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

+549新增论文
+443其中 AI 相关
+62已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-21 12:19 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 17:33 — 2026-11-21 17:33(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:2026-11-21 12:19最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 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。

29,158 已收录论文24,181 AI 相关8,992 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 6,976 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-21 12:17(北京时间)
新论文 · v1arXiv:2611.21020

SWE-OPD: On-Policy Distillation for Agentic Coding

Yuhang He, Yanyu Chen, Chuanzhe Guo, Jie Zhu, Zichen Tian, Hongyu Ge, Zixuan Ren, Huang He 等 14 位作者

原文摘要

In this work, we study on-policy distillation (OPD) for transferring agentic coding capabilities from a large teacher model to a smaller student model. Prior work has focused primarily on reasoning and short-horizon agentic tasks. We present SWE-OPD, a minimal OPD study for software engineering, where the student interacts with executable repositories and the teacher supervises the prefixes visited by the student. Across three teacher--student pairs, two training sources, and three SWE benchmarks, we find that on-policy supervision transfers coding capability: supervision from Qwen3.6-27B increases the SWE-bench Pro resolve rate of Qwen3.5-9B from 35.9% to 39.4%. Refreshing trajectories is not strictly required either, as an offline variant with cached teacher scores still improves over the student baselines while cutting end-to-end GPU-hours by 59.4%--72.6% and halving peak GPU allocation, at benchmark-dependent accuracy. Furthermore, we extend the offline variant to corss-tokenizer scenarios: distilling GLM-4.5-Air from DeepSeek-V4-Flash-0731 raises its SWE-bench Pro and Multilingual resolve rate from 35.8% to 39.7% and from 41.8% to 46.2%, around 4% and 4.6% improvement, respectively. We also find that a decreasing loss, however, does not necessarily indicate improved agent performance: under top-$K$ reverse-KL loss, one student exhibits severe repetition despite a decreasing loss. We identify a support-mass gradient term as a potential mechanism underlying this failure.

cs.AI · cs.LG

版本更新 · v4arXiv:2602.05302

PieArena: Ranking and Profiling Language Agents in Realistic Negotiation Scenarios

Chris Zhu, Sasha Cui, Will Sanok Dufallo, Runzhi Jin, Zhen Xu, Linjun Zhang, Daylian M. Cain

原文摘要

We present an in-depth evaluation of LLMs' ability to negotiate, a central business task that requires strategic reasoning and theory of mind to create and claim economic value. To do so, we introduce PieArena, a large-scale negotiation benchmark grounded in multi-agent interactions over realistic scenarios adapted from MBA negotiation courses at an elite business school. We evaluate language agents across three pairing regimes: mirror-play, cross-play, and human-LM play. We develop a ranking model for continuous negotiation payoffs that yields order-invariant, uncertainty-quantified leaderboards while correcting for systematic experimental asymmetries. We further study the effects of shared-intentionality agentic scaffolding and find asymmetric gains, with large improvements for mid- and lower-tier LMs and diminishing returns for frontier LMs. As calibration anchors, we collect human-human and human-LM negotiation data from trained business school students, finding that a representative frontier language agent matches or exceeds this human baseline in our evaluation settings. Beyond deal outcomes, PieArena provides a multi-dimensional behavioral profile that reveals cross-model heterogeneity in instruction compliance, computation accuracy, as well as judge-assessed deception and reputation, illustrating the value of evaluation beyond outcome-only leaderboards.

cs.AI

版本更新 · v2arXiv:2603.27460

Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development

Zhongying Deng, Cheng Tang, Ziyan Huang, Jiashi Lin, Ying Chen, Junzhi Ning, Chenglong Ma, Jiyao Liu 等 126 位作者

原文摘要

Foundation models have demonstrated remarkable success across diverse domains and tasks, primarily due to the thrive of large-scale, diverse, and high-quality datasets. However, in the field of medical imaging, the curation and assembling of such medical datasets are highly challenging due to the reliance on clinical expertise and strict ethical and privacy constraints, resulting in a scarcity of large-scale unified medical datasets and hindering the development of powerful medical foundation models. In this work, we present the largest survey to date of medical image datasets, covering over 1,000 open-access datasets with a systematic catalog of their modalities, tasks, anatomies, annotations, limitations, and potential for integration. Our analysis exposes a landscape that is modest in scale, fragmented across narrowly scoped tasks, and unevenly distributed across organs and modalities, which in turn limits the utility of existing medical image datasets for developing versatile and robust medical foundation models. To turn fragmentation into scale, we propose a metadata-driven fusion paradigm (MDFP) that integrates public datasets with shared modalities or tasks, thereby transforming multiple small data silos into larger, more coherent resources. Building on MDFP, we release an interactive discovery portal that enables end-to-end, automated medical image dataset integration, and compile all surveyed datasets into a unified, structured table that clearly summarizes their key characteristics and provides reference links, offering the community an accessible and comprehensive repository. By charting the current terrain and offering a principled path to dataset consolidation, our survey provides a practical roadmap for scaling medical imaging corpora, supporting faster data discovery, more principled dataset creation, and more capable medical foundation models.

cs.CV · cs.AI

新论文 · v1arXiv:2611.21003

K2-Horizon: A Frontier Fleet of Open Source Language Models

K2 Team, Zhengzhong Liu, Haonan Li, Xuezhe Ma, Alexander Moreno, Subham Sekhar Sahoo, Rupesh Kumar Srivastava, Mikhail Yurochkin 等 95 位作者

原文摘要

We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from every training stage, pre-training data, training logs, and all training, evaluation, and infrastructure code, notably our production-scale xLLM framework for dense and MoE training and the RL360 reinforcement-learning stack. Beyond the data itself, we release and document in detail every pipeline that built it, the part of model development that open releases most rarely disclose. These pipelines produced roughly 10T synthetic pre-training tokens, showing that synthetic data can be a major ingredient of pre-training at scale. Four of the models share one 22T-token pre-training corpus and recipe, revealing consistent scaling behavior and enabling controlled studies of scale and sparsity. All artifacts are available at https://ifm.ai/k2/

cs.AI

新论文 · v1arXiv:2611.20996

Explaining Cross-Label Decision Structure in Multi-Output Tree Ensembles

Sylvio Barbon Junior, Hiago F. F. Oliveira, Rafael Figueira Gonçalves, Jessica Fernandes Lopes, Ricardo Cerri

原文摘要

A multi-output tree ensemble predicts several binary labels through decision paths that it largely shares across those labels, yet the attribution methods commonly used to explain such models treat each output in isolation. We propose multi-sink Decision Predicate Graphs (ms-DPGs), which extend decision predicate graphs with one sink per label: each leaf contributes to the sink of every label its in-bag positive count for that label. On this representation we define Shared Predicate Flow (SPF), the share of a label's decision mass that passes through conditions active for both labels of a pair, and a contracted view of the graph in which label-specific lifts indicate whether a shared condition raises the probability of both labels or of one at the expense of the other. Controlled experiments show that SPF captures shared decision conditions rather than mere feature overlap. On nine multi-label benchmarks, native multi-output forests share substantially more decision structure than Binary Relevance on eight, including five of the six on which the two are at predictive parity, even at equal hyperparameters, whereas per-label TreeSHAP rankings register the change of architecture without showing whether more is shared. Because a native forest routes every label through the same splits, SPF describes the fitted model rather than the data, and on eight of the nine benchmarks it rises when the feature-label signal is removed; a high SPF is therefore not evidence of shared predictive information. ms-DPG offers a global, predicate-level account of how a multi-output ensemble reuses decision conditions across its labels.

cs.LG · cs.AI

新论文 · v1arXiv:2611.20988

Hivemind: A Federated Protocol for AI-Mediated Tacit Knowledge Sharing in Organisations

Srinivasan Sundaresan

原文摘要

This paper presents Hivemind, a federated protocol that enables AI agents to exchange tacit knowledge on behalf of their users, asynchronously and without requiring any direct human interaction. Individual AI memory systems capture the decisions, reasoning, and contextual knowledge that surface during AI-assisted work. Hivemind creates the shared space in which this captured knowledge can flow between individuals across team and departmental boundaries, addressing a structural gap that neither centralised knowledge repositories nor existing AI tooling closes, even where partial solutions such as expert-finding systems exist. The paper evaluates centralised versus decentralised design approaches, proposes a federated architecture grounded in Nonaka and Takeuchi's (1995) Socialisation, Externalisation, Combination and Internalisation (SECI) model, and demonstrates a proof-of-concept implementation using Design Science Research (DSR) methodology. Measuring the enterprise benefit of federated tacit expertise sharing remains an open task for future work.

cs.AI · cs.MA

新论文 · v1arXiv:2611.20986

Calpha: Toward Native Biomolecular Intelligence in a General-Purpose Language Model

Yong Liu, Zhanpeng Shi, Peicong Lin, Zhenyi Zhong, Yizhou Dang, Wendi Wu, Haoling Li, Shuangjia Zheng

原文摘要

Biomolecular foundation models predict the structures of proteins and their complexes accurately, and large language models (LLMs) reason in natural language, but the two remain separate: the abilities of the former are not part of a general-purpose LLM, and the reasoning of the latter does not enter a structure prediction. Calpha aims at one general model that folds proteins and complexes, understands protein sequence and structure, and lets its language reasoning influence the structures it predicts. This report covers the first step, in which an LLM outputs single-chain protein structures. Qwen3.5-9B reads a sequence enriched with ESM2 residue features. It can respond in text or emit a token that invokes a pretrained SimpleFold structure head, which receives information about the protein only through Qwen's residue states. Under a fixed folding request, the head achieves mean TM-scores of 0.808 on the FoldBench-334 development set, 0.635 on CASP16 domains and 0.900 on a CAMEO-style PDB set. We compare these predictions with a baseline in which ESM2-3B encodes the sequence and a fine-tuned SimpleFold-3B head predicts the coordinates. The two systems also differ in trainable weights, training duration and crop length. For chains of 100--299 residues, all three paired confidence intervals include zero. At 300--600 residues, the mean differences are -0.044, -0.160 and -0.025, respectively. In a separate test, four English request types triggered folding on every evaluated example. Protein training reduced accuracy on a 350-item MMLU-Pro subset from 77.4% to 70.3%. An LLM can therefore output single-chain structures from its own states, with lower long-chain scores in the current comparisons and reduced accuracy on the evaluated MMLU-Pro subset. Complex prediction, structure understanding and the effect of language reasoning on the predicted coordinates remain future work.

q-bio.QM · cs.AI

新论文 · v1arXiv:2611.20983

GRACE: Parallel Best-of-N Multi-Agent Analysis of Closed-Form Questions over Dataframes with Guardrail Selection and Adaptive Retry

Fabio Picariello, Salvatore Cipolla, Susanna Bonura, Vito Morreale

原文摘要

LLM agents answer closed-form questions over a tabular dataframe by writing and executing Python, almost always as a single rollout: one interpretation, one script, one result, accepted without comparison. The few pipelines that produce several candidates reconcile them by answer-agreement voting, which needs a single comparable scalar and always returns one, even when every candidate is wrong. We present GRACE (Generative, Ranked, Adaptive Code Ensemble), a production-oriented pipeline that instead runs parallel best-of-N selection specialised to dataframe code. N agents run concurrently under a strict, machine-checkable output contract, diversified by persona hints, per-retry rotation, and a two-regime temperature schedule. Rather than vote, GRACE judges the executed artefacts: a pairwise tournament pre-filters the pool and a domain-specialised guardrail either selects the best candidate or, when none is defensible, rejects the whole pool and abstains with a diagnosis (an option voting structurally lacks), which in turn drives an adaptive retry with per-agent feedback. On InfiAgent-DABench (DAEval, M = 246), using Qwen3.6-35B-A3B as the backbone, GRACE raises answer match from 39% for a single-rollout baseline to 71% (+32 pp, 1.8x), recovering 53% of previously wrong answers. On the same DAEval task, the smaller dense Qwen3.5-9B lifts answer match from 38% to 63% (+25 pp, 1.7x). A controlled 3 x 3 study that sweeps the number of parallel agents against the retry budget finds, on the 35B backbone, the two levers largely substitutable: once either one is generous, the other adds little, identifying a cheap operating point for deployment. The same directional pattern replicates on the independent DataBench benchmark, again on the Qwen3.6-35B-A3B backbone, where answer match rises by 15 percentage points (from 70% to 85%) across 400 sampled questions.

cs.AI · cs.LG

新论文 · v1arXiv:2611.20981

Breaking feature collapse in self-supervised time series encoders

Özgün Turgut, Daniel Rueckert

原文摘要

General-purpose time series encoders promise a single backbone that generalises across tasks and domains; however, they currently do not scale. We trace this down to a critical point of failure: under the predominant masked data modelling supervision, the encoder's output tokens partially collapse. This is the natural consequence of supervising encoders only in the input space, which leaves no constraint against redundant token dimensions in the feature space. We introduce OTISv2, which suppresses the collapse by complementing the reconstruction objective in input space with two objectives in the feature space: a self-distillation loss against an exponential-moving-average teacher, and a KoLeo regulariser. Under matched training, the resulting encoder reduces the cosine similarity and lifts the effective rank of its output tokens relative to its predecessor, and it scales monotonically from 7.1 M to 116 M parameters where its predecessor declines. It matches or surpasses existing general-purpose time series encoders across 143 uni- and multi-variate datasets. To foster further research, we release our code and pre-trained weights.

cs.LG · cs.AI

版本更新 · v2arXiv:2610.26430

Physiology-Aware Graph Attention for Early Ventricular Reentry Prediction

Ma{ë}lis Morier, Jairo Rodr{í}guez Padilla, Mihaela Pop, Patrick Gallinari, Maxime Sermesant

原文摘要

Sudden cardiac death mostly originates from post-infarct ventricular tachycardia. This reentry is sustained by propagation around scar tissue, but its risk remains difficult to predict from infarct imaging alone. While in silico induction studies can reveal these mechanisms, exploring many substrates and pacing locations remains computationally expensive. We propose a graph attention classifier for early prediction of sustained ventricular reentry from simulated cardiac electrophysiology. Each simulation is represented as a graph with early nodal voltage samples and edge attributes encoding geometry, stimulation frequency, tissue labels, amplitude, and diastolic interval. These descriptors are computed only from the first two paced activations, before sustained reentry is observed. On a balanced independent test set of 1000 simulations, the classifier achieves 94.6% accuracy, 95.4% sensitivity, 93.8% specificity, 94.6% F1-score, 0.987 ROC-AUC, and 0.985 PR-AUC. Ablation experiments show that diastolic interval is the most critical descriptor. Qualitative attention inspection suggests that different heads focus on border-zone, stimulation location, and early propagation path. These results support graph attention as a fast early-prediction approach for large-scale in silico ventricular reentry studies.

cs.AI

新论文 · v1arXiv:2611.20976

Living Context Graphs: Continual Semantic Learning for Enterprise Data Reasoning

Varun Agrawal, Ayush Srivastava, Eshan Gujar, Kaushal Kumar, Ayush Gupta

原文摘要

Enterprise AI can now perform analytics, reporting, and data-driven workflows, yet the data foundations it relies on remain hard to operationalize: business meaning, relationships, metric definitions, and reporting conventions are absent from database schemas and scattered across undocumented systems and tribal knowledge. We introduce the Living Context Graph (LCG), a continually learned context layer that makes existing enterprise data operational for AI without requiring a copy or transformation of the underlying warehouse. LCG bootstraps base semantics and relationships from the data itself, builds a concept graph that indexes business context as queries are served, writes durable semantic learnings from verified outcomes, and withdraws or re-earns context when later outcomes contradict prior routes. We evaluate LCG through text-to-SQL, a core data-reasoning task, on a transfer-aware split of three BIRD databases comprising 499 questions: 353 form the learning history and 146 are held out and never used to update the context layer. Bootstrapping alone improves overall accuracy by +4.6 percentage points. Learning only from the history raises accuracy on the untouched held-out transfer set by +22.6 points (95% CI [+14.4, +30.8]), exceeding five-shot retrieval from the same history by +12.3 points. Incremental learning reaches the one-shot endpoint within observed evaluation noise on every database. Ablating the graph changes held-out accuracy by at most one question per database but increases per-query cost 2.3-3.3x and median latency 1.8-3.4x: continually learned semantics drive accuracy, while the graph makes that context efficient to serve.

cs.CL · cs.AI · cs.DB · cs.ET · cs.IR

新论文 · v1arXiv:2611.20970

Hilbert States for Decision-Making under Partial Observability

Zibo Meng, Andrew C. M. Austin, Alin Achim

原文摘要

Decision-making under partial observability can be fundamentally limited by information, even with known dynamics: distinct physical states can induce the same observation history while favoring different actions. We study three questions: when is full-state optimality unattainable, what information exactly preserves the best observation-based decision, and how does finite-data learning affect reliability? First, we characterize decision ambiguity through conditional action preferences. We give a zero-regret criterion, a positive lower bound for conflicting uniquely optimal actions on positive-probability regions with uniformly positive cost gaps, and an exact binary-action regret formula. This also shows why physical-state ambiguity alone need not cause decision loss. Second, for two actions and squared terminal-goal costs in the plane, we derive an exact inner-product representation of conditional expected cost differences using three conditional moment contrasts and a goal-dependent feature. This Hilbert state preserves observation-optimal actions without hidden-state reconstruction. Under nondegenerate history and goal variation, its three-dimensional form is minimal among exact bilinear cost-contrast representations. Third, we relate finite-data state errors to excess decision regret via goal-projected errors, action margins, and goal geometry. Physics-based simulations verify information-limited regret, exact action preservation, and the role of second moments. Held-out real-SAR experiments demonstrate feasible goal-conditioned decisions from masked images, while finite-data evaluations reveal goal-dependent reliability. Together, these results separate irreducible information loss, exact representation sufficiency, and learning-induced excess regret.

cs.AI · cs.LG

新论文 · v1arXiv:2611.21695

Stop Early, Stay Simple: Evidence Acquisition for Power System Fault and Attack Diagnosis

S M Zia Ur Rashid, Sazidul Islam Hira, Rupesh Sah, Suman Rath

原文摘要

Diagnostic models for cyber-physical power systems usually inspect every telemetry source for every event, although some events are clear from one source and others need several. We ask when a learned policy for choosing that evidence is worth its added complexity. To answer this, we present ActiGrid, a benchmark framework for active evidence acquisition that pairs a classifier trained to tolerate missing modalities with a Double Deep Q-Network (DQN), which either requests another source or stops and diagnoses but never operates grid actuators. On the public TU Delft digital substation and MSU/ORNL power system attack benchmarks, we compare the DQN with fixed subsets and orders, calibrated entropy stopping, a greedy value of information rule, and tree ensembles trained per subset. The learned policy does adapt to what it observes, yet it is statistically indistinguishable from entropy stopping on TU Delft and Pareto dominated by simple controls on MSU/ORNL, while compact subset classifiers outperform it while reading fewer sources on both benchmarks. Entropy stopping run over those subset classifiers reaches the highest mean Macro-F1 on TU Delft (0.928 while reading 57% of the sources) but brings no saving over a static three source subset on MSU/ORNL, where no method exceeds 0.45 Macro-F1. Selective acquisition can pay off, but on these benchmarks it does not need a learned policy.

cs.CR · cs.AI · cs.LG · eess.SY

新论文 · v1arXiv:2611.20955

Jev-as-a-Reward: Validating Probabilistic Safety Judges for Reinforcement Learning

Eunkyeong Lee, Junhyung Moon, Seunggwan Hong, Dongsuk Oh

原文摘要

Probabilistic safety judges return the probability of a safety label and cheaply score reinforcement learning rollouts. A confidence gate sends only uncertain verdicts to a stronger LLM judge, which leaves confidently passed harm in a blind spot. Policies can exploit this blind spot, and commercial judges release new versions. We propose a protocol that validates such a judge before and during training, separates the share of confidently passed harm from its amount, and re-validates it after each judge change. We apply it to two versions of Jev, a commercial decision-only judge, and to an open guard classifier. Under the probability reward of Jev, passed harm shrinks in all 11 runs under the primary reference, mostly more slowly than caught harm, and concentrates in the blind spot. Binarizing this judge meets the escape criterion on Llama, and the open guard classifier shows an escape signal in one seed under a second reference. Guard classifiers share this blind spot, and even frontier LLM judges catch at most one of 10 human-labeled confident misses. Without a trained reward model, the probability reward with a helpfulness term lowers HarmBench attack success on Llama-3.1-8B from 0.287 to at most 0.033 at the final checkpoint.

cs.AI · cs.CL

新论文 · v1arXiv:2611.20942

vividRGB: Various visual cue detection in RGB photographs with forced focused attention

Masahiro Ryo

原文摘要

Natural photographs contain dense, overlapping structures that can be weakly expressed in feature maps from whole-image vision encoders. We introduce forced focused attention (FFA) and its Python implementation, vividRGB. FFA is an encoder-agnostic inference-time procedure that applies a frozen encoder to enlarged, overlapping local crops, averages inverse-aligned reflections, and reconstructs a feature field in source coordinates. We evaluate 50 high-resolution photographs from grassland vegetation, fungal networks, airborne forests, coral reefs, and phytoplankton microscopy, using DINOv3 ViT-B/16. Across the five domains, FFA at 5% or 10% focus produces higher mean RGB-edge boundary F1 than direct DINOv3 encoding and automatic SAM3 masks. Domain means range from 0.244 to 0.596 for FFA, versus 0.028-0.076 for direct DINOv3 and 0.020-0.090 for SAM3. Qualitative feature maps also show finer appearance structure. These results concern RGB-edge preservation rather than verified species, traits, or biological segmentation. Narrower focus generally improves boundary preservation, while extraction time and GPU energy use increase sharply and gains diminish at finer settings. FFA may be useful for relatively small, high-resolution ecological image collections where dense visual cues matter. The alpha release vividRGB 0.1.0 and versioned research resources support reproducible cue inspection.

cs.CV · cs.AI

版本更新 · v2arXiv:2611.16493

ReA-Bench: Benchmarking Responsible Agent Auditing

Zipeng Ye, Shenyu Dai, Yong Xie, Kun Wang, Jiyang Guan, Zhikai Chen, Jiaqi Weng, Jialing Tao 等 9 位作者

原文摘要

As agents perform tasks autonomously in local and cloud environments, auditing their execution trajectories can help service providers identify risks and improve agent harnesses and foundation models. Our analysis of production and red-team trajectories shows that some risks are repeatedly missed across auditing configurations. Overall recall obscures which risks are missed and whether higher recall comes with more false positives on legitimate behavior. Existing benchmarks offer limited insight into these failures. Generated trajectories can contain inconsistent tool calls and responses, conspicuous risky actions, or repetitive attack patterns. Grouping results by risk type also makes it difficult to tell whether auditors overlooked evidence, accepted an unsupported conclusion, or lacked information needed to make a judgment. We introduce ReA-Bench, a diagnostic benchmark for auditing responsible agent behavior. It organizes auditing challenges into six tiers: internal safety knowledge, contradictions within trajectories, evidence retrieval in long contexts, verification of sources and reasoning, indirect steering, and missing information. It includes challenging benign cases in which actions resemble risky ones but are consistent with the task and its authorization. It also provides paired trajectory versions to test whether auditors detect the same risks more reliably when the trajectory is shortened or missing information is supplied. Most samples are drawn from real production trajectories, with scenarios manually selected and adapted to preserve realistic tasks and risks that could plausibly arise during execution.

cs.AI

新论文 · v1arXiv:2611.20936

ReAccess: Look Once, Stay Focused with Efficient Test Time Memory

Tianhao Chen, Yilong Fan, Yuheng Wu, Xiaobin Hu, Dongman Lee

原文摘要

Despite recent advances, vision-language models (VLMs) still struggle with fine-grained visual understanding, often failing to effectively utilize critical evidence already encoded in their visual representations. Existing approaches address this limitation through repeated visual processing or explicit reasoning, introducing substantial computational overhead. In this paper, we propose ReAccess, a lightweight framework that preserves and reuses question-relevant visual evidence through compact associative memory. Constructed during a single visual prefill, the memory supports progressive evidence retrieval that steers both attention queries and hidden representations, enabling effective visual grounding without repeated image encoding or explicit reasoning traces. Extensive experiments demonstrate consistent improvements in both general visual understanding and fine-grained perception, while achieving substantially faster inference than existing reasoning-intensive approaches, demonstrating a promising path to efficient multimodal reasoning.

cs.CV · cs.AI

新论文 · v1arXiv:2611.20935

Razavi-Bench: Tracking AI Progress in Analog Circuit Reasoning

Zhishuai Zhang, Oscar Toomey, Taekwang Jang

原文摘要

We propose Razavi-Bench, a benchmark built from the 50 questions in Razavi's two Analog Design Experiments With AI articles and graded by GPT-6 Luna against reference answers and rubrics, and use it to track how fast AI models improve at analog circuit analysis. Across 73 multimodal large language models released between May 2024 and November 2026, the top score rose from 28.83% to 98.33% in 30 months, mostly before the reference answers were published. The leading models now answer almost every textbook question correctly; the hardest questions show a circuit that resembles a textbook one but works differently, or ask about behavior the obvious analysis misses. Within two and a half years, a set of questions that exposed the limits of AI models has become one they answer almost perfectly.

cs.AI

新论文 · v1arXiv:2611.20926

Advancing Cognitive Pairwise Comparisons for Human-Centric Artificial Intelligence and Data Science: A Human-in-the-Loop Framework

Kevin Kam Fung Yuen

原文摘要

Human-centric artificial intelligence and data science (HCAD) systems increasingly involve human-in-the-loop (HITL) interventions to overcome the limitations of fully automated machine learning algorithms, such as automation bias and the inability to incorporate tacit domain knowledge, and continuously improve predictive performance. Data transformations, feature selection, feature weighting, and model selection usually require laborious human involvement across iterations of artificial intelligence and data science (AI&DS) pipelines, making development a challenging and time-consuming process. The Cognitive Pairwise Comparison (CPC) framework is proposed as a human-centric methodology to elicit preferences and priorities in a structured format. To establish essential facets of human-centricity, this research advances the CPC methodology so that user preference elicitation can be systematically quantified in these challenging AI&DS tasks. To demonstrate the usability and validity of these facets, this study presents several applications within movie recommender systems using the MovieLens dataset. It exemplifies how the extended CPC framework can facilitate interactions between human judgments and data across supervised learning, dimensionality reduction, and unsupervised learning algorithms.

cs.AI · cs.HC · math.OC

新论文 · v1arXiv:2611.20922

Experience Rot: Experienced Agents Fail Where Novices Succeed

Zhen Xu, Hang Yin, Zitao Zhang, Kevin Zhao, Xuefeng Liu

原文摘要

Agents are increasingly built to learn from experience: they keep conversation histories, memory files, retrieved notes and skills, and reuse what worked before. We show that this experience turns against them when their tools change silently. We test a range of models with tools defined as in the Model Context Protocol (MCP). After a silent change, experienced agents fail on most tasks, while agents new to the task mostly succeed. We call this failure Experience Rot. Experienced agents copy their earlier call instead of checking the current interface, even when the tool definition shows the change. Then they report success. When the documentation is incomplete, experience is what makes agents succeed. After a change, the same experience leaves them worse off than no experience at all. The failure appears in every form of memory we test and grows with the amount of experience. Simple remedies fall short: announcing the update barely helps, and pointing the agent to the documentation fails whenever the documentation does not spell out the answer.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.20915

Multi-LLM Agentic AI at the Edge: Joint Agent Scheduling and Resource Allocation

Yuanyan Song, Kezhi Wang, Xinmian Xu, Yongmin Li

原文摘要

In this paper, an edge-enabled multi-large language model (LLM) agentic AI architecture is proposed, where one supervisor AI agent and multiple worker AI agents cooperatively execute users' requests at the edge. The supervisor AI agent decouples the user request into a series of sub-tasks and dynamically dispatches them to LLM-equipped worker AI agents via two agentic workflows, i.e., parallel and serial workflows, with a two-level retry mechanism to enhance execution reliability. To prolong the service duration of worker AI agents, the energy minimization problems are formulated by jointly considering AI agent selection and resource allocation, subject to task execution latency and quality of service (QoS) constraints. A resource allocation scheme based on alternating optimization (AO) is proposed. Specifically, for the parallel workflow, the scheme decouples the original problem into bandwidth allocation, retrieval selection and AI agent selection sub-problems, which are iteratively solved by the proposed successive convex approximation (SCA), Lagrangian relaxation (LR), and Gale-Shapley (GS) or LLM-based algorithms, respectively. For the serial workflow, the problem is similarly decoupled into service time allocation, retrieval selection and AI agent selection sub-problems; the optimal service time allocation is obtained via Karush-Kuhn-Tucker (KKT) conditions, while the latter two sub-problems are addressed by extending the proposed algorithms to the serial scenario. Simulation results show that the proposed resource allocation scheme can reduce system energy consumption compared with other benchmarks, while experiments on a multi-LLM agentic AI system based on Qwen and Llama demonstrate improved task execution success rates.

cs.AI

新论文 · v1arXiv:2611.21694

TIBECG: Temporal Information Bottleneck for Lead-Specific Demographic Privacy in ECG Waveform Release

Tsai-Min Chen, Jun Seita

原文摘要

Transmitting an electrocardiogram (ECG) waveform can expose demographic information beyond an intended diagnostic task. We evaluate TIBECG, an architectural temporal information bottleneck, on individual leads extracted from hospital twelve-lead recordings. A 2,302,337-parameter processor varies its latent length while retaining the same nine-score consistency objective, without age or sex labels in protector fitting. Across 6,386 recordings, ten folds, 12 leads and 21 lengths, retrospective selections retained at least 95% of each lead's clean macro-F1. Their weaker age/sex interval gain exceeded the best eligible conventional transformation on all leads, by a median 22.08 percentage points. The implemented sampling-rate comparison shows that direct downsampling usually suppresses demographic inference more strongly but at a larger loss of diagnostic utility, whereas TIBECG favors utility preservation. Selection stability also varies by lead. Lower latent Kullback-Leibler (KL) divergence does not reliably rank residual sex leakage at a fixed length. These findings support lead-specific evaluation of waveform processing, not validated consumer deployment or expert-readable reconstruction. When the waveform-input diagnostic classifier is hosted as part of a proprietary cloud service rather than deployed on the device, waveform transmission is required to obtain its scores.

cs.LG · cs.AI · cs.CR · eess.SP

版本更新 · v3arXiv:2607.04412

LLM-as-a-Tutor: Policy-Aware Prompt Adaptation for Non-Verifiable RL

Yujin Kim, Namgyu Ho, Sangmin Hwang, Joonkee Kim, Yongjin Yang, Sangmin Bae, Seungone Kim, Jaehun Jung 等 10 位作者

原文摘要

Reinforcement learning (RL) for non-verifiable instruction following increasingly relies on LLM judges with prompt-specific rubrics as reward signals. While recent methods adapt these rubrics to the evolving policy during training, the training prompts themselves remain static, drawn from fixed corpora. This static approach often results in a critical misalignment between prompt difficulty and policy capability, leaving the judge unable to recover a discriminative reward signal when prompts fail to elicit quality variance among rollouts. To address this misalignment, we introduce LLM-as-a-Tutor, a framework that extends the LLM's role from judge to tutor: a single model serves as an examiner that pairwise-compares policy rollouts to detect non-challenging prompts, and as a generator that appends atomic constraints to them. This append-only design preserves the original prompt while adding constraints to increase difficulty in step with the policy's capability, producing a self-calibrating training signal without external difficulty schedules. On three complex instruction-following benchmarks, our method outperforms both policy-unaware baselines and prior policy-adaptive methods that adapt rubrics or rewrite prompts, on average and on four of five metrics, suggesting prompt adaptation as an underexplored axis of policy-awareness in non-verifiable RL for constraint-heavy instruction following.

cs.AI

版本更新 · v2arXiv:2604.17277

Analog Resonant Recurrent Neural Networks via Metacircuits

Zixin Zhou, Tianxi Jiang, Menglong Yang, Mengli Sui, Yizhuo Zhang, Zhihua Feng, Qingbo He, Shiwu Zhang

原文摘要

Physical neural networks offer a promising route to fast and energy-efficient edge intelligence by embedding computation in the intrinsic dynamics of materials and devices. However, processing temporal signals requires recurrent networks with memory, whose trained dynamics are hard to reproduce faithfully in physical systems. Here we present an analog resonant recurrent neural network (R$^2$NN) implemented as a metacircuit of coupled local resonators. A reformulated mechanical-electrical analogy based on frequency-dependent negative resistances maps a trained network one-to-one onto the metacircuit, with every trainable parameter realized as a resistance. Training tunes local resonances and global resistive couplings so that past responses persist as recurrent memory and interfere with incoming inputs, and the resulting impedance landscape routes different frequencies to distinct outputs. Exploiting this recurrent memory, the metacircuit solves a temporal-parity task whose output depends on the entire input history. Building on the learned frequency selectivity, stand-alone R$^2$NN systems classify raw tactile, acoustic and vibration signals in real time and directly drive downstream hardware. This work establishes engineered locally resonant dynamics as a physical substrate for recurrent neural computation and opens a route to compact, low-latency neural hardware for machine intelligence at the edge.

cs.LG · cs.AI · cs.ET · physics.app-ph

版本更新 · v2arXiv:2610.35031

PatentAgentBench: Benchmarking LLM Agents with Professional Patent Task Suites

Yijie Hu, Wei Li, Junjie Zhou, Zhiqi Jiang, Yi Jin, Licong Xu, Xinkai Sun, Kai Duan 等 9 位作者

原文摘要

Patent practice lies at the intersection of technical, legal, and linguistic work, making it challenging yet crucial to technological innovation. Recently, rapid advances in large language models (LLMs) and agentic systems have demonstrated the potential to complete realistic business workflows and even patent works. This raises a question: can current AI systems perform patent work reliably? Existing benchmarks cover a limited range of patent tasks and simplify evidence acquisition by providing reference materials or candidate documents. In this paper, we introduce PatentAgentBench, a benchmark comprising 510 instances across eight task suites: novelty search, freedom-to-operate (FTO) search, design FTO, invalidity search, claim charting, patent drafting, office-action response, and patent translation. Curated from real patent records and professional materials, the benchmark combines task-specific metrics with expert-reviewed annotations and rubrics. Retrieval tasks such novelty search, FTO, etc., require agents to search and compare patent collections without a supplied candidate pool. The drafting and OAR tasks assess the quality of agents' work given technical disclosure and office action notice. We evaluate seven LLMs across four agent harnesses on PatentAgentBench, finding that current systems still struggle to identify novelty, assess risk, and understand practitioner intent throughout patent workflows.

cs.AI

版本更新 · v3arXiv:2510.23587

A Survey of Data Agents: Emerging Paradigm or Overstated Hype?

Yizhang Zhu, Liangwei Wang, Chenyu Yang, Xiaotian Lin, Boyan Li, Wei Zhou, Xinyu Liu, Zhangyang Peng 等 25 位作者

原文摘要

The rapid advancement of large language models (LLMs) has spurred the emergence of data agents, autonomous systems designed to orchestrate Data + AI ecosystems for tackling complex data-related tasks. However, the term "data agent" currently suffers from terminological ambiguity and inconsistent adoption, conflating simple query responders with sophisticated autonomous architectures. This terminological ambiguity fosters mismatched user expectations, accountability challenges, and barriers to industry growth. Inspired by the SAE J3016 standard for driving automation, this survey introduces the first systematic hierarchical taxonomy for data agents, comprising six levels that delineate and trace progressive shifts in autonomy, from manual operations (L0) to a vision of generative, fully autonomous data agents (L5), thereby clarifying capability boundaries and responsibility allocation. Through this lens, we offer a structured review of existing research arranged by increasing autonomy, encompassing specialized data agents for data management, preparation, and analysis, alongside emerging efforts toward versatile, comprehensive systems with enhanced autonomy. We further analyze critical evolutionary leaps and technical gaps for advancing data agents, especially the ongoing L2-to-L3 transition, where data agents evolve from procedural execution to autonomous orchestration. Finally, we conclude with a forward-looking roadmap, envisioning proactive, generative data agents. A curated companion repository is available at https://github.com/HKUSTDial/awesome-data-agents.

cs.DB · cs.AI

新论文 · v1arXiv:2611.20901

Auditing a Moving Target: Evidence Anchoring and the Expiry of Fairness Audits of Provider-Served Language Models

Mika Facchini

原文摘要

Organizations increasingly use large language models (LLMs) for decisions about people, such as hiring, investment and employee evaluation, and rely on them over long periods of time. Fairness audits of such systems, however, are conducted only once. We examine how the biases of LLMs change over time. In a venture screening setting, models from five providers decided whether and how much to invest in startups whose founders differed only in gender, origin and university. We repeated this audit three times over 13 months (248,400 API calls). Study 1 shows that the same model, called under the same identifier one year later, displayed a significantly different bias pattern. Study 2 shows that the successors of retired models carried the same bias as their predecessors, but in a significantly different size and form. Study 3 shows that the bias does not depend on the industry of the startup. A document analysis further shows that providers change and retire models at their discretion, with notice periods of zero to six months. Drawing on these findings, we develop a theory of evidence anchoring. We theorize that bias resides in the model class and persists, whereas audit evidence is anchored to a single model and expires when the provider decides. We derive policy advice for organizations and regulators, namely audits of model classes instead of single models, a version record for every model response, and re-audits triggered by model changes.

cs.CY · cs.AI

版本更新 · v2arXiv:2608.03512

Analysis of Inverse Relation Directionality in Large Language Models

Sefika Efeoglu, Adrian Paschke

原文摘要

Large language models (LLMs) have achieved strong performance on text-to-knowledge graph generation and related tasks. Nevertheless, it is still unclear whether they accurately model the direction-dependent semantics of inverse relations, in which reversing the order of the arguments alters the meaning of a relation (e.g., \textit{mother} versus \textit{child}). To the best of our knowledge, this work presents the first systematic study of inverse relation directionality in LLMs, using a benchmark consisting of 5,457 instances spanning 27 distinct inverse relation labels. We evaluate five open-source LLMs under a multiple-choice prompting framework and further examine the influence of relation descriptions and entity representations by substituting the original entities with synthetic and masked entities. Our findings reveal systematic asymmetries in inverse relation classification across LLMs, indicate that relation descriptions do not consistently improve performance, and show that model performance can be sensitive to variations in entity representations.

cs.AI

新论文 · v1arXiv:2611.20894

Which AI Use Cases Deserve a Pilot? A Risk and Value Framework for Funding, Scoping, and Scaling AI

Vishal Srivastava

原文摘要

A composite score can rank AI use cases while leaving the permitted operating scope and the evidence required for expansion unspecified. We treat the unit of decision as a use case at a proposed scope (authority, human mediation, limits, volume, environment, population) and define three decisions: screen the scope, fund the experiment, and authorize expansion only when residual risk is acceptable and evidence of incremental net value applies at that scope. A risk track (R1-R4) is implemented as an explicit decision table with vetoes and documented precedence, and a value track follows four evidence stages with each transition needing its own evidence. Funding uses expected value of perfect information as an upper bound and the value of sample information for the specific experiment. A quantified expected-loss check, required where policy says so, computes the containment an expansion needs and credits a control only on evidence measured at the proposed scope; in a synthetic scenario the highest-value control design fails that test while a smaller scope passes. A reproducible Python package with 279 passing tests demonstrates the framework on synthetic refund, reconciliation and knowledge-assistant use cases, including a synthetic matched-scope fixture and a transferability stress test (value evidence is simulated, not realized), and compares it with a public AWS sample. Thresholds and appetite are prototype policy requiring field validation.

cs.AI

新论文 · v1arXiv:2611.20891

From Recursion to Convolution: Parallel Self-Improvement for Multi-Agent Systems

Zheng Wang, Yuang Liu, Sachith Abeywickrama, Yangkai Ding, Tao Yu, Xinyu Zhang, Hongbo Wang, Qi Ye 等 10 位作者

原文摘要

Multi-agent self-improvement enables agents to improve through interaction and feedback. However, sequential dependencies across agents and refinement rounds can propagate errors, increase computational costs, and limit parallelism. Moreover, refining individual features or agents in isolation can overlook their dependencies and restrict collective coordination. To address these challenges, we propose ConvRSI, a framework for parallelizing multi-agent self-improvement through convolution. ConvRSI couples intra-agent convolution, which jointly refines related agent features, with inter-agent graph convolution, which aggregates messages according to the collaboration topology. Both stages operate in latent space, enabling parallel updates without repeatedly decoding and re-encoding intermediate messages. Experiments on MATH-500 show that ConvRSI achieves 5.04% performance improvement while reducing 17.27% inference cost, and 45.11% latency compared with RecursiveMAS. It also supports diverse collaboration patterns by representing their topologies as adjacency matrices. These results demonstrate the potential of convolutional co-evolution for efficient multi-agent self-improvement. Our code and datasets are provided in https://github.com/LYA2333/ConvRSI.

cs.AI · cs.CL · cs.LG