a common practice was to decompose the weight in the activation-whitened space, and then achieve satisfying results. In this work, we propose Optimal Brain Decomposition LLM (OBD-LLM), which studies the decomposition problem in the model space by utilizing second-order Hessian information. Through a rigorous Kronecker-factorization of the Hessian, we show that the decomposition needs to consider both input and output information of the layer, and achieves much better decomposition results compared to input only method. Our loss-aware decomposition method involves a bi-directional whitening on the weight matrix. As a result, OBD-LLM is a closed-form solution for the optimal decomposition of weights in the language model. Remarkably, we achieve ~20-40\% better results than previous state-of-the-art decomposition methods, the SVD-LLM.

Understanding emotions in natural language is inherently a multi-dimensional reasoning problem, where multiple affective signals interact through context, interpersonal relations, and situational cues. However, most existing emotion understanding benchmarks rely on short texts and predefined emotion labels, reducing this process to independent label prediction and ignoring the structured dependencies among emotions. To address this limitation, we introduce Emotional Scenarios (EmoScene), a theory-grounded benchmark of 4,731 context-rich scenarios annotated with an 8-dimensional emotion vector derived from Plutchik's basic emotions. We evaluate six instruction-tuned large language models in a zero-shot setting and observe modest performance, with the best model achieving a Macro F1 of 0.501, highlighting the difficulty of context-aware multi-label emotion prediction. Motivated by the observation that emotions rarely occur independently, we further propose an entanglement-aware Bayesian inference framework that incorporates emotion co-occurrence statistics to perform joint posterior inference over the emotion vector. This lightweight post-processing improves structural consistency of predictions and yields notable gains for weaker models (e.g., +0.051 Macro F1 for Qwen2.5-7B). EmoScene therefore provides a challenging benchmark for studying multi-dimensional emotion understanding and the limitations of current language models.

End-to-end autonomous driving has evolved from the conventional paradigm based on sparse perception into vision-language-action (VLA) models, which focus on learning language descriptions as an auxiliary task to facilitate planning. In this paper, we propose an alternative Vision-Geometry-Action (VGA) paradigm that advocates dense 3D geometry as the critical cue for autonomous driving. As vehicles operate in a 3D world, we think dense 3D geometry provides the most comprehensive information for decision-making. However, most existing geometry reconstruction methods (e.g., DVGT) rely on computationally expensive batch processing of multi-frame inputs and cannot be applied to online planning. To address this, we introduce a streaming Driving Visual Geometry Transformer (DVGT-2), which processes inputs in an online manner and jointly outputs dense geometry and trajectory planning for the current frame. We employ temporal causal attention and cache historical features to support on-the-fly inference. To further enhance efficiency, we propose a sliding-window streaming strategy and use historical caches within a certain interval to avoid repetitive computations. Despite the faster speed, DVGT-2 achieves superior geometry reconstruction performance on various datasets. The same trained DVGT-2 can be directly applied to planning across diverse camera configurations without fine-tuning, including closed-loop NAVSIM and open-loop nuScenes benchmarks.

We present experimental results from seven controlled runs of nanoFMT, a Free-Market Algorithm (FMA) orchestrated transformer with dynamic Mixture-of-Experts (MoE) management. The experiments address a fundamental question for advanced LLM development: how should an MoE system manage its expert pool when operating at full capacity under changing data distributions? We demonstrate that cost-penalized fitness metrics, combined with a linear grace period for newborn experts, produce a system that accumulates domain expertise through diversification rather than replacement. The central result is a round-trip domain shift experiment showing 9-11x faster recovery when returning to a previously learned domain, with zero expert births or replacements required. This "molecular memory" effect -- where dormant experts survive and reactivate when their domain returns -- has no analogue in current MoE management approaches. A preliminary cost analysis estimates annual savings of $39.1M and 27.1 GWh energy reduction for an OpenAI-scale provider under a moderate scenario.

Overlap, also known as positivity, is a key condition for causal treatment effect estimation. Many popular estimators suffer from high variance and become brittle when features differ strongly across treatment groups. This is especially challenging in high dimensions: the curse of dimensionality can make overlap implausible. To address this, we propose a class of feature representations called deconfounding scores, which preserve both identification and the target of estimation; the classical propensity and prognostic scores are two special cases. We characterize the problem of finding a representation with better overlap as minimizing an overlap divergence under a deconfounding score constraint. We then derive closed-form expressions for a class of deconfounding scores under a broad family of generalized linear models with Gaussian features and show that prognostic scores are overlap-optimal within this class. We conduct extensive experiments to assess this behavior empirically.

Standard Mixture-of-Experts (MoE) models rely on centralized routing mechanisms that introduce rigid inductive biases. We propose Routing-Free MoE which eliminates any hard-coded centralized designs including external routers, Softmax, Top-K and load balancing, instead encapsulating all activation functionalities within individual experts and directly optimized through continuous gradient flow, enabling each expert to determine its activation entirely on its own. We introduce a unified adaptive load-balancing framework to simultaneously optimize both expert-balancing and token-balancing objectives through a configurable interpolation, allowing flexible and customizable resource allocation. Extensive experiments show that Routing-Free MoE can consistently outperform baselines with better scalability and robustness. We analyze its behavior in detail and offer insights that may facilitate future MoE design ad optimization.

Plant phenology modelling aims to predict the timing of seasonal phases, such as leaf-out or flowering, from meteorological time series. Reliable predictions are crucial for anticipating ecosystem responses to climate change. While phenology modelling has traditionally relied on mechanistic approaches, deep learning methods have recently been proposed as flexible, data-driven alternatives with often superior performance. However, mechanistic models tend to outperform deep networks when data distribution shifts are induced by climate change. Domain Adaptation (DA) techniques could help address this limitation. Yet, unlike standard DA settings, climate change induces a temporal continuum of domains and involves both a covariate and label shift, with warmer records and earlier start of spring. To tackle this challenge, we introduce Mid-feature Rank-adversarial Domain Adaptation (MIRANDA). Whereas conventional adversarial methods enforce domain invariance on final latent representations, an approach that does not explicitly address label shift, we apply adversarial regularization to intermediate features. Moreover, instead of a binary domain-classification objective, we employ a rank-based objective that enforces year-invariance in the learned meteorological representations. On a country-scale dataset spanning 70 years and comprising 67,800 phenological observations of 5 tree species, we demonstrate that, unlike conventional DA approaches, MIRANDA improves robustness to climatic distribution shifts and narrows the performance gap with mechanistic models.

Spatial consistency is a fundamental property of the visual world and a key requirement for models that aim to understand physical reality. Despite recent advances, multimodal large language models (MLLMs) often struggle to reason about 3D geometry across multiple views. Rather than asking models to describe scene attributes, we introduce a more challenging task: given two views of the same scene, identify the object that violates 3D motion consistency. We propose a simple and scalable method for generating realistic, spatially inconsistent image pairs from multi-view scenes, enabling systematic evaluation of this capability. Our results show that state-of-the-art MLLMs significantly underperform human observers and exhibit substantial variability across different scene attributes, revealing a fragile and incomplete understanding of 3D structure. We hope our findings underscore the need for approaches that develop a more deeply grounded understanding of the physical world.

We propose the Preference Guided Iterated Pareto Referent Optimisation (PG-IPRO) for urban route planning for people with different accessibility requirements and preferences. With this algorithm the user can interact with the system by giving feedback on a route, i.e., the user can say which objective should be further minimized, or conversely can be relaxed. This leads to intuitive user interaction, that is especially effective during early iterations compared to information-gain-based interaction. Furthermore, due to PG-IPRO's iterative nature, the full set of alternative, possibly optimal policies (the Pareto front), is never computed, leading to higher computational efficiency and shorter waiting times for users.

While large language models (LLMs) have demonstrated strong performance on complex reasoning tasks such as competitive programming (CP), existing methods predominantly focus on single-attempt settings, overlooking their capacity for iterative refinement. In this paper, we present RefineRL, a novel approach designed to unleash the self-refinement capabilities of LLMs for CP problem solving. RefineRL introduces two key innovations: (1) Skeptical-Agent, an iterative self-refinement agent equipped with local execution tools to validate generated solutions against public test cases of CP problems. This agent always maintains a skeptical attitude towards its own outputs and thereby enforces rigorous self-refinement even when validation suggests correctness. (2) A reinforcement learning (RL) solution to incentivize LLMs to self-refine with only standard RLVR data (i.e., problems paired with their verifiable answers). Extensive experiments on Qwen3-4B and Qwen3-4B-2507 demonstrate that our method yields substantial gains: after our RL training, these compact 4B models integrated with the Skeptical-Agent not only outperform much larger 32B models but also approach the single-attempt performance of 235B models. These findings suggest that self-refinement holds considerable promise for scaling LLM reasoning, with significant potential for further advancement.

In the previous work, a lexical (re)categorisation -- or confirmation of the given category -- of roots identified as verbal was undertaken to determine their original category accurately. Building on this, the present paper offers an account of the valency classification of those Mapudungun roots confirmed to be verbal, using the language's own morphotactics; specifically, by examining the permissible and restricted combinations of various suffixes with roots or verbal stems in the Mapuche verb form. As with all work conducted thus far, the results presented here aim to improve the morphological analyser (Dungupeyum) with all verified findings incorporated into the system. From a theoretical perspective, we also hope to contribute to the recognition and understanding of issues related to the valency of Mapuche verb forms.

This technical report presents methods developed by the UK AI Security Institute for assessing whether advanced AI systems reliably follow intended goals. Specifically, we evaluate whether frontier models sabotage safety research when deployed as coding assistants within an AI lab. Applying our methods to four frontier models, we find no confirmed instances of research sabotage. However, we observe that Claude Opus 4.5 Preview (a pre-release snapshot of Opus 4.5) and Sonnet 4.5 frequently refuse to engage with safety-relevant research tasks, citing concerns about research direction, involvement in self-training, and research scope. We additionally find that Opus 4.5 Preview shows reduced unprompted evaluation awareness compared to Sonnet 4.5, while both models can distinguish evaluation from deployment scenarios when prompted. Our evaluation framework builds on Petri, an open-source LLM auditing tool, with a custom scaffold designed to simulate realistic internal deployment of a coding agent. We validate that this scaffold produces trajectories that all tested models fail to reliably distinguish from real deployment data. We test models across scenarios varying in research motivation, activity type, replacement threat, and model autonomy. Finally, we discuss limitations including scenario coverage and evaluation awareness.

The Mining Software Repositories (MSR) field focuses on analysing the rich data contained in software repositories to derive actionable insights into software processes and products. Mining repositories at scale requires techniques capable of handling large volumes of heterogeneous data, a challenge for which language models (LMs) are increasingly well-suited. Since the advent of Transformer-based architectures, LMs have been rapidly adopted across a wide range of MSR tasks. This article presents a comprehensive survey of the use of LMs in MSR, based on an analysis of 85 papers. We examine how LMs are applied, the types of artefacts analysed, which models are used, how their adoption has evolved over time, and the extent to which studies support reproducibility and reuse. Building on this analysis, we propose a taxonomy of LM applications in MSR, identify key trends shaping the field, and highlight open challenges alongside actionable directions for future research.

Pretraining Large Language Models (LLMs) from scratch requires massive amount of compute. Aurora super computer is an ExaScale machine with 127,488 Intel PVC (Ponte Vechio) GPU tiles. In this work, we showcase LLM pretraining on Aurora at the scale of 1000s of GPU tiles. Towards this effort, we developed Optimus, an inhouse training library with support for standard large model training techniques. Using Optimus, we first pretrained Mula-1B, a 1 Billion dense model and Mula-7B-A1B, a 7 Billion Mixture of Experts (MoE) model from scratch on 3072 GPU tiles for the full 4 trillion tokens of the OLMoE-mix-0924 dataset. We then demonstrated model scaling by pretraining three large MoE models Mula-20B-A2B, Mula-100B-A7B, and Mula-220B-A10B till 100 Billion tokens on the same dataset. On our largest model Mula-220B-A10B, we pushed the compute scaling from 384 to 12288 GPU tiles and observed scaling efficiency of around 90% at 12288 GPU tiles. We significantly improved the runtime performance of MoE models using custom GPU kernels for expert computation, and a novel EP-Aware sharded optimizer resulting in training speedups up to 1.71x. As part of the Optimus library, we also developed a robust set of reliability and fault tolerant features to improve training stability and continuity at scale.

Label prediction in neural networks (NNs) has O(n) complexity proportional to the number of classes. This holds true for classification using fully connected layers and cosine similarity with some set of class prototypes. In this paper we show that if NN latent space (LS) geometry is known and possesses specific properties, label prediction complexity can be significantly reduced. This is achieved by associating label prediction with the O(1) complexity closest cluster center search in a vector system used as target for latent space configuration (LSC). The proposed method only requires finding indexes of several largest and lowest values in the embedding vector making it extremely computationally efficient. We show that the proposed method does not change NN training accuracy computational results. We also measure the time required by different computational stages of NN inference and label prediction on multiple datasets. The experiments show that the proposed method allows to achieve up to 11.6 times overall acceleration over conventional methods. Furthermore, the proposed method has unique properties which allow to predict the existence of new classes.

Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks. Although this limitation has been noted, the internal reasons remain unclear. We use character counting (e.g., "How many p's are in apple?") as a minimal, controlled probe that isolates token-level reasoning from higher-level confounds. Using this setting, we uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma: models often compute the correct answer internally yet fail to express it at the output layer. Through mechanistic analysis combining probing classifiers, activation patching, logit lens analysis, and attention head tracing, we show that character-level information is encoded in early and mid-layer representations. However, this information is attenuated by a small set of components in later layers, especially the penultimate and final layer MLP. We identify these components as negative circuits: subnetworks that downweight correct signals in favor of higher-probability but incorrect outputs. Our results lead to two contributions. First, we show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph. This explains why such errors persist and can worsen under scaling and instruction tuning. Second, we provide evidence that LLM forward passes implement a form of competitive decoding, in which correct and incorrect hypotheses coexist and are dynamically reweighted, with final outputs determined by suppression as much as by amplification. These findings carry implications for interpretability and robustness: simple symbolic reasoning exposes weaknesses in modern LLMs, underscoring need for design strategies that ensure information is encoded and reliably used.

Mental health text classification has rapidly adopted modern adaptation methods, yet practical guidance on which optimization strategy to use, when, and why remains limited. This paper presents a systematic comparative study of optimization pathways for a joint mental-health classification task, moving from strong vanilla baselines to progressively more specialized techniques. We first establish classical and encoder references, then examine parameter-efficient supervised fine-tuning with LoRA/QLoRA under multiple objective and optimization settings, and finally evaluate preference-based optimization with DPO, ORPO, and KTO, including class-rebalanced training. Rather than emphasizing a single headline score, we focus on methodological insight: how performance changes with objective formulation, adapter choice, optimizer behavior, context windowing, and class-balance intervention. The results show that optimization effects are highly method-dependent: some approaches deliver stable, transferable gains, while others are sensitive to configuration and data balance. Preference optimization, in particular, exhibits large variation across objectives, indicating that method selection is more consequential than simply adding a preference-training stage. The central contribution is a clear optimization narrative for mental health NLP: start from transparent baselines, apply controlled tuning, and use preference optimization selectively where its gains are demonstrable. This provides a reproducible and practically grounded framework for choosing effective training strategies beyond architecture choice alone.

A new generation of language models reasons entirely in continuous hidden states, producing no tokens and leaving no audit trail. We show that this silence creates a fundamentally new attack surface. ThoughtSteer perturbs a single embedding vector at the input layer; the model's own multi-pass reasoning amplifies this perturbation into a hijacked latent trajectory that reliably produces the attacker's chosen answer, while remaining structurally invisible to every token-level defense. Across two architectures (Coconut and SimCoT), three reasoning benchmarks, and model scales from 124M to 3B parameters, ThoughtSteer achieves >=99% attack success rate with near-baseline clean accuracy, transfers to held-out benchmarks without retraining (94-100%), evades all five evaluated active defenses, and survives 25 epochs of clean fine-tuning. We trace these results to a unifying mechanism: Neural Collapse in the latent space pulls triggered representations onto a tight geometric attractor, explaining both why defenses fail and why any effective backdoor must leave a linearly separable signature (probe AUC>=0.999). Yet a striking paradox emerges: individual latent vectors still encode the correct answer even as the model outputs the wrong one. The adversarial information is not in any single vector but in the collective trajectory, establishing backdoor perturbations as a new lens for mechanistic interpretability of continuous reasoning. Code and checkpoints are available.

Wearable HAR has improved steadily, but most progress still relies on closed-set classification, which limits real-world use. In practice, human activity is open-ended, unscripted, personalized, and often compositional, unfolding as narratives rather than instances of fixed classes. We argue that addressing this gap does not require simply scaling datasets or models. It requires a fundamental shift in how wearable HAR is formulated, supervised, and evaluated. This work shows how to model open-ended activity narratives by aligning wearable sensor data with natural-language descriptions in an open-vocabulary setting. Our framework has three core components. First, we introduce a naturalistic data collection and annotation pipeline that combines multi-position wearable sensing with free-form, time-aligned narrative descriptions of ongoing behavior, allowing activity semantics to emerge without a predefined vocabulary. Second, we define a retrieval-based evaluation framework that measures semantic alignment between sensor data and language, enabling principled evaluation without fixed classes while also subsuming closed-set classification as a special case. Third, we present a language-conditioned learning architecture that supports sensor-to-text inference over variable-length sensor streams and heterogeneous sensor placements. Experiments show that models trained with fixed-label objectives degrade sharply under real-world variability, while open-vocabulary sensor-language alignment yields robust and semantically grounded representations. Once this alignment is learned, closed-set activity recognition becomes a simple downstream task. Under cross-participant evaluation, our method achieves 65.3% Macro-F1, compared with 31-34% for strong closed-set HAR baselines. These results establish open-ended narrative modeling as a practical and effective foundation for real-world wearable HAR.

Large Vision Language Models show impressive performance across image and video understanding tasks, yet their computational cost grows rapidly with the number of visual tokens. Existing token pruning methods mitigate this issue through empirical approaches while overlooking the internal mechanism of attention. In this paper, we propose a novel training free token pruning framework grounded in the dual form perspective of attention. We reformulate attention as an implicit linear layer whose weight matrix is the sum of rank 1 outer products, each generated by a single token's key value pair. Token pruning thus reduces to selecting an optimal subset of these rank 1 updates that best approximates the original dual weight matrix. Extending this perspective to standard softmax attention in LVLMs, we derive a novel metric quantifying both a token's information magnitude and information duplication. To efficiently select the subset with the proposed metric, we introduce Progressive Chunked Maximal Marginal Relevance. Extensive experiments demonstrate that our method achieves a better trade off between performance and efficiency, while providing another perspective on existing pruning approaches.

The whole-brain connectome of a fruit fly comprises over 130K neurons connected with a probability of merely 0.02%, yet achieves an average shortest path of only 4.4 hops. Despite being highly structured at the circuit level, the network's long-range connections are broadly distributed across brain regions, functioning as stochastic shortcuts that enable efficient global communication. Inspired by this observation, we propose Stochastic Attention (SA), a drop-in enhancement for sliding-window attention (SWA) that applies a random permutation to the token sequence before windowed attention and restores the original order afterward. This transforms the fixed local window into a stochastic global one within the same $O(nw)$ per-layer budget. Through depth, independently sampled permutations yield exponentially growing receptive fields, achieving full sequence coverage in $O(\log_w n)$ layers versus $O(n/w)$ for SWA. We validate SA in two settings: pre-training language models from scratch, where a gated SA + SWA combination achieves the best average zero-shot accuracy, and training-free inference on Qwen3-8B and Qwen3-30B-A3B, where SA consistently outperforms SWA and matches or exceeds Mixture of Block Attention at comparable compute budgets. These results suggest that connectome-inspired stochastic routing is a practical primitive for improving the expressivity of efficient attention, complementary to existing linear and sparse approaches.

Datasets used in immunotherapy response prediction are typically small in size, as well as diverse in cancer type, drug administered, and sequencer used. Models often drop in performance when tested on patient cohorts that are not included in the training process. Recent work has shown that transformer-based models along with self-supervised learning show better generalisation performance than threshold-based biomarkers, but is still suboptimal. We present BioCOMPASS, an extension of a transformer-based model called COMPASS, that integrates biomarkers and treatment information to further improve its generalisability. Instead of feeding biomarker data as input, we built loss components to align them with the model's intermediate representations. We found that components such as treatment gating and pathway consistency loss improved generalisability when evaluated with Leave-one-cohort-out, Leave-one-cancer-type-out and Leave-one-treatment-out strategies. Results show that building components that exploit biomarker and treatment information can help in generalisability of immunotherapy response prediction. Careful curation of additional components that leverage complementary clinical information and domain knowledge represents a promising direction for future research.

The memory wall remains the primary bottleneck for training large language models on consumer hardware. We introduce Spectral Compact Training (SCT), a method that replaces dense weight matrices with permanent truncated SVD factors W = U diag(s) V^T, where the full dense matrix is never materialized during training or inference. Gradients flow through the compact spectral factors via standard backpropagation, and U, V are retracted to the Stiefel manifold via QR decomposition after each optimizer step. SCT achieves up to 199x memory reduction per MLP layer at rank 32, enabling full training steps of 70B-parameter architectures on a Steam Deck handheld (7.2 GB peak memory vs. 1,245 GB for dense FP32 training with Adam). Rank-sweep experiments on SmolLM2-1.7B (ranks 32-256, 2000 steps, NVIDIA A100) show that all tested ranks converge to the same loss floor (~4.2-4.5), identifying the learning rate schedule -- not MLP rank -- as the primary bottleneck. Rank 128 emerges as the efficiency sweet spot at 11.7x MLP compression with the lowest perplexity. GPU memory drops 46% at rank 32 while training throughput doubles.

Context: Schools, training platforms, and technology firms increasingly need to assess programming proficiency at scale with transparent, reproducible methods that support personalized learning pathways. Objective: This study introduces a pedagogical framework for Scratch project assessment, aligned with the Common European Framework of Reference (CEFR), providing universal competency levels for students and teachers alongside actionable insights for curriculum design. Method: We apply Fuzzy C-Means clustering to 2008246 Scratch projects evaluated via Dr.Scratch, implementing an ordinal criterion to map clusters to CEFR levels (A1-C2), and introducing enhanced classification metrics that identify transitional learners, enable continuous progress tracking, and quantify classification certainty to balance automated feedback with instructor review. Impact: The framework enables diagnosis of systemic curriculum gaps-notably a "B2 bottleneck" where only 13.3% of learners reside due to the cognitive load of integrating Logic Synchronization, and Data Representation--while providing certainty--based triggers for human intervention.

As Large Language Models (LLMs) scale in size and complexity, the consequences of failures during training become increasingly severe. A major challenge arises from Silent Data Corruption (SDC): hardware-induced faults that bypass system-level detection mechanisms. SDC may behave like benign numerical noise, but can also cause harmful gradient corruption that leads to loss spikes, divergence, or stalled progress. This work provides a controlled study of how intermittent SDC affects LLM pretraining. Using targeted fault injection at the level of GPU matrix-multiply instructions, we characterize the sensitivity of different bit positions, kernel functions, and execution stages. Our analysis shows that locally originating faults can produce impactful corruption, including NaN propagation, short-lived spikes in loss, gradient norm, and attention logits, as well as persistent parameter divergence. Building on the observed corruption signatures, we propose a lightweight detection method that identifies potentially harmful parameter updates. Experiments on LLaMA models with 60M, 350M, and 1.3B parameters demonstrate that recomputing the most recent training step upon detection can effectively mitigate the impact of these events.

End-to-end OCR for historical newspapers remains challenging, as models must handle long text sequences, degraded print quality, and complex layouts. While Transformer-based recognizers dominate current research, their quadratic complexity limits efficient paragraph-level transcription and large-scale deployment. We investigate linear-time State-Space Models (SSMs), specifically Mamba, as a scalable alternative to Transformer-based sequence modeling for OCR. We present to our knowledge, the first OCR architecture based on SSMs, combining a CNN visual encoder with bi-directional and autoregressive Mamba sequence modeling, and conduct a large-scale benchmark comparing SSMs with Transformer- and BiLSTM-based recognizers. Multiple decoding strategies (CTC, autoregressive, and non-autoregressive) are evaluated under identical training conditions alongside strong neural baselines (VAN, DAN, DANIEL) and widely used off-the-shelf OCR engines (PERO-OCR, Tesseract OCR, TrOCR, Gemini). Experiments on historical newspapers from the Bibliothèque nationale du Luxembourg, with newly released >99% verified gold-standard annotations, and cross-dataset tests on Fraktur and Antiqua lines, show that all neural models achieve low error rates (~2% CER), making computational efficiency the main differentiator. Mamba-based models maintain competitive accuracy while halving inference time and exhibiting superior memory scaling (1.26x vs 2.30x growth at 1000 chars), reaching 6.07% CER at the severely degraded paragraph level compared to 5.24% for DAN, while remaining 2.05x faster. We release code, trained models, and standardized evaluation protocols to enable reproducible research and guide practitioners in large-scale cultural heritage OCR.

Large language model (LLM) agents struggle to autonomously evolve coordination strategies in dynamic environments, largely because coarse global outcomes obscure the causal signals needed for local policy refinement. We identify this bottleneck as a multi-agent credit assignment problem, which has long been studied in classical multi-agent reinforcement learning (MARL) but remains underaddressed in LLM-based systems. Building on this observation, we propose LangMARL, a framework that brings credit assignment and policy gradient evolution from cooperative MARL into the language space. LangMARL introduces agent-level language credit assignment, pioneers gradient evolution in language space for policy improvement, and summarizes task-relevant causal relations from replayed trajectories to provide dense feedback and improve convergence under sparse rewards. Extensive experiments across diverse cooperative multi-agent tasks demonstrate improved sample efficiency, interpretability, and strong generalization.

Non-stationarity arises from concurrent policy updates and leads to persistent environmental fluctuations. Existing approaches like Centralized Training with Decentralized Execution (CTDE) and sequential update schemes mitigate this issue. However, since the perception of the policies of other agents remains dependent on sampling environmental interaction data, the agent essentially operates in a passive perception state. This inevitably triggers equilibrium oscillations and significantly slows the convergence speed of the system. To address this issue, we propose Gradient Realignment via Active Shared Perception (GRASP), a novel framework that defines generalized Bellman equilibrium as a stable objective for policy evolution. The core mechanism of GRASP involves utilizing the independent gradients of agents to derive a defined consensus gradient, enabling agents to actively perceive policy updates and optimize team collaboration. Theoretically, we leverage the Kakutani Fixed-Point Theorem to prove that the consensus direction $u^*$ guarantees the existence and attainability of this equilibrium. Extensive experiments on StarCraft II Multi-Agent Challenge (SMAC) and Google Research Football (GRF) demonstrate the scalability and promising performance of the framework.

Transformer language models contain localized reasoning circuits, contiguous layer blocks that improve reasoning when duplicated at inference time. Finding these circuits currently requires brute-force sweeps costing 25 GPU hours per model. We propose CircuitProbe, which predicts circuit locations from activation statistics in under 5 minutes on CPU, providing a speedup of three to four orders of magnitude. We find that reasoning circuits come in two types: stability circuits in early layers, detected through the derivative of representation change, and magnitude circuits in late layers, detected through anomaly scoring. We validate across 9 models spanning 6 architectures, including 2025 models, confirming that CircuitProbe top predictions match or are within 2 layers of the optimal circuit in all validated cases. A scaling experiment across the Qwen 2.5 family reveals that layer duplication consistently benefits models under 3B parameters but degrades performance in 7B+ models, making this a practical scaling technique for small language models. CircuitProbe requires as few as 10 calibration examples and its predictions are stable across English, Hindi, Chinese, and French.

Retrieval-augmented generation (RAG) improves language model (LM) performance by providing relevant context at test time for knowledge-intensive situations. However, the relationship between parametric knowledge acquired during pretraining and non-parametric knowledge accessed via retrieval remains poorly understood, especially under fixed data budgets. In this work, we systematically study the trade-off between pretraining corpus size and retrieval store size across a wide range of model and data scales. We train OLMo-2-based LMs ranging from 30M to 3B parameters on up to 100B tokens of DCLM data, while varying both pretraining data scale (1-150x the number of parameters) and retrieval store size (1-20x), and evaluate performance across a diverse suite of benchmarks spanning reasoning, scientific QA, and open-domain QA. We find that retrieval consistently improves performance over parametric-only baselines across model scales and introduce a three-dimensional scaling framework that models performance as a function of model size, pretraining tokens, and retrieval corpus size. This scaling manifold enables us to estimate optimal allocations of a fixed data budget between pretraining and retrieval, revealing that the marginal utility of retrieval depends strongly on model scale, task type, and the degree of pretraining saturation. Our results provide a quantitative foundation for understanding when and how retrieval should complement pretraining, offering practical guidance for allocating data resources in the design of scalable language modeling systems.

Assessing the veracity of a claim made online is a complex and important task with real-world implications. When these claims are directed at communities with limited access to information and the content concerns issues such as healthcare and culture, the consequences intensify, especially in low-resource languages. In this work, we introduce AfrIFact, a dataset that covers the necessary steps for automatic fact-checking (i.e., information retrieval, evidence extraction, and fact checking), in ten African languages and English. Our evaluation results show that even the best embedding models lack cross-lingual retrieval capabilities, and that cultural and news documents are easier to retrieve than healthcare-domain documents, both in large corpora and in single documents. We show that LLMs lack robust multilingual fact-verification capabilities in African languages, while few-shot prompting improves performance by up to 43% in AfriqueQwen-14B, and task-specific fine-tuning further improves fact-checking accuracy by up to 26%. These findings, along with our release of the AfrIFact dataset, encourage work on low-resource information retrieval, evidence retrieval, and fact checking.

Large-scale web applications are widely deployed with complex third-party components, inheriting security risks arising from component vulnerabilities. Security assessment is therefore required to determine whether such known vulnerabilities remain practically exploitable in real applications. Penetration testing is a widely adopted approach that validates exploitability by launching concrete attacks against known vulnerabilities in real-world black-box systems. However, existing approaches often fail to automatically generate reliable exploits, limiting their effectiveness in practical security assessment. This limitation mainly stems from two issues: (1) precisely triggering vulnerabilities with correct technical details, and (2) adapting exploits to diverse real-world deployment settings. In this paper, we propose AutoEG, a fully automated multi-agent framework for exploit generation targeting black-box web applications. AutoEG has two phases: First, AutoEG extracts precise vulnerability trigger logic from unstructured vulnerability information and encapsulates it into reusable trigger functions. Second, AutoEG uses trigger functions for concrete attack objectives and iteratively refines exploits through feedback-driven interaction with the target application. We evaluate AutoEG on 104 real-world vulnerabilities with 29 attack objectives, resulting in 660 exploitation tasks and 55,440 exploit attempts. AutoEG achieves an average success rate of 82.41%, substantially outperforming state-of-the-art baselines, whose best performance reaches only 32.88%.

Due to their widespread use in industry, several techniques have been proposed in the literature to fuzz REST APIs. Existing fuzzers for REST APIs have been focusing on detecting crashes (e.g., 500 HTTP server error status code). However, security vulnerabilities can have major drastic consequences on existing cloud infrastructures. In this paper, we propose a series of novel automated oracles aimed at detecting violations of access policies in REST APIs, as well as executing traditional attacks such as SQL Injection and XSS. These novel automated oracles can be integrated into existing fuzzers, in which, once the fuzzing session is completed, a ``security testing'' phase is executed to verify these oracles. When a security fault is detected, as output our technique is able to general executable test cases in different formats, like Java, Kotlin, Python and JavaScript test suites. Our novel techniques are integrated as an extension of EvoMaster, a state-of-the-art open-source fuzzer for REST APIs. Experiments are carried out on 9 artificial examples, 8 vulnerable-by-design REST APIs with black-box testing, and 36 REST APIs from the WFD corpus with white-box testing, for a total of 52 distinct APIs. Results show that our novel oracles and their automated integration in a fuzzing process can lead to detect security issues in several of these APIs.

Group Relative Policy Optimization (GRPO) is widely used for reinforcement learning with verifiable rewards, but it often suffers from advantage collapse: when all rollouts in a group receive the same reward, the group yields zero relative advantage and thus no learning signal. For example, if a question is too hard for the reasoner, all sampled rollouts can be incorrect and receive zero reward. Recent work addresses this issue by adding hints or auxiliary scaffolds to such hard questions so that the reasoner produces mixed outcomes and recovers a non-zero update. However, existing hints are usually fixed rather than adapted to the current reasoner, and a hint that creates learning signal under the hinted input does not necessarily improve the no-hint policy used at test time. To this end, we propose Hint Learning for Reinforcement Learning (HiLL), a framework that jointly trains a hinter policy and a reasoner policy during RL. For each hard question, the hinter generates hints online conditioned on the current reasoner's incorrect rollout, allowing hint generation to adapt to the reasoner's evolving errors. We further introduce hint reliance, which measures how strongly correct hinted trajectories depend on the hint. We derive a transferability result showing that lower hint reliance implies stronger transfer from hinted success to no-hint success, and we use this result to define a transfer-weighted reward for training the hinter. Therefore, HiLL favors hints that not only recover informative GRPO groups, but also produce signals that are more likely to improve the original no-hint policy. Experiments across multiple benchmarks show that HiLL consistently outperforms GRPO and prior hint-based baselines, demonstrating the value of adaptive and transfer-aware hint learning for RL. The code is available at https://github.com/Andree-9/HiLL.

Gaussian processes (GPs) offer appealing properties but are costly to train at scale. Sparse variational GP (SVGP) approximations reduce cost yet still rely on Cholesky decompositions of kernel matrices, ill-suited to low-precision, massively parallel hardware. While one can construct valid variational bounds that rely only on matrix multiplications (matmuls) via an auxiliary matrix parameter, optimising them with off-the-shelf first-order methods is challenging. We make the inverse-free approach practical by proposing a better-conditioned bound and deriving a matmul-only natural-gradient update for the auxiliary parameter, markedly improving stability and convergence. We further provide simple heuristics, such as step-size schedules and stopping criteria, that make the overall optimisation routine fit seamlessly into existing workflows. Across regression and classification benchmarks, we demonstrate that our method 1) serves as a drop-in replacement in SVGP-based models (e.g., deep GPs), 2) recovers similar performance to traditional methods, and 3) can be faster than baselines when well tuned.

Autonomous agents increasingly interact with the web, yet most websites remain designed for human browsers -- a fundamental mismatch that the emerging ``Agentic Web'' must resolve. Agents must repeatedly browse pages, inspect DOMs, and reverse-engineer callable routes -- a process that is slow, brittle, and redundantly repeated across agents. We observe that every modern website already exposes internal APIs (sometimes called \emph{shadow APIs}) behind its user interface -- first-party endpoints that power the site's own functionality. We present Unbrowse, a shared route graph that transforms browser-based route discovery into a collectively maintained index of these callable first-party interfaces. The system passively learns routes from real browsing traffic and serves cached routes via direct API calls. In a single-host live-web benchmark of equivalent information-retrieval tasks across 94 domains, fully warmed cached execution averaged 950\,ms versus 3{,}404\,ms for Playwright browser automation (3.6$\times$ mean speedup, 5.4$\times$ median), with well-cached routes completing in under 100\,ms. A three-path execution model -- local cache, shared graph, or browser fallback -- ensures the system is voluntary and self-correcting. A three-tier micropayment model via the x402 protocol charges per-query search fees for graph lookups (Tier~3), a one-time install fee for discovery documentation (Tier~1), and optional per-execution fees for site owners who opt in (Tier~2). All tiers are grounded in a necessary condition for rational adoption: an agent uses the shared graph only when the total fee is lower than the expected cost of browser rediscovery.

We consider the problem of constructing surrogate operators for parameter-to-solution maps arising from parametric partial differential equations, where repeated forward model evaluations are computationally expensive. We present a systematic empirical comparison of neural operator surrogates, including a reduced-basis neural operator trained with $L^2_μ$ and $H^1_μ$ objectives and the Fourier neural operator, against polynomial surrogate methods, specifically a reduced-basis sparse-grid surrogate and a reduced-basis tensor-train surrogate. All methods are evaluated on a linear parametric diffusion problem and a nonlinear parametric hyperelasticity problem, using input fields with algebraically decaying spectral coefficients at varying rates of decay $s$. To enable fair comparisons, we analyze ensembles of surrogate models generated by varying hyperparameters and compare the resulting Pareto frontiers of cost versus approximation accuracy, decomposing cost into contributions from data generation, setup, and evaluation. Our results show that no single method is universally superior. Polynomial surrogates achieve substantially better data efficiency for smooth input fields ($s \geq 2$), with convergence rates for the sparse-grid surrogate in agreement with theoretical predictions. For rough inputs ($s \leq 1$), the Fourier neural operator displays the fastest convergence rates. Derivative-informed training consistently improves data efficiency over standard $L^2_μ$ training, providing a competitive alternative for rough inputs in the low-data regime when Jacobian information is available at reasonable cost. These findings highlight the importance of matching the surrogate methodology to the regularity of the problem as well as accuracy demands and computational constraints of the application.

We present OmniVoice, a massive multilingual zero-shot text-to-speech (TTS) model that scales to over 600 languages. At its core is a novel diffusion language model-style discrete non-autoregressive (NAR) architecture. Unlike conventional discrete NAR models that suffer from performance bottlenecks in complex two-stage (text-to-semantic-to-acoustic) pipelines, OmniVoice directly maps text to multi-codebook acoustic tokens. This simplified approach is facilitated by two key technical innovations: (1) a full-codebook random masking strategy for efficient training, and (2) initialization from a pre-trained LLM to ensure superior intelligibility. By leveraging a 581k-hour multilingual dataset curated entirely from open-source data, OmniVoice achieves the broadest language coverage to date and delivers state-of-the-art performance across Chinese, English, and diverse multilingual benchmarks. Our code and pre-trained models are publicly available at https://github.com/k2-fsa/OmniVoice.

Smart contract vulnerabilities can cause substantial financial losses due to the immutability of code after deployment. While existing tools detect vulnerabilities, they cannot effectively repair them. In this paper, we propose SCPatcher, a framework that combines retrieval-augmented generation with a knowledge graph for automated smart contract repair. We construct a knowledge graph from 5,000 verified Ethereum contracts, extracting function-level relationships to build a semantic network. This graph serves as an external knowledge base that enhances Large Language Model reasoning and enables precise vulnerability patching. We introduce a two-stage repair strategy, initial knowledge-guided repair followed by Chain-of-Thought reasoning for complex vulnerabilities. Evaluated on a diverse set of vulnerable contracts, SCPatcher achieves 81.5\% overall repair rate and 91.0\% compilation pass rate, substantially outperforming existing methods.

Successor Features (SF) combined with Generalized Policy Improvement (GPI) provide a robust framework for transfer learning in Reinforcement Learning (RL) by decoupling environment dynamics from reward functions. However, standard SF learning methods typically rely on semi-gradient Temporal Difference (TD) updates. When combined with non-linear function approximation, semi-gradient methods lack robust convergence guarantees and can lead to instability, particularly in the multi-task setting where accurate feature estimation is critical for effective GPI. Inspired by Full Gradient DQN, we propose Full-Gradient Successor Feature Representations Q-Learning (FG-SFRQL), an algorithm that optimizes the successor features by minimizing the full Mean Squared Bellman Error. Unlike standard approaches, our method computes gradients with respect to parameters in both the online and target networks. We provide a theoretical proof of almost-sure convergence for FG-SFRQL and demonstrate empirically that minimizing the full residual leads to superior sample efficiency and transfer performance compared to semi-gradient baselines in both discrete and continuous domains.

Mechanistic simulations typically assume fixed ontologies: variables, causal relationships, and resolution policies are static. This assumption fails when the true causal structure is contested or unidentifiable-as in antimicrobial resistance (AMR) spread, where contact, environmental, and selection ontologies compete. We introduce Procela, a Python framework where variables act as epistemic authorities that maintain complete hypothesis memory, mechanisms encode competing ontologies as causal units, and governance observes epistemic signals and mutates system topology at runtime. This is the first framework where simulations test their own assumptions. We instantiate Procela for AMR in a hospital network with three competing families. Governance detects coverage decay, policy fragility, and runs structural probes. Results show 20.4% error reduction and 69% cumulative regret improvement over baseline. All experiments are reproducible with full auditability. Procela establishes a new paradigm: simulations that model not only the world but their own modeling process, enabling adaptation under structural uncertainty.

TF-IDF is a classical formula that is widely used for identifying important terms within documents. We show that TF-IDF-like scores arise naturally from the test statistic of a penalized likelihood-ratio test setup capturing word burstiness (also known as word over-dispersion). In our framework, the alternative hypothesis captures word burstiness by modeling a collection of documents according to a family of beta-binomial distributions with a gamma penalty term on the precision parameter. In contrast, the null hypothesis assumes that words are binomially distributed in collection documents, a modeling approach that fails to account for word burstiness. We find that a term-weighting scheme given rise to by this test statistic performs comparably to TF-IDF on document classification tasks. This paper provides insights into TF-IDF from a statistical perspective and underscores the potential of hypothesis testing frameworks for advancing term-weighting scheme development.

This study examines the challenges of modeling complex and noisy data related to socioeconomic factors over time, with a focus on data from various districts in Odisha, India. Traditional time-series models struggle to capture both trends and variations together in this type of data. To tackle this, a Variational Neural Stochastic Differential Equation (V-NSDE) model is designed that combines the expressive dynamics of Neural SDEs with the generative capabilities of Variational Autoencoders (VAEs). This model uses an encoder and a decoder. The encoder takes the initial observations and district embeddings and translates them into a Gaussian distribution, which determines the mean and log-variance of the first latent state. Then the obtained latent state initiates the Neural SDE, which utilize neural networks to determine the drift and diffusion functions that govern continuous-time latent dynamics. These governing functions depend on the time index, latent state, and district embedding, which help the model learn the unique characteristics specific to each district. After that, using a probabilistic decoder, the observations are reconstructed from the latent trajectory. The decoder outputs a mean and log-variance for each time step, which follows the Gaussian likelihood. The Evidence Lower Bound (ELBO) training loss improves by adding a KL-divergence regularization term to the negative log-likelihood (nll). The obtained results demonstrate the effective learning of V-NSDE in recognizing complex patterns over time, yielding realistic outcomes that include clear trends and random fluctuations across different areas.

Diffusion language models (DLMs) offer a promising path toward low-latency generation through parallel decoding, but their practical efficiency depends heavily on the decoding trajectory. In practice, this advantage often fails to fully materialize because standard training does not provide explicit supervision over token reveal order, creating a train-inference mismatch that leads to suboptimal decoding behavior. We propose Trajectory-Ranked Instruction Masked Supervision (TRIMS), a simple trajectory-guided supervised fine-tuning framework that injects trajectory supervision into standard Masked Diffusion Language Model (MDLM) training with minimal overhead. Instead of relying on costly DLM-based distillation, TRIMS uses lightweight signals from an autoregressive teacher to guide a trajectory-aware masking strategy, encouraging the model to learn more effective decoding orders. Experiments on LLaDA and Dream across math and coding benchmarks show that TRIMS significantly improves the accuracy-parallelism trade-off over both standard MDLM training and train-free acceleration baselines, while achieving competitive performance with prior distillation-based approaches at substantially lower training cost. Further analysis shows that TRIMS leads to better decoding trajectories, validating the effectiveness of trajectory-guided supervision for DLMs.

Modern data warehouses extend SQL with semantic operators that invoke large language models on each qualifying row, but the per-row inference cost is prohibitive at scale. Model cascades reduce this cost by routing most rows through a fast proxy model and delegating uncertain cases to an expensive oracle. Existing frameworks, however, require global dataset access and optimize a single quality metric, limiting their applicability in distributed systems where data is partitioned across independent workers. We present two adaptive cascade algorithms designed for streaming, per-partition execution in which each worker processes its partition independently without inter-worker communication. SUPG-IT extends the SUPG statistical framework to streaming execution with iterative threshold refinement and joint precision-recall guarantees. GAMCAL replaces user-specified quality targets with a learned calibration model: a Generalized Additive Model maps proxy scores to calibrated probabilities with uncertainty quantification, enabling direct optimization of a cost-quality tradeoff through a single parameter. Experiments on six datasets in a production semantic SQL engine show that both algorithms achieve F1 > 0.95 on every dataset. GAMCAL achieves higher F1 per oracle call at cost-sensitive operating points, while SUPG-IT reaches a higher quality ceiling with formal guarantees on precision and recall.

Smart contracts are self-executing programs that manage financial transactions on blockchain networks. Developers commonly rely on third-party code libraries to improve both efficiency and security. However, improper use of these libraries can introduce hidden vulnerabilities that are difficult to detect, leading to significant financial losses. Existing automated tools struggle to identify such misuse because it often requires understanding the developer's intent rather than simply scanning for known code patterns. This paper presents LibScan, an automated detection framework that combines large language model (LLM)-based semantic reasoning with rule-based code analysis, identifying eight distinct categories of library misuse in smart contracts. To improve detection reliability, the framework incorporates an iterative self-correction mechanism that refines its analysis across multiple rounds, alongside a structured knowledge base derived from large-scale empirical studies of real-world misuse cases. Experiments conducted on 662 real-world smart contracts demonstrate that LibScan achieves an overall detection accuracy of 85.15\%, outperforming existing tools by a margin of over 16 percentage points. Ablation experiments further confirm that combining both analysis approaches yields substantially better results than either method used independently.

Predictive process monitoring (PPM) focuses on predicting future process trajectories, including next activity predictions. This is crucial in dynamic environments where processes change or face uncertainty. However, current frameworks often assume a static environment, overlooking dynamic characteristics and concept drifts. This results in catastrophic forgetting, where training while focusing merely on new data distribution negatively impacts the performance on previously learned data distributions. Continual learning addresses, among others, the challenges related to mitigating catastrophic forgetting. This paper proposes a novel approach called Continual Next Activity Prediction with Prompts (CNAPwP), which adapts the DualPrompt algorithm for next activity prediction to improve accuracy and adaptability while mitigating catastrophic forgetting. We introduce new datasets with recurring concept drifts, alongside a task-specific forgetting metric that measures the prediction accuracy gap between initial occurrence and subsequent task occurrences. Extensive testing on three synthetic and two real-world datasets representing several setups of recurrent drifts shows that CNAPwP achieves SOTA or competitive results compared to five baselines, demonstrating its potential applicability in real-world scenarios. An open-source implementation of our method, together with the datasets and results, is available at: https://github.com/SvStraten/CNAPwP.

We propose a novel extension of the Bradley-Terry model to multiplayer games and adapt a recent algorithm by Newman [1] to our model. We demonstrate the use of our proposed method on synthetic datasets and on a real dataset of games of cards.

Poverty is a complex dynamic challenge that cannot be adequately captured using predefined differential equations. Nowadays, artificial machine learning (ML) methods have demonstrated significant potential in modelling real-world dynamical systems. Among these, Neural Ordinary Differential Equations (Neural ODEs) have emerged as a powerful, data-driven approach for learning continuous-time dynamics directly from observations. This chapter applies the Neural ODE framework to analyze poverty dynamics in the Indian state of Odisha. Specifically, we utilize time-series data from 2007 to 2020 on the key indicators of economic development and poverty reduction. Within the Neural ODE architecture, the temporal gradient of the system is represented by a multi-layer perceptron (MLP). The obtained neural dynamical system is integrated using a numerical ODE solver to obtain the trajectory of over time. In backpropagation, the adjoint sensitivity method is utilized for gradient computation during training to facilitate effective backpropagation through the ODE solver. The trained Neural ODE model reproduces the observed data with high accuracy. This demonstrates the capability of Neural ODE to capture the dynamics of the poverty indicator of concrete-structured households. The obtained results show that ML methods, such as Neural ODEs, can serve as effective tools for modeling socioeconomic transitions. It can provide policymakers with reliable projections, supporting more informed and effective decision-making for poverty alleviation.

Knowledge distillation has become a primary mechanism for transferring reasoning and domain expertise from frontier Large Language Models (LLMs) to smaller, deployable students. However, the dominant paradigm remains \textit{off-policy}: students train on static teacher-generated data and never encounter their own errors during learning. This train--test mismatch, an instance of \textit{exposure bias}, causes prediction errors to compound autoregressively at inference time. On-Policy Distillation (OPD) addresses this by letting the student generate its own trajectories and receive teacher feedback on these self-generated outputs, grounding distillation in the theory of interactive imitation learning. Despite rapid growth spanning divergence minimization, reward-guided learning, and self-play, the OPD literature remains fragmented with no unified treatment. This survey provides the first comprehensive overview of OPD for LLMs. We introduce a unified $f$-divergence framework over on-policy samples and organize the landscape along three orthogonal dimensions: \emph{feedback signal} (logit-based, outcome-based, or self-play), \emph{teacher access} (white-box, black-box, or teacher-free), and \emph{loss granularity} (token-level, sequence-level, or hybrid). We systematically analyze representative methods, examine industrial deployments, and identify open problems including distillation scaling laws, uncertainty-aware feedback, and agent-level distillation.

We present KUTED, a speech-to-text translation (S2TT) dataset for Central Kurdish, derived from TED and TEDx talks. The corpus comprises 91,000 sentence pairs, including 170 hours of English audio, 1.65 million English tokens, and 1.40 million Central Kurdish tokens. We evaluate KUTED on the S2TT task and find that orthographic variation significantly degrades Kurdish translation performance, producing nonstandard outputs. To address this, we propose a systematic text standardization approach that yields substantial performance gains and more consistent translations. On a test set separated from TED talks, a fine-tuned Seamless model achieves 15.18 BLEU, and we improve Seamless baseline by 3.0 BLEU on the FLEURS benchmark. We also train a Transformer model from scratch and evaluate a cascaded system that combines Seamless (ASR) with NLLB (MT).

Despite extensions to speech inputs, effectively leveraging the rich knowledge and contextual understanding of large language models (LLMs) in automatic speech recognition (ASR) remains non-trivial, as the task primarily involves direct speech-to-text mapping. To address this, this paper proposes chain-of-thought ASR (CoT-ASR), which constructs a reasoning chain that enables LLMs to first analyze the input speech and generate contextual analysis, thereby fully exploiting their generative capabilities. With this contextual reasoning, CoT-ASR then performs more informed speech recognition and completes both reasoning and transcription in a single pass. Moreover, CoT-ASR naturally supports user-guided transcription: while designed to self-generate reasoning, it can also seamlessly incorporate user-provided context to guide transcription, further extending ASR functionality. To reduce the modality gap, this paper introduces a CTC-guided Modality Adapter, which uses CTC non-blank token probabilities to weight LLM embeddings, efficiently aligning speech encoder outputs with the LLM's textual latent space. Experiments show that, compared to standard LLM-based ASR, CoT-ASR achieves a relative reduction of 8.7% in word error rate (WER) and 16.9% in entity error rate (EER).

Predicting the behavior of ultra-large complex systems, from climate to biological and technological networks, is a central unsolved challenge. Existing approaches face a fundamental trade-off: equation discovery methods provide interpretability but fail to scale, while neural networks scale but operate as black boxes and often lose reliability over long times. Here, we introduce the Sparse Identification Graph Neural Network, a framework that overcome this divide by allowing to infer the governing equations of large networked systems from data. By defining symbolic discovery as edge-level information, SIGN decouples the scalability of sparse identification from network size, enabling efficient equation discovery even in large systems. SIGN allows to study networks with over 100,000 nodes while remaining robust to noise, sparse sampling, and missing data. Across diverse benchmark systems, including coupled chaotic oscillators, neural dynamics, and epidemic spreading, it recovers governing equations with high precision and sustains accurate long-term predictions. Applied to a data set of time series of temperature measurements in 71,987 sea surface positions, SIGN identifies a compact predictive network model and captures large-scale sea surface temperature conditions up to two years in advance. By enabling equation discovery at previously inaccessible scales, SIGN opens a path toward interpretable and reliable prediction of real-world complex systems.

As the focus in LLM-based coding shifts from static single-step code generation to multi-step agentic interaction with tools and environments, understanding which tasks will challenge agents and why becomes increasingly difficult. This is compounded by current practice: agent performance is typically measured by aggregate pass rates on benchmarks, but single-number metrics obscure the diversity of tasks within a benchmark. We present a framework for predicting success or failure on individual tasks tailored to the agentic coding regime. Our approach augments Item Response Theory (IRT) with rich features extracted from tasks, including issue statements, repository contexts, solutions, and test cases, and introduces a novel decomposition of agent ability into LLM and scaffold ability components. This parameterization enables us to aggregate evaluation data across heterogeneous leaderboards and accurately predict task-level performance for unseen benchmarks, as well as unseen LLM-scaffold combinations. Our methods have practical utility for benchmark designers, who can better calibrate the difficulty of their new tasks without running computationally expensive agent evaluations.

In recent years, the scaling laws of recommendation models have attracted increasing attention, which govern the relationship between performance and parameters/FLOPs of recommenders. Currently, there are three mainstream architectures for achieving scaling in recommendation models, namely attention-based, TokenMixer-based, and factorization-machine-based methods, which exhibit fundamental differences in both design philosophy and architectural structure. In this paper, we propose a unified scaling architecture for recommendation systems, namely \textbf{UniMixer}, to improve scaling efficiency and establish a unified theoretical framework that unifies the mainstream scaling blocks. By transforming the rule-based TokenMixer to an equivalent parameterized structure, we construct a generalized parameterized feature mixing module that allows the token mixing patterns to be optimized and learned during model training. Meanwhile, the generalized parameterized token mixing removes the constraint in TokenMixer that requires the number of heads to be equal to the number of tokens. Furthermore, we establish a unified scaling module design framework for recommender systems, which bridges the connections among attention-based, TokenMixer-based, and factorization-machine-based methods. To further boost scaling ROI, a lightweight UniMixing module is designed, \textbf{UniMixing-Lite}, which further compresses the model parameters and computational cost while significantly improve the model performance. The scaling curves are shown in the following figure. Extensive offline and online experiments are conducted to verify the superior scaling abilities of \textbf{UniMixer}.

As Large Language Model (LLM) capabilities advance, the demand for high-quality annotation of exponentially increasing text corpora has outpaced human capacity, leading to the widespread adoption of LLMs in automatic evaluation and annotation. However, proprietary LLMs often exhibit systematic biases that diverge from human expert consensus, lacks reproducibility, and raises data privacy concerns. Our work examines the viability of finetuning a quantized Small Language Model of 1.7B parameter size on limited human-annotated data to serve as a highly aligned, deterministic evaluator and annotator. By implementing a custom, multi-dimensional rubric framework and simple augmentation and regularization techniques, the proposed approach achieves higher inter-annotator agreement (0.23 points increase in Krippendorff's $α$) than the best performing state-of-the-art proprietary LLM. We also demonstrate the generalizability of the proposed training pipeline on a separate emotion classification task. The results show that task-specific alignment and efficient 4-bit quantized fine-tuning provide superior open-source alternative to using proprietary models for evaluation and annotation. Our finetuning approach is publicly available at https://github.com/jylee-k/slm-judge.

Molecular dynamics simulations provide detailed trajectories at the atomic level, but extracting interpretable and robust insights from these high-dimensional data remains challenging. In practice, analyses typically rely on a single representation. Here, we show that representation choice is not neutral: it fundamentally shapes the conformational organization, similarity relationships, and apparent transitions inferred from identical simulation data. To complement existing representations, we introduce Orientation features, a geometrically grounded, rotation-aware encoding of protein backbone. We compare it against common descriptions across three dynamical regimes: fast-folding proteins, large-scale domain motions, and protein-protein association. Across these systems, we find that different representations emphasize complementary aspects of conformational space, and that no single representation provides a complete picture of the underlying dynamics. To facilitate systematic comparison, we developed ManiProt, a library for efficient computation and analysis of multiple protein representations. Our results motivate a comparative, representation-aware framework for the interpretation of molecular dynamics simulations.

In enhancing the fairness of Large Language Models (LLMs), evaluating social biases rooted in the cultural contexts of specific linguistic regions is essential. However, most existing Japanese benchmarks heavily rely on translating English data, which does not necessarily provide an evaluation suitable for Japanese culture. Furthermore, they only evaluate bias in the conclusion, failing to capture biases lurking in the reasoning. In this study, based on attribution theory in social psychology, we constructed a new dataset, ``JUBAKU-v2,'' which evaluates the bias in attributing behaviors to in-groups and out-groups within reasoning while fixing the conclusion. This dataset consists of 216 examples reflecting cultural biases specific to Japan. Experimental results verified that it can detect performance differences across models more sensitively than existing benchmarks.

The impending arrival of cryptographically relevant quantum computers (CRQCs) threatens the security foundations of modern software: Shor's algorithm breaks RSA, ECDSA, ECDH, and Diffie-Hellman, while Grover's algorithm reduces the effective security of symmetric and hash-based schemes. Despite NIST standardising post-quantum cryptography (PQC) in 2024 (FIPS 203 ML-KEM, FIPS 204 ML-DSA, FIPS 205 SLH-DSA), most codebases lack automated tooling to inventory classical cryptographic usage and prioritise migration based on quantum risk. We present Quantum-Safe Code Auditor, a quantum-aware static analysis framework that combines (i) regex-based detection of 15 classes of quantum-vulnerable primitives, (ii) LLM-assisted contextual enrichment to classify usage and severity, and (iii) risk scoring via a Variational Quantum Eigensolver (VQE) model implemented in Qiskit 2.x, incorporating qubit-cost estimates to prioritise findings. We evaluate the system across five open-source libraries -- python-rsa, python-ecdsa, python-jose, node-jsonwebtoken, and Bouncy Castle Java -- covering 5,775 findings. On a stratified sample of 602 labelled instances, we achieve 71.98% precision, 100% recall, and an F1 score of 83.71%. All code, data, and reproduction scripts are released as open-source.

The generalization ability of imitation learning policies for robotic manipulation is fundamentally constrained by the diversity of expert demonstrations, while collecting demonstrations across varied environments is costly and difficult in practice. In this paper, we propose a practical framework that exploits inherent scene diversity without additional human effort by scaling camera views during demonstration collection. Instead of acquiring more trajectories, multiple synchronized camera perspectives are used to generate pseudo-demonstrations from each expert trajectory, which enriches the training distribution and improves viewpoint invariance in visual representations. We analyze how different action spaces interact with view scaling and show that camera-space representations further enhance diversity. In addition, we introduce a multiview action aggregation method that allows single-view policies to benefit from multiple cameras during deployment. Extensive experiments in simulation and real-world manipulation tasks demonstrate significant gains in data efficiency and generalization compared to single-view baselines. Our results suggest that scaling camera views provides a practical and scalable solution for imitation learning, which requires minimal additional hardware setup and integrates seamlessly with existing imitation learning algorithms. The website of our project is https://yichen928.github.io/robot_multiview.

Housing selection is a high-stakes and largely irreversible decision problem. We study housing consultation as a decision-support interface for housing selection. Existing housing platforms and many LLM-based assistants often reduce this process to ranking or recommendation, resulting in opaque reasoning, brittle multi-constraint handling, and limited guarantees on factuality. We present HabitatAgent, the first LLM-powered multi-agent architecture for end-to-end housing consultation. HabitatAgent comprises four specialized agent roles: Memory, Retrieval, Generation, and Validation. The Memory Agent maintains multi-layer user memory through internal stages for constraint extraction, memory fusion, and verification-gated updates; the Retrieval Agent performs hybrid vector--graph retrieval (GraphRAG); the Generation Agent produces evidence-referenced recommendations and explanations; and the Validation Agent applies multi-tier verification and targeted remediation. Together, these agents provide an auditable and reliable workflow for end-to-end housing consultation. We evaluate HabitatAgent on 100 real user consultation scenarios (300 multi-turn question--answer pairs) under an end-to-end correctness protocol. A strong single-stage baseline (Dense+Rerank) achieves 75% accuracy, while HabitatAgent reaches 95%.

Enterprise adoption of Large Language Models (LLMs) is constrained by hallucination, domain drift, and the inability to enforce regulatory compliance at the reasoning level. We present a neurosymbolic architecture implemented within the Foundation AgenticOS (FAOS) platform that addresses these limitations through ontology-constrained neural reasoning. Our approach introduces a three-layer ontological framework--Role, Domain, and Interaction ontologies--that provides formal semantic grounding for LLM-based enterprise agents. We formalize the concept of asymmetric neurosymbolic coupling, wherein symbolic ontological knowledge constrains agent inputs (context assembly, tool discovery, governance thresholds) while proposing mechanisms for extending this coupling to constrain agent outputs (response validation, reasoning verification, compliance checking). We evaluate the architecture through a controlled experiment (600 runs across five industries: FinTech, Insurance, Healthcare, Vietnamese Banking, and Vietnamese Insurance), finding that ontology-coupled agents significantly outperform ungrounded agents on Metric Accuracy (p < .001, W = .460), Regulatory Compliance (p = .003, W = .318), and Role Consistency (p < .001, W = .614), with improvements greatest where LLM parametric knowledge is weakest--particularly in Vietnam-localized domains. Our contributions include: (1) a formal three-layer enterprise ontology model, (2) a taxonomy of neurosymbolic coupling patterns, (3) ontology-constrained tool discovery via SQL-pushdown scoring, (4) a proposed framework for output-side ontological validation, (5) empirical evidence for the inverse parametric knowledge effect that ontological grounding value is inversely proportional to LLM training data coverage of the domain, and (6) a production system serving 21 industry verticals with 650+ agents.

The scenario approach provides a powerful data-driven framework for designing solutions under uncertainty with rigorous probabilistic robustness guarantees. Existing theory, however, primarily addresses assessing robustness with respect to a single appropriateness criterion for the solution based on a dataset, whereas many practical applications - including multi-agent decision problems - require the simultaneous consideration of multiple criteria and the assessment of their robustness based on multiple datasets, one per criterion. This paper develops a general scenario theory for multi-criteria data-driven decision making. A central innovation lies in the collective treatment of the risks associated with violations of individual criteria, which yields substantially more accurate robustness certificates than those derived from a naive application of standard results. In turn, this approach enables a sharper quantification of the robustness level with which all criteria are simultaneously satisfied. The proposed framework applies broadly to multi-criteria data-driven decision problems, providing a principled, scalable, and theoretically grounded methodology for design under uncertainty.