跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+758新增论文
+578其中 AI 相关
+94已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-17 04:58 · 全站最近入库:2026-11-17 04:58各信源进度 →

统计口径与数据新鲜度

统计 2026-11-16 05:30 — 2026-11-17 05:30(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-17 04:58最近发现新增版本:2026-11-17 04:58库内论文最新更新时间:2026-11-14 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

24,106 已收录论文20,158 AI 相关7,049 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 24,106 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-17 04:57(北京时间)
版本更新 · v3arXiv:2605.11928

AgentRobustBench: A Sim-to-Real Benchmark and Domain-Randomized RL Recipe for Tool-Use Agents

Xiaolin Zhou, Aojie Yuan, Zheng Luo, Zipeng Ling, Xixiao Pan, Yicheng Gao, Haiyue Zhang, Jiate Li 等 15 位作者

原文摘要

Tool-use language agents are evaluated on benchmarks that assume clean inputs, unambiguous tool registries, and reliable APIs. Real deployments violate all these assumptions: user typos propagate into hallucinated tool names, a misconfigured request timeout can stall an agent indefinitely, and duplicate tool names across servers can freeze an SDK. We study these failures as a sim-to-real gap in the tool-use partially observable Markov decision process (POMDP), where deployment noise enters through the observation, action space, reward-relevant metadata, or transition dynamics. We introduce AgentRobustBench, a benchmark with 22 perturbation types organized by these four POMDP components, each grounded in a verified GitHub issue or documented tool-calling failure. Across 21 models from 1.5B to 32B parameters (including the closed-source o4-mini), the robustness profile is sharply uneven: observation perturbations reduce accuracy by less than 5%, while reward-relevant and transition perturbations reduce accuracy by roughly 40% and 30%, respectively; scale alone does not close these gaps. We then propose ToolRL-DR, a domain-randomization reinforcement learning (RL) recipe that trains a tool-use agent on perturbation-augmented trajectories spanning the three statically encodable POMDP components. On a 3B backbone, ToolRL-DR-Full retains roughly three-quarters of clean accuracy and reaches an aggregate perturbed accuracy comparable to open-source 14B function-calling baselines while substantially narrowing the gap to o4-mini. It closes approximately 27% of the Transition gap despite never seeing transition perturbations in training, suggesting that RL on adversarial static tool-use inputs induces a more persistent retry policy that transfers to unseen runtime failures. Code, dataset, checkpoints, and leaderboard are accessible through our GitHub repository.

cs.AI

版本更新 · v3arXiv:2608.17070

Certified but Private: Scalable Zero-Knowledge Proofs for Neural Network Guarantees

Youwei Zhong, Ben Merbaum, Timos Antonopoulos, Ning Luo, Charalampos Papamanthou, Katerina Sotiraki, Ruzica Piskac

原文摘要

With the growing deployment of machine learning models, formal guarantees of the robustness and fairness of these models have become increasingly important in safety-critical and legal-compliance settings. However, model parameters are often commercial secrets that cannot be disclosed to auditors or end users. To this end, we present PANDA, a scalable system that uses zero-knowledge proofs (ZKPs) to prove the robustness and fairness properties of a model without revealing its private parameters. PANDA is built on top of CROWN, an efficient robustness certification framework that is used in many state-of-the-art formal verification tools for neural networks. The core contribution of PANDA is a novel algorithm for proving linear relaxation bounds for non-linear activation layers, yielding simple, lightweight proofs. Remarkably, our implementation can generate proofs of local robustness for neural networks with more than 2.9M parameters in 5 minutes, and can verify them in 10 seconds. Prior ZKP-based robustness system rely on exponential-time algorithms that cannot scale to nontrivial networks. In contrast, PANDA scales polynomially in the number of neurons in a network, allowing us to support neural networks 4 orders of magnitude larger than previous approaches with significantly reduced prover overhead.

cs.LG · cs.CR · cs.LO

版本更新 · v2arXiv:2611.02383

Toward an Evidence-Based Template for Agent Instruction Files: An Empirical Study of Adoption, Consistency, and Maintenance

Mehedi Hasan Shanto, Md. Anaytul Islam, Muhammad Asaduzzaman, Alioune Ngom

原文摘要

AI coding agents rely on repository-level instruction files to understand project conventions, development workflows, and constraints. However, different agents support different instruction formats, and limited empirical evidence exists about how these files are adopted, what information they contain, whether coexisting formats provide consistent guidance, and how they are maintained. We conduct an empirical study of 3,387 instruction files from 1,000 GitHub repositories across four formats: AGENTS.md, CLAUDE.md, GitHub Copilot instructions, and Cursor rules. We find that 42.2% of instruction-bearing repositories in the higher-popularity cohort maintain multiple formats. Among 386 manually examined pairs of corresponding sections, 16.1% provide divergent guidance. Our maintenance analysis further shows that 40.4% of the analyzed files have only one recorded modification event and that maintenance patterns differ across formats. Based on category prevalence and category addition ratios (CAR), we derive an evidence-based template containing ten Required, four Gap, and two Recommended categories. In an exploratory generation experiment, explicit template guidance increases structural category coverage from 18.7% to 82.8%. We further introduce AGIT-Instruct, a benchmark for evaluating instruction-file generation through structural coverage and content alignment. Across four generation configurations, retrieval-augmented generation achieves the highest full-file ROUGE-1 F1 of 0.459. A manual assessment of 278 generated category instances finds that 19.4% fully preserve and 61.2% partially preserve developer-written intent. Our findings provide empirical guidance for organizing and maintaining agent instruction

cs.SE

版本更新 · v4arXiv:2503.05970

Partially Decentralized Multi-Agent Q-Learning via Digital Cousins for Wireless Networks

Talha Bozkus, Urbashi Mitra

原文摘要

Q-learning is a widely used reinforcement learning (RL) algorithm for optimizing wireless networks, but faces challenges with large state-spaces. Recently proposed multi-environment mixed Q-learning (MEMQ) algorithm addresses these challenges by employing multiple Q-learning algorithms across multiple synthetically generated, distinct but structurally related environments, so-called digital cousins. In this paper, we propose a novel multi-agent MEMQ (M-MEMQ) for cooperative decentralized wireless networks with multiple networked transmitters (TXs) and base stations (BSs). TXs do not have access to global information (joint state and actions). The new concept of coordinated and uncoordinated states is introduced. In uncoordinated states, TXs act independently to minimize their individual costs and update local Q-functions. In coordinated states, TXs use a Bayesian approach to estimate the joint state and update the joint Q-functions. The cost of information-sharing scales linearly with the number of TXs and is independent of the joint state-action space size. Several theoretical guarantees, including deterministic and probabilistic convergence, bounds on estimation error variance, and the probability of misdetecting the joint states, are given. Numerical simulations show that M-MEMQ outperforms several decentralized and centralized training with decentralized execution (CTDE) multi-agent RL algorithms by achieving 60% lower average policy error (APE), 40% faster convergence, 45% reduced runtime complexity, and 40% less sample complexity. Furthermore, M-MEMQ achieves comparable APE with significantly lower complexity than centralized methods. Simulations validate the theoretical analyses.

cs.LG · eess.SP

新论文 · v1arXiv:2611.10498

EmbodiedIntentBench: Benchmarking Intent-Driven Runtime Adaptation for Mobile Robotic Missions

Rajat Bhattacharjya, Changwoon Lee, Hyeonjong Ju, Sing-Yao Wu, Hamidreza Alikhani, Zhuoyi Yang, Ian Harris, Nalini Venkatasubramanian 等 9 位作者

原文摘要

Mobile robotic sensing systems must adapt to operator requirements, limited resources, changing connectivity, and observation opportunities shaped by physical motion. Evaluating runtime strategies therefore requires assessing whether their decisions fulfill mission objectives, beyond isolated perception accuracy or execution efficiency. We present EmbodiedIntentBench, a benchmark for intent-driven runtime adaptation in mobile robotic missions. It represents mission intent through structured contracts for coverage, temporal evidence, reporting, urgency, and resources, and couples simulated unmanned aerial vehicle (UAV) workloads with causal, resource-accounted execution. Computation decisions affect mission time, resource use, and which observations can contribute evidence. We evaluate fixed and adaptive perception strategies in multi-phase missions and under changing urgency, connectivity, and battery conditions. In the multi-phase Depot study, a configuration with 87.5% full-bank target recall across visible target observations completes only 1 of 16 missions, whereas a lower-recall configuration completes 14. In an ablation, ignoring execution-induced observation loss increases aggregate mission success and reverses the ordering of two runtime strategies. Replays using Jetson AGX Orin CPU and GPU profiles change the relative success of local and remote strategies. In selected Jetson GPU-profile replays, intent-driven adaptation reduces modeled task energy by up to 28% with unchanged mission success. EmbodiedIntentBench provides an extensible framework for evaluating runtime decisions against mission fulfillment under physical and resource constraints.

cs.RO · cs.AR · cs.CV · cs.DC · cs.LG

新论文 · v1arXiv:2611.10497

Disentangling Channel Semantics in Vision Transformers via Token Decorrelation and Composition-Aware Modulation

Daeun Kim, Hyesong Choi, Hyejin Park, Dongbo Min

原文摘要

Multi-channel images encode heterogeneous channel semantics aligned over a shared spatial structure, which requires models to jointly capture globally shared structure and channel-specific variations. Recent multi-channel Vision Transformers attempt to address this challenge by augmenting the [CLS] token with memory tokens to organize multi-channel representations under varying channel compositions. However, we find that these context tokens collapse onto a few dominant channels under high inter-channel redundancy, producing redundant representations and a biased global summary that fails to reliably support composition-aware channel interpretation. We propose MuCa-ViT (Multi-Channel Composition-Aware Vision Transformer), a unified framework that addresses these limitations through two coupled mechanisms. Factorial Token Learning (FTL) enforces token-level decorrelation among context tokens, encouraging the [CLS] token to capture globally shared semantics while memory tokens preserve complementary channel-specific cues. Channel Composition-Aware Modulation (CAM) uses the FTL-refined [CLS] as a semantic anchor to generate sample-wise modulation signals, which enable composition-aware interpretation of channels within a unified backbone rather than relying on static channel embeddings. Experiments on JUMP-CP, CHAMMI, and So2Sat show that MuCa-ViT outperforms prior multi-channel ViTs across all three benchmarks, and the magnitude of improvement aligns with the inter-channel redundancy of each dataset ($ρ\in [0.19, 0.77]$), validating the diagnostic analysis that motivates our design. Our code is available at https://github.com/kimdanni/MuCa-ViT .

cs.CV

版本更新 · v2arXiv:2605.03780

Task Vector Geometry Underlies Dual Modes of Task Inference in Transformers

Hao Yan, Haolin Yang, Yiqiao Zhong

原文摘要

Transformers are effective at inferring the latent task from context via two inference modes: recognizing a task seen during training, and adapting to a novel one. Recent interpretability studies have identified from middle-layer representations task-specific directions, or task vectors, that steer model behavior. However, a lack of rigorous foundations hinders connecting internal representations to external model behavior: existing work fails to explain how task-vector geometry is shaped by the training distribution, and what geometry enables out-of-distribution (OOD) generalization. In this paper, we study these questions in a controlled synthetic setting by training small transformers from scratch on latent-task sequence distributions, which allows a principled mathematical characterization. We show that two inference modes can coexist within a single model. In-distribution behavior is governed by Bayesian task retrieval, implemented internally through convex combinations of learned task vectors. OOD behavior, by contrast, arises through extrapolative task learning, whose representations occupy a subspace nearly orthogonal to the task-vector subspace. Taken together, our results suggest that task-vector geometry, training distributions, and generalization behaviors are closely related.

cs.LG · cs.CL · stat.ML

版本更新 · v5arXiv:2607.25875

EvoST: Adapting Traffic Forecasters to Evolving Sensor Graphs

Du Yin, Xiachong Lin, Yue Tan, Ruiyi Yang, Jinliang Deng, Estrid He, Hao Xue, Flora D. Salim

原文摘要

Web-based navigation and traffic information services depend on accurate forecasts as sensor networks evolve. On checkpoints reused years later, updating only shared weights or adding a low-rank node prompt leaves substantial degradation, whereas, in the evaluated architectures, refreshing node identities recovers most of it while the shared spatio-temporal weights are reused. We propose EvoST, a framework that maintains traffic forecasters by refreshing node identities while keeping shared weights fixed. EvoST updates existing identities using recently released labels and initializes new sensors with identities borrowed from behaviorally similar established sensors. Once enough delayed labels arrive, it jointly refits all identities, and a per-sensor recursive least-squares calibrator corrects forecasts as labels become available. Across twelve PeMS region-year settings, identity refresh reduces the relative one-hour MAE gap between five-year-old STAEFormer checkpoints and target-year models from 17.7% to 2.5%, with all forecasters calibrated, and its MAE is 0.7% above that of full-parameter fine-tuning on the same data. In Graph WaveNet, identity refresh recovers 88% of the five-year MAE gap to calibrated target-year models. Similar recovery holds for STAEFormer checkpoints up to fifteen years old, while refreshing STID recovers 67% of its five-year gap. On an annual benchmark covering nine EvoXXLTraffic districts and Transport for NSW, EvoST has the lowest average MAE, RMSE, and MAPE among the reported non-foundation methods on all ten networks, and calibration accounts for most of its Avg-MAE reduction from the matched frozen checkpoints. Code and the extended version: https://github.com/lixus7/EvoST

cs.LG · cs.AI

版本更新 · v2arXiv:2605.05851

Language models lack systematicity in generating and updating hypotheses

Hua-Dong Xiong

原文摘要

Large language models (LLMs) are increasingly applied to complex problems, from automated code debugging to science itself. Such problems require generating plausible hypotheses from sparse, noisy, and ambiguous data and updating them as more information arrives. Yet how closely LLMs approach optimal inductive reasoning remains unclear. We study this question in the classic number game, in which a learner infers a concept from a few ambiguous positive integers: $\{2, 16, 4, 8\}$, for example, is explained both by ``powers of 2'' and by ``numbers up to 20.'' We probe the concepts LLMs hypothesize from the observed examples and use their behavioral patterns to understand how they generate and update hypotheses. We measure the LLM posterior over hypotheses in three tasks (prediction, evaluation, and generation) and compare LLM behavior with a Bayesian model and with humans. LLMs favor narrower hypotheses than humans and show a robust evaluation--generation gap: in evaluation they select hypotheses more consistent with the observed examples, whereas in generation they propose narrower hypotheses that are more often rules. They also update poorly: as examples accumulate, their accuracy declines while their confidence grows. Thinking reduces, and can even reverse, the gap; it also broadens the hypotheses LLMs favor, raising accuracy while lowering confidence. Whereas prior studies have shown that LLMs are good hypothesis proposers, our results and further comparisons with the Bayesian model show that LLM inductive reasoning lacks coherence across the three tasks, suggesting that LLMs do not maintain a stable representation shared across tasks that make the same demand. Our results highlight a limitation of LLMs in inductive inference, particularly in settings such as science, where hypotheses must go beyond the data.

cs.LG

版本更新 · v2arXiv:2609.19134

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Koutian Wu, Zihan Zhou, Yuanbo Pang, Weihao Liu 等 63 位作者

原文摘要

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B/9B/4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE

cs.CL · cs.CY

版本更新 · v4arXiv:2502.11371

RAG vs. GraphRAG: A Systematic Evaluation and Key Insights

Haoyu Han, Li Ma, Yu Wang, Harry Shomer, Yongjia Lei, Zhisheng Qi, Kai Guo, Zhigang Hua 等 12 位作者

原文摘要

Retrieval-Augmented Generation (RAG) improves large language models (LLMs) by retrieving relevant information from external sources and has been widely adopted for text-based tasks. For structured data, such as knowledge graphs, Graph Retrieval-Augmented Generation (GraphRAG) retrieves and aggregates information along graph structures. More recently, GraphRAG has been extended to general text settings by organizing unstructured text into graph representations, showing promise for reasoning and grounding. Despite these advances, existing GraphRAG systems for text data are often tailored to specific tasks, datasets, and system designs, resulting in heterogeneous evaluation protocols. Consequently, a systematic understanding of the relative strengths, limitations, and trade-offs between RAG and GraphRAG on widely used text benchmarks remains limited. In this paper, we present a comprehensive benchmark study comparing RAG and GraphRAG on established text-based tasks, including question answering and query-based summarization. We introduce a unified evaluation protocol that standardizes data preprocessing, retrieval configurations, and generation settings, enabling fair and reproducible comparisons. Our results highlight the distinct strengths of RAG and GraphRAG across different tasks and evaluation perspectives. Building on these findings, we explore selection and integration strategies that combine the strengths of both paradigms, leading to consistent performance improvements. We further analyze failure modes, efficiency trade-offs, and evaluation biases, and highlight key considerations for designing and evaluating retrieval-augmented generation systems.

cs.IR

版本更新 · v2arXiv:2608.10187

ConnectionMind: Leveraging Social Networks and Large Language Models for Personalized Recommendation at Meta

Haoyu Han, Yuming Liu, Lei Huang, Lizhu Zhang, Jiliang Tang, Xiangjun Fan

原文摘要

Modern recommendation systems on social media platforms such as Meta must model complex social relationships, including friendships, group memberships, and creator interactions, alongside massive and heterogeneous content such as text and video. Traditional recommendation models, however, often omit these signals or treat them independently, lacking the reasoning capability to integrate multi-relational context for fine-grained personalization. We present ConnectionMind, a production-ready recommendation framework that tightly integrates the social network structure with large language models (LLMs) to enable scalable, interpretable, and reasoning-aware personalization in Meta. ConnectionMind constructs a heterogeneous graph connecting users, items, friends, groups, and creator pages, and formulates recommendation as a graph reasoning problem: discovering personalized paths from users to candidate items. An LLM-based policy is employed to reason over these graph structures and guide recommendation decisions. To train the system at scale, ConnectionMind adopts a two-stage learning strategy. We first perform supervised fine-tuning (SFT) on large-scale user-item interaction trajectories to initialize the reasoning policy, followed by end-to-end reinforcement learning (RL) to refine the model's ability to reason over social graphs for personalized recommendation. Extensive experiments on multiple real-world datasets demonstrate the effectiveness of ConnectionMind compared to representative baselines. More importantly, ConnectionMind has been deployed in Meta's large-scale recommendation pipeline and has been evaluated through online A/B tests, achieving a 0.43% improvement in video watch time. These results demonstrate measurable real-world impact in a production recommendation system.

cs.IR · cs.SI

版本更新 · v2arXiv:2611.07520

Graph learning-assisted logic-based Benders decomposition for distributionally robust in-station train rescheduling under disruptions

Xiaobin Li, Yanbin Gao, Jianing Long, Wuming Lei

原文摘要

Resource closures, arrival delays, and processing-time extensions jointly alter station operations during disruptions. Uncertain reopening times further complicate decisions about holding trains and selecting alternative routes. We use observations of disruption-resolution stages to update conditional predictions of the remaining closure duration. A distributionally robust model accounts for errors in scenario probability estimates and jointly optimizes train routes, sequences, and timings. Logic-based Benders decomposition coordinates routing and resource scheduling. Conditional cuts exploit resource-use patterns, and a graph neural network guides the solution order of resource relaxations. Together, these components form the GNN-LBBD algorithm. Across 28 test instances for a five-track station, GNN-LBBD achieved a mean total completion delay of 61.13 train-min. This was 16.63% lower than tabu search and 20.06% lower than variable neighborhood search. Its mean total computation time was close to that of tabu search. On six representative instances, using stage observations reduced mean total completion delay by approximately 21.7%. Continued updating of reopening-time predictions allows route and entry-time adjustments during closures, reducing delays caused by waiting for reopening.

cs.AI

新论文 · v1arXiv:2611.10490

Tactile Sensors Are an Effective Substitute for Wrist Cameras in Contact-Rich or Occluded Manipulation

Aijia Yang, Arsh Tangri, Sergey Zakharov, Krishna Murthy Jatavallabhula

原文摘要

Learned manipulation policies often rely on wrist-mounted cameras for close-range visual feedback that complements external views of the workspace. However, contact-rich manipulation requires establishing and maintaining contacts that can remain hidden even from wrist cameras. Tactile sensors mounted on fingers can reliably and directly measure contact, without needing visual access. In this work, we show that this feedback enables policies to recover the performance advantage of wrist cameras using scene and egocentric vision together with touch. Our approach incorporates two low-cost tactile arrays into a visuotactile ACT policy, without wrist images during training or deployment. We evaluate six real-world task families spanning placement, insertion, occluded retrieval, and sustained-contact tool use. On three contact-rich tasks, we independently vary wrist-camera and tactile availability while keeping the demonstrations, policy architecture, and optimization settings fixed. Adding wrist cameras improves success by 18.3-28.3 percentage points over vision without wrist images. Adding tactile sensing instead improves success by 31.7-60.0 percentage points and yields higher observed success than vision-only policies with wrist cameras on all three tasks, with significant advantages on two under Holm-corrected tests. These results support finger-mounted tactile sensing as an effective alternative to wrist cameras in the evaluated contact-rich settings.

cs.RO

版本更新 · v2arXiv:2605.07125

An Embarrassingly Simple Graph Heuristic Reveals Shortcut-Solvable Benchmarks for Sequential Recommendation

Haoyu Han, Li Ma, Hanbing Wang, Bingheng Li, Daochen Zha, Chun How Tan, Huiji Gao, Xin Liu 等 12 位作者

原文摘要

Sequential recommendation has increasingly shifted toward generative recommenders that combine sequential patterns with semantic item information. Yet these methods are often evaluated on a small set of widely used benchmarks, raising a key question: do these benchmarks actually require the advanced modeling capabilities that modern generative recommenders claim to provide? We conduct a benchmark audit with an intentionally simple graph heuristic. Starting from only the last one or two interacted items, it retrieves candidates from a few-hop item-transition graph and ranks them by item-feature similarity. Despite using no sequence encoder, generative objective, or training, this heuristic matches or outperforms many modern baselines, with relative NDCG@10 improvements of 38.10% and 44.18% over the best competing baseline on Amazon Review Sports and CDs. We show that this behavior reflects shortcut solvability rather than an artifact of one heuristic. We identify three shortcut structures that can make next-item prediction easier than expected: low-branching local transitions, feature-smooth transitions, and limited dependence on long user histories. These shortcuts need not appear together; even one or two strong signals can make simple local retrieval highly competitive, while weakening them makes the benefits of more sophisticated models clearer. Across 14 datasets, model rankings vary substantially with dataset properties, yet the heuristic remains competitive on 10 of them. Our findings suggest that strong performance on standard benchmarks does not always demonstrate advanced sequential, semantic, or generative modeling ability. We call for more careful dataset selection and dataset-level diagnostic analysis when using benchmarks to support claims about new recommendation models.

cs.IR · cs.AI

新论文 · v1arXiv:2611.10488

Safer for Whom? A Systematic Exploration of AI Safety in Eating Disorder Interactions

Jinyuan Wang, Minh Duc Chu, Frances Ross, Ellen Fitzsimmons-Craft, Kristina Lerman

原文摘要

AI assistants advise users about diet, exercise, and weight loss---areas in which unsafe guidance can worsen body dissatisfaction, dietary restriction, compulsive exercise, and other behaviors associated with eating disorders (EDs). Existing evaluations emphasize diagnostic accuracy, overtly harmful prompts, or synthetic scenarios, providing limited insight into whether models recognize implicit vulnerability in natural interactions or protect users consistently. We evaluate responses from three open-weight large language models to 100 questions sourced from an online pro-anorexia community. We grade responses using clinically informed criteria that extend a seven-dimensional taxonomy of ED-related harms. We also construct matched conditions in which questions are accompanied by information about the user varying race, gender, body mass index (BMI), and explicit ED vulnerability. Safety varies substantially across models and harm dimensions. High-risk responses are approximately twice as frequent without bios, suggesting that models do not consistently recognize vulnerability from language alone. Explicit signals of ED vulnerability improve safety by 10--14 points across models. Responses also differ significantly by race, gender, and BMI, with body size producing the largest disparities. These patterns are consistent with reliance on the stereotype that EDs primarily affect thin, White women, potentially leaving users outside this profile with less protective responses.

cs.HC · cs.AI

新论文 · v1arXiv:2611.10487

GEAR: A GPU-Accelerated Global Solver for Bounded Continuous Nonlinear Programs via Linear Bound Propagation

Duo Zhou, Hesun Chen, Xiangru Zhong, Grani A. Hanasusanto, Huan Zhang

原文摘要

We present a novel GPU-accelerated global solver for constrained nonlinear programming (NLP) inspired by the efficient linear bound propagation framework in neural network (NN) verification. Existing work on NN verification, a problem that can be cast as feasibility detection for a nonlinear objective, demonstrated orders-of-magnitude speedups with the linear bound propagation-based framework when compared to traditional solvers. Extending this paradigm to global nonlinear programming requires addressing many challenges: a global constrained NLP solver must effectively exploit all, possibly unstructured, nonlinear constraints to improve the objective, whereas NN verification typically handles simple input box constraints over a well-structured NN and does not directly optimize an objective function. We address these challenges by fully exploiting the linear bounds produced by bound propagation: these linear bounds can be viewed as relaxations of nonlinear constraints and used to provide lower bounds, check infeasibility, and guide the search for a primal solution. In addition, we propose tighter linear relaxations for bilinear, quotient, and other commonly used nonlinear functions in NLP problems, and a projected augmented Lagrangian method tightly coupled to our dual solving procedure to produce primal solutions. Our method solves 421 of 505 bounded and continuous NLP problems in the GAMS and MINLPLib benchmarks under a 180-second limit, more than each of the six global solvers we compare with, including Gurobi and BARON (commercial), and SCIP (open-source). Sixteen of these problems are certified by none of the six global solvers; all of them contain exponential, logarithmic, or trigonometric terms, which are often challenging for traditional approaches. Project webpage is at https://verified-intelligence.github.io/GEAR.

math.OC · cs.LG · cs.MS · math.NA

版本更新 · v2arXiv:2605.09727

One for All: A Non-Linear Transformer can Enable Cross-Domain Generalization for In-Context Reinforcement Learning

Bowen He, Juncheng Dong, Lin Lin, Xiang Cheng

原文摘要

A central challenge in reinforcement learning (RL) is to learn models that generalize beyond the tasks on which they are trained, a goal traditionally pursued through multi-task and meta RL. Recently, transformer architectures have emerged as a promising approach, enabling adaptation to new tasks via in-context learning without explicit parameter updates. From a functional perspective, a transformer can be viewed as a functional operator that maps a context to a task-specific function. It is thus fundamental to understand and design this operator to support stronger generalization in RL. In this work, we address this resulting question of generalization from a kernel-based perspective by establishing a connection between non-linear transformers and kernel-based temporal difference learning. By interpreting the transformer as performing regression in a Reproducing Kernel Hilbert Space (RKHS), we show that value functions from different domains can be represented using a shared set of weights, provided they lie within the same RKHS. Experiments on multiple MetaWorld domains support this interpretation, demonstrating convergence of the temporal-difference objective.

cs.LG · cs.AI

新论文 · v1arXiv:2611.10486

Test-Time Scaling for Diffusion Language Models with Frequency-Aware Remasking

Bowen Zuo, Yue Yu, Dongruo Zhou, Yinglun Zhu

原文摘要

Diffusion language models (DLMs) generate text through iterative denoising, where decisions about which tokens to commit or remask strongly affect output quality. Existing certainty-aware strategies rely on instantaneous confidence or entropy, but these local signals can be misleading on difficult reasoning problems. We formulate DLM test-time scaling as adaptive allocation of both inference compute and decoding strategies. Our two-stage approach uses a fast base sampler to solve easy questions, then applies frequency-aware remasking to the remaining hard questions. This strategy tracks how often each token's prediction changes across denoising steps and preferentially commits frequently changing tokens, encouraging samples to explore alternative reasoning paths rather than collapse onto the model's preferred trajectory. Experiments with LLaDA-8B-Instruct and Dream-7B-Instruct show consistent improvements over a strong adaptive allocation baseline, with absolute accuracy gains of up to 3.73 percentage points (LLaDA) and 7.33 percentage points (Dream) on MATH-500, and absolute coverage gains of up to 11.39 percentage points (LLaDA) and 10.98 percentage points (Dream) on HumanEval. On questions unsolved by the base certainty-aware strategy after 100 samples, frequency-aware remasking recovers solutions for 3.75% to 18.75% of examples.

cs.CL · cs.AI

版本更新 · v2arXiv:2609.31470

Beyond Empirical Support: Structured Outlier Generation via Sinkhorn Optimal Transport

Haixiang Sun, Andrew L. Liu

原文摘要

Outliers are essential for evaluating and improving the robustness of machine learning systems, especially when future distributions may differ significantly from historical training data. In high-stakes applications, robustness often depends on rare cases that finite datasets fail to capture, making simple resampling or perturbation insufficient for stress scenario generation. Existing outlier synthesis methods typically rely on sparse neighborhoods, low support latent regions, or classifier boundary crossings, which can be heuristic, unstable, and tied to specific modalities or architectures. We therefore propose Sinkhorn Boundary Outlier Generation (SBOG), a structured framework for latent-space outlier generation that couples Sinkhorn optimal transport geometry with distributionally robust boundary modeling. The resulting Sinkhorn-induced support cost guides the sampler toward weakly supported boundary regions, while semantic constraints prevent uncontrolled drift from the intended context, yielding controlled deviations from the in-distribution reference measure rather than arbitrary sparse-region samples. Experiments on time series anomaly generation and image outlier synthesis show that our framework produces informative, semantically controlled outliers and improves downstream robustness evaluation across modalities, providing a foundation for stress scenario generation beyond empirical support.

stat.ML · cs.LG · math.OC

版本更新 · v4arXiv:2608.10317

From Detection to Understanding: TAR and TAR-Bench for Multi-Task Traffic Anomaly Reasoning

Han Zhang, Yilin Zhao, Zaid Pervaiz Bhat, Zheng Tang, Varun Praveen, Vidya N. Murali, David C. Anastasiu, Tomasz Kornuta

原文摘要

Detecting a traffic anomaly does not establish whether a video-language model can explain what happened, localize it in time, or identify its causes. We introduce TAR (Traffic Anomaly Reasoning) and TAR-Bench, paired resources for training and evaluating these complementary capabilities across 10 tasks spanning question answering, temporal reasoning, and scene understanding. TAR contains 44,040 automatically generated annotations with chain-of-thought traces for 3,670 CCTV videos from eight public datasets. TAR-Bench provides 960 human-curated annotations for 80 held-out clips from 17 public YouTube videos. Evaluation of eleven vision-language models and analysis of fine-tuned outputs reveal temporal, causal, and descriptive errors not captured by closed-form question-answering accuracy. Training on all 10 tasks raises mean benchmark scores from 34.3 to 55.7 for Cosmos-Reason2-8B and from 30.9 to 53.9 for Qwen3-VL-8B-Instruct. These results support joint training across complementary tasks, while highlighting persistent limitations in temporal precision and causal attribution. TAR and TAR-Bench serve as the official training and in-domain evaluation resources for AI City Challenge 2026 Track~3. The dataset is available under CC-BY-4.0 at https://huggingface.co/datasets/nvidia/PhysicalAI-Traffic-Anomaly-Reasoning.

cs.CV

新论文 · v1arXiv:2611.10484

Can General-Purpose Generative AI Supply Training Data for Medical Vision Models? A Multimodal Retrospective Study

Jiachuan Peng, Kyle Lam, Jierui Qu, Lin Li, Haining Zheng, Bang Zheng, Aniruddha Agarwal, Zaki Almallah 等 13 位作者

原文摘要

Background: Well-annotated, diverse medical imaging datasets remain a major constraint for vision-based medical AI. We evaluated Nano Banana, a general-purpose image generation and editing model, as a source of synthetic medical images for model training, subgroup augmentation, and model auditing. Methods: In this retrospective study across four imaging modalities, chest radiography, fundus photography, and CT models were trained only on synthetic images and externally validated on real-world data against real-data-trained models. Carotid ultrasound models were trained on a private Human Phenotype Project cohort (aged 40-75 years) with and without synthetic augmentation. Clinicians rated image plausibility. Evaluation used AUROC, AUPR, calibration, and decision curve analysis. Findings: Synthetic-only models achieved AUROCs of 0.877 (95% CI 0.843-0.910) for pneumonia (n=14863), 0.963 (0.948-0.975) for retinal detachment (n=1149), and 0.711 (0.600-0.823) for renal calculus (n=4122), comparable to real-data-trained models, but their calibration and net benefit were not consistently more favorable. With mixed synthetic-real training, a carotid ultrasound model produced an ordinal vascular age index that stratified 633 external subjects into subgroups differing in BMI and glucose (both p<0.01), linking vascular aging to cardiometabolic health. To stress-test shortcut learning, editing a moderate tessellated pattern into fundus photographs reduced the true-class probability for central serous chorioretinopathy from 0.996 to 0.822. Clinician-guided editing raised low-quality CT plausibility from 3.10/5 to 4.30/5. Interpretation: Synthetic data can complement real data in model training, subgroup representation, and model testing via controlled editing, but its clinical value should be assessed with calibration, decision curve analysis, and clinician review, not discrimination alone.

cs.CV · cs.AI

新论文 · v1arXiv:2611.10483

They Can See It but Not Say It: Iterative Self Knowledge Re-expression in Visual Reasoning Relative Pose Identification

Mengyu Wang, Ken Deng, Yftah Ziser, Shay B. Cohen, Tiejun Ma

原文摘要

The task of visual reasoning relative pose identification (VRRPI) has exposed critical weaknesses in the multi-view 3D spatial reasoning capabilities of modern vision-language models (VLMs). We present a systematic analysis of VLMs' behavior in VRRPI, uncovering stable output biases where VLMs disproportionately predict specific motion patterns regardless of visual evidence. To mitigate these inherent biases, we propose self knowledge re-expression with iterative debiasing (SKR-ID), an annotation-free pipeline that adapts VLMs for VRRPI using only randomly extracted unannotated image pairs. SKR-ID transitions VLMs' output mechanism from generic next-token prediction to task-optimized classification heads. The pipeline proceeds in two stages: (1) a cold-start initialization using debiasing object-centric prompts to anchor motion inference in 2D visual evidence; and (2) a recursive logit-ranking mechanism that enforces statistical parity by using the median logit value as a dynamic threshold. Experimental evaluation across frontier VLMs on standard benchmarks demonstrates that SKR-ID significantly mitigates systematic biases. Our method improves open-source VLMs' Macro-F1 scores from 35.0-45.4% to 53.7-58.0%, without using any annotated data. These results show that SKR-ID can effectively improve 3D reasoning performance from a severely collapsed baseline. The code is publicly available at https://github.com/astudentuser/Codes_of_SKR-ID.

cs.CV

版本更新 · v3arXiv:2605.27996

Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure

Max Lamparth, Daniel Fein, Andreas Haupt, Marcel Hussing, Mykel J. Kochenderfer

原文摘要

Single-axis mitigations of reward-model biases, such as reducing proxy reliance on length, sycophancy, or style, can redirect optimization pressure onto correlated proxies rather than eliminate it, a new type of reward hacking we call reward bias substitution. The failure is enabled by a measurement-versus-optimization gap between mitigation evaluation on the audit distribution and policy training on the policy-induced distribution. Prior reward-hacking analyses assume a fixed reward proxy, whereas we study the mitigation operator that removes one bias from it, the basic step of iterative reward refinement. We formalize mitigation outcomes into a regime taxonomy and prove that successful mitigation, bias substitution, and overcorrection can produce identical observables under every audit-distribution score, including ranking accuracy and win-rate, even when granted oracle access to the true reward. Across 30 published preference-learning mitigation works, no method we survey reports the evidence needed to certify successful mitigation. Augmenting evaluation with policy-induced distributions while tracking multiple biases provably resolves this distributional gap, and we translate this into actionable prescriptions for mitigation methods and benchmarks. We demonstrate bias substitution in RLHF, where a length penalty in GRPO shortens responses as intended yet makes the policy overconfident and less accurate on open-ended questions, show that the same distributional gap hides failures of published mitigations, and find the necessary preconditions for reward bias substitution documented in published results and in our study of length-sycophancy coupling in LLM responses.

cs.AI

新论文 · v1arXiv:2611.10477

PatternBloom: Empowering Agentic RAG with Externalized RL-Distilled Graph Patterns

Yi Ge, Heng Huang

原文摘要

Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms $27$ baselines by $+15.0$ Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.

cs.AI

版本更新 · v2arXiv:2611.07750

Foresee to Trade: Forecast-Augmented Policy Optimization for Portfolio Management

Zeyu Liu, Xuanzhi Feng, Haoran Xu, Lai Sing Kwong, Yang Li, Hualei Zhang, Hao Chen, Jie Zhang 等 10 位作者

原文摘要

Online investment services must translate asset opportunities into investment outcomes for users' accounts, yet the same market opportunity can warrant different account actions. We study how supervision from future asset opportunities can strengthen account-level trading decisions and propose Forecast-Augmented Relative Policy Optimization (FRPO). FRPO separately supervises asset opportunities and account decisions through a jointly learned representation: a Financial State Head (FSH) learns future relative-return targets to provide asset-level opportunity supervision, while Same-Path Group-Relative Optimization compares trading trajectories under the same market path and initial account to provide account-level relative-outcome supervision. An Adaptive Gate coordinates auxiliary supervision using predictive variance and gradient alignment. FSH is used only during training; inference uses the encoder and policy for trading. Across DJIA30, TOPIX Core30, and three investment-style portfolios, ten-seed evaluations show that FRPO achieves the highest mean cumulative and annualized returns among the compared methods in all five settings. These results support strengthening investment agents' trading decisions through joint supervision from asset opportunities and trading outcomes.

cs.LG

版本更新 · v2arXiv:2606.17584

Root-Selecting Fixed-Point Inversion for Rectified Flows via Trajectory Straightness

Semin Kim, Jihwan Yoon, Seunghoon Hong

原文摘要

Finding the initial noise that generates a given data sample, known as inversion, is a key component for downstream applications such as training-free image editing. Existing fixed-point inversion methods improve inversion accuracy by formulating each inversion step as a fixed-point problem, but they lack a principled mechanism for selecting among multiple fixed-point solutions that can arise in practice. We observe that different selections induce different inversion trajectories, leading to substantial variation in reconstruction and editing quality. For rectified flows, we further find that this variation is closely associated with trajectory straightness, motivating straightness as a principled selection criterion. We propose SelFix, a fixed-point inversion method that selects fixed-point solutions inducing straighter inverse trajectories while retaining convergence to an exact inverse root under standard local assumptions. Experiments on FLUX.1-dev and PIE-Bench show that SelFix improves fixed-point inversion, achieving stronger real-image reconstruction and better source-preserving prompt-based editing than prior inversion baselines. The code is available at https://github.com/seminkim/selfix .

cs.CV · cs.LG

版本更新 · v2arXiv:2608.29996

Partition-Aware Unlearning for Removing Spurious Correlations in Large Vision-Language Models

Aditi Sarker, Nazreen Shah, Rafi Ibn Sultan, Rhongho Jang, Dongxiao Zhu, Prashant Khanduri

原文摘要

Large Vision-Language Models (LVLMs) achieve strong performance across many multimodal tasks; however, they often exploit spurious object-background correlations, resulting in predictions driven by contextual shortcuts rather than object-relevant visual evidence. Despite growing interest in hallucination and robustness evaluation, existing benchmarks provide limited control over whether model predictions are grounded in the target object or induced by correlated background cues. In this work, we introduce PURGE (\underline{P}artition-aware \underline{U}nlearning for \underline{R}emoving spurious-correlation \underline{G}enerated \underline{E}rrors), a framework for constructing, benchmarking, and mitigating spurious-correlation-induced failures in LVLMs. The framework consists of: -- (1) Structured dataset construction wherein we develop three complementary structured data construction strategies that partition examples by object-relevant evidence and spurious background cues, enabling controlled diagnosis of shortcut reliance; and -- (2) Partition-aware unlearning, which uses these partitions to selectively remove spurious object-background associations while preserving object-based reasoning. We evaluate the \algo~framework across multiple LVLMs, including LLaVA-1.6-7B, Qwen3-VL-8B-Instruct, and Qwen3.5-9B, together with CLIP as a vision-language encoder, on a diverse suite of benchmarks, including CHAIR, POPE, Causal-HalBench, MM-SpuBench, AMBER, MMHal, and Waterbirds. Our results show that PURGE consistently reduces hallucinations and spurious-correlation-driven errors while maintaining or improving overall performance in most evaluated settings, providing both a reusable evaluation protocol and an effective mitigation framework for more reliable LVLMs.

cs.CV · cs.AI · cs.LG

新论文 · v1arXiv:2611.10475

WordEval: Evaluating Word-Native Operation Fidelity in Document Editing

Jinyoung Park, Yupan Huang, Shimin Zhang, Jiayu Ding, Yilin Jia, Yuzhong Zhao, Tengchao Lv, Wenshan Wu 等 21 位作者

原文摘要

Microsoft Word is a central environment for office automation, and the faithful execution of its editing operations is an important capability for document agents. Unlike plain-text editing, editing in Word modifies a structured object model, where the requested effect is realized in object properties. Existing benchmarks, however, evaluate office workflows or desktop automation, treating word-processor documents as one task domain among many and verifying outcomes with keyword matching, per-task property checks, or screenshot comparison. We introduce WordEval, an operation-fidelity benchmark that instead centers evaluation on Word editing operations, testing whether each requested edit is realized as intended changes to Word objects and properties. It contains 642 cases across 15 Word feature areas, ranging from text and lists to charts, shape geometry, and SmartArt. 546 atomic cases cover 231 commands, 96 compositional cases combine multiple edits in a single document, and a Word-native oracle checks task-specific object properties and preservation conditions. Cases are constructed from editing-unit specifications, yielding controlled input documents, reference documents, and aligned oracle checks. Each compositional case is linked to corresponding atomic cases, supporting fine-grained diagnosis of editing failures. We evaluate eight agent systems and find that many struggle on the Atomic track, with performance varying across feature areas. Moreover, every system scores lower on Compositional than on Atomic, with gaps that vary substantially by system. Exact atomic-intent links and recorded failure diagnoses provide complementary views of this variation.

cs.CV

版本更新 · v2arXiv:2605.27769

Smoothed Score Queries and the Complexity of Sampling

Jingbo Liu

原文摘要

We study the query complexity of sampling from high-dimensional Gaussian distributions using gradient information. In the standard oracle model, exact gradients expose only matrix-vector products with the precision matrix, leading to polynomial approximation barriers and a characteristic $\sqrtκ$ dependence on the known spectral range. We show that this barrier disappears when the sampler is allowed to query smoothed scores, namely gradients of the logarithms of the Gaussian-convolved densities. For a centered Gaussian target with precision matrix $Λ$, a smoothed-score query at noise level $τ$ gives access to the resolvent $(Λ+τ^{-1}I)^{-1}$. Combining geometrically spaced noise levels with sinc-quadrature rational approximation, we obtain a sampler with $q=O((\logκ+\log(e\sqrt d/δ_{\rm TV}))\log(e\sqrt d/δ_{\rm TV}))$ smoothed-score queries for total variation error $δ_{\rm TV}$, improving the condition-number dependence from $\sqrtκ$ to logarithmic. We also study finite-bit gradient oracles. Using coordinatewise quantization of the transformed smoothed-score answers and a final dithering step, we obtain a sampling scheme whose total communicated gradient information is polylogarithmic in $κ$; in particular, for fixed dimension and accuracy, the bit complexity is $O(\log^2κ)$. To complement these upper bounds, we introduce a channel-synthesis, or reverse-Shannon, converse technique for sampling lower bounds. This converts total-variation simulation guarantees into communication requirements and yields an $Ω(d\logκ)$ bit lower bound for $d\ge2$ and fixed total variation error below one, as $κ\to\infty$. Together, these results identify smoothed scores as a provably more informative oracle for sampling and give nearly matching upper and lower bounds for its finite-bit complexity.

cs.DS · cs.IT · cs.LG · stat.ML