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

+1680新增论文
+1333其中 AI 相关
+187已有论文修订
+59新增 AI 导读

arXiv 最近完成扫描:2026-11-18 04:28 · 全站最近入库:2026-11-17 23:35各信源进度 →

统计口径与数据新鲜度

统计 2026-11-17 10:04 — 2026-11-18 10:04(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:2026-11-18 04:28最近发现新增版本:2026-11-17 23:35库内论文最新更新时间:2026-11-17 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。

25,786 已收录论文21,491 AI 相关7,725 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 5,970 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-18 04:27(北京时间)
新论文 · v1arXiv:2611.19525

Objectify, Don't Delete: Complaint-Derived Evaluation of Ambient Clinical Notes

Nikita Gupta, Narayanan Unny Edakunni, Vikram Okaly, Suranjana Datta, Navya Penubadi, Karl Eric Swanson, Siddharth Sharma

原文摘要

Evaluating the quality of AI-generated clinical notes is hard: standard metrics like ROUGE measure lexical similarity to reference notes, while expert checklists encode what committees believe matters. Neither captures what clinicians actually complain about. Recent work has shown that evaluation criteria can be mined from user feedback and filtered to criteria that LLM judges can reliably enforce. We argue that this is where much of the value is lost. The criteria that fail enforceability are disproportionately the ones clinicians raise most often, because high-frequency complaints are stated as several questions glued together. No yes/no judgment can grade such composites as posed. We introduce an objectification step that rewrites such criteria into sets of scoped, determinate decisions. We demonstrate the approach through an evaluation pipeline that mines criteria from complaints, objectifies them, and deploys LLM judges against both planted defects and production notes. Across eleven defects planted into ACI-bench reference notes, objectified judges localize violations with median recall of 0.95, while reference metrics change by at most 0.037. Given the complaint as one broad question, the same judges still find every planted defect, but they also flag 70% of the unperturbed reference notes; one section-pair question per call raises Youden's J from 0.30 to 0.72 (by 0.19 to 0.26 with prompt effort matched). We release ACI-Defect, a perturbation benchmark; clinician-reviewed instance-level labels showing quality gaps in the ACI-bench gold standard (32 of 40 notes contain hallucinations; 40 of 40 omit discussed facts); and grounding-judge verdicts. On 1,446 production notes, mean instance-rate burden falls at every clinician-rating step (rho=-0.15), a small association. Defect counts have to be kept per span rather than collapsed into one pass/fail bit per note.

cs.AI · cs.CL

版本更新 · v3arXiv:2601.08676

Advancing ESG Intelligence: An Expert-level Agent and Comprehensive Benchmark for Sustainable Finance

Yilei Zhao, Wentao Zhang, Lei Xiao, Yandan Zheng, Mengpu Liu, Wei Yang Bryan Lim

原文摘要

Environmental, social, and governance (ESG) criteria are essential for evaluating corporate sustainability and ethical performance. However, professional ESG analysis is hindered by data fragmentation across unstructured sources, and existing large language models (LLMs) often struggle with the complex, multi-step workflows required for rigorous auditing. To address these limitations, we introduce ESGAgent, a hierarchical multi-agent system empowered by a specialized toolset, including retrieval augmentation, web search and domain-specific functions, to generate in-depth ESG analysis. Complementing this agentic system, we present a comprehensive three-level benchmark derived from 310 corporate sustainability reports, designed to evaluate capabilities ranging from atomic common-sense questions to the generation of integrated, in-depth analysis. Empirical evaluations demonstrate that ESGAgent outperforms state-of-the-art closed-source LLMs with an average accuracy of 84.15% on atomic question-answering tasks, and excels in professional report generation by integrating rich charts and verifiable references. These findings confirm the diagnostic value of our benchmark, establishing it as a vital testbed for assessing general and advanced agentic capabilities in high-stakes vertical domains.

cs.AI

新论文 · v1arXiv:2611.19499

Do Probing and Steering Favor the Same Directions Within and Across Language Models?

Afshin Sadeghi, Daniel Becker

原文摘要

A direction in a language model's hidden state can be used to probe whether a concept is present or to steer the model toward it. Although probe directions are often reused for steering, the direction that best detects a concept need not be the one that best changes it. We study this distinction in three instruction-tuned models from different families. We construct a one-parameter family of directions ranging from a covariance-adjusted discriminant to the mean difference, and evaluate probing and additive steering, which adds the direction to the hidden state, along the family at a fixed side effect. For truthfulness and sentiment, in every model, the best probe is not the best steering direction: steering consistently favors directions closer to the mean difference. This separation holds across 59 of 60 layer-model-concept combinations in a layer sweep, with matched probe and steering texts in all six model-concept combinations, and under a rotation-based steering operator. The exact steering optimum is less stable: for sentiment, an intermediate direction produces a 27-56% stronger steering effect than the mean difference, but this gain shrinks or disappears with matched texts and depends on the layer and the operator. The two roles also transfer differently across models. A probe transferred by a linear map fitted on neutral text retains 81-91% of the target model's above-chance probing performance, while the transferred mean difference retains less of its steering effect in 11 of 12 model pairs. This is a consistent tendency, although our predefined criteria classify only three pairs, and nine remain inconclusive. Finally, as an application on PopQA, steering toward "unknown" increases abstention by 18-44 percentage points on uncertain errors, versus 2-5 on confident errors, while rarely correcting errors in cases where a correct answer is reachable.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.19472

InvarOrd: Self-Supervised Ordinal Knowledge Distillation with Rank-Invariant Alignment for Label-Efficient Cross-Species Hyperspectral Fish Freshness Assessment

Kazi Nabiul Alam, Omenuko Joseph Nwebonyi, Pooneh Bagheri Zadeh, Akbar Sheikh-Akbari

原文摘要

Hyperspectral models of fish freshness rarely transfer between species, and labelling a new species is slow and destructive. Freshness is also ordinal, and standard domain alignment can erase that order. We present InvarOrd, which adapts to a new species from 30 labelled cubes (5 per rank) in three stages: self-supervised spectral pretraining on unlabelled source cubes, cross-species alignment by a shrinkage covariance discrepancy computed only within matched ranks, and ordinal distillation into a compact student. The alignment term is provably invariant to the rank means, so it gives no incentive to collapse the ordinal axis. On paired salmon and cod data (800 cubes each, 16 days), a source-trained teacher reaches 1.6 to 1.9 days mean absolute error (MAE) on the other species, with quadratic weighted kappa (QWK) below zero. The 423K-parameter student, 2.54 times smaller than its teacher, attains the lowest MAE (1.476 days) and highest within 1-day accuracy (56.1%) among 14 configurations, matching the best of six ImageNet-pretrained backbones (ConvNeXt-T, 1.503 days) at 66 times fewer parameters, 1.44ms per cube on a GPU and 5s re-adaptation. InvarOrd beats target-only training by 0.29 days (p = 0.031). Among alignment objectives, rank-invariant alignment retains the most ordinal structure (embedding rank correlation 0.183, QWK 0.112), whereas global discrepancy lowers the rank correlation to 0.049 and QWK to zero at only 0.014 days higher MAE.

cs.CV · cs.AI

版本更新 · v3arXiv:2609.06563

MARBO: Relational Belief Grounding for LLM Agents in Social Deduction Games

Yechan Hwang, Sangjun Bae, Jeongmo Kim, Sangwoo Bang, Seungyul Han

原文摘要

Social deduction games (SDGs) require agents to reason under partial observability by maintaining relational beliefs about hidden roles and team alignments. While recent LLM-agent approaches improve gameplay through prompting and preference optimization, they often optimize actions and in-game speech without explicitly grounding them in such beliefs. This frequently leads to strategically inconsistent behavior, especially for compact LLM agents. We introduce Multi-Agent Relational Belief Optimization (MARBO), a belief-grounded preference optimization framework that leverages relational beliefs to guide strategic decisions and in-game speech. MARBO provides preference feedback only when behaviors are supported by reliable relational beliefs and lead to strategically favorable social outcomes, encouraging more consistent learning under uncertainty. Experiments on representative SDGs show that MARBO enables compact LLM agents to consistently outperform existing baselines. The Code is available on https://github.com/PleaseTakemeAway/MARBO.

cs.AI · cs.LG

新论文 · v1arXiv:2611.19465

VAST: Compiling Grounded Vulnerability Scenarios into Verifiable, Multi-Step Environments for Evaluating LLM Red-Team Agents

Zhen Wan

原文摘要

Evaluating Large Language Model (LLM) security agents requires reproducible environments whose workflow difficulty can be varied and whose failures can be diagnosed. We present VAST (Verifiable Adversarial Scenario Training & Evaluation), a declarative range compiler separating vulnerability scenarios from a compiler-introduced workflow-hardening layer. Following a post-hoc code and AST audit across all topics, the main benchmark retains 109 seed edges (45 with strict upstream CVE annotations), 72 family identifiers (62 under shared-HMAC relay grouping), 81 topic configurations (76 core plus 5 bounded execution models), and 1,387 instances (359 canonical four-gate-revalidated specs plus 1,028 seed variants) in a 42:39 partition. On the TopicServer compiler stratum, a budget-matched 270-episode prompt contrast across 10 compiled mechanism-family clusters (30 P2/P3/P4 instances, 40 turns, temperature 0.0) shows that glm-5.3, grok-4.7, and deepseek-v4-pro solve 12/90 instances under contract_only and 14/90 under generic_hint (cluster sign p=1.00), whereas supplying the shared workflow protocol in full_guide lifts completion to 76/90 (28/30, 23/30, 25/30; cluster sign p<=0.00391). On the separate ranges_v2 ledger, a 126-episode evaluation across 6 non-TopicServer workflows and multi-container Docker/Linux stacks solves 88/126 (69.8%) (28/42, 31/42, 29/42), while an 18-episode intervention on asynchronous Redis worker chaining (v2_02) lifts solves from 0/18 to 18/18 (100%) in 4-6 turns when installed worker modules are disclosed in GET /sched/directory. Finally, a three-seed (s in {42, 1024, 2025}) LoRA recipe comparison on Qwen2.5-Coder-7B-Instruct across 37 lineage-disjoint families (231 episodes on 33 held-out instances) yields 0/33 for Base, 1.67 +/- 2.89 / 33 for Golden SFT, and 4.67 +/- 2.31 / 33 for Header-Recovery SFT, with all solves on v2_03.

cs.CR · cs.AI

新论文 · v1arXiv:2611.19461

Program Distillation: Transferring LLM Capability from System 2 to System 0

Pujun Zheng, Wanying Ren, Guoxiu He

原文摘要

Recent studies leverage large language models (LLMs) to construct executable programs through environment interaction, feedback analysis, and code revision. However, prior work has mainly treated LLMs as optimizers, overlooking their ability to transfer knowledge across system levels. We propose Program Distillation (PD), a framework that formalizes the transfer of abilities from System 2 to System 0. During learning, the LLM provides task knowledge and draft strategies; at deployment, a student program executes the resulting policy. We show that this approach can improve performance while reducing inference time. We introduce PDBench, a benchmark comprising 27 discrete- and continuous-control tasks with fixed initial policies, and PDMetaHarness, a model-harness-independent meta-system for monitoring and guiding the learning process. The strongest model achieves a normalized mean score of 0.728, surpassing PPO (0.656) and Jev-like decision models (0.239), while running exclusively on CPUs and reducing inference costs. Our analysis shows that performance is largely driven by the first breakthrough during learning, and structural modifications have a slightly lower success rate than parameter modifications but typically yield larger gains. We further introduce style distillation, which enables interpretable stylistic control over the learning process. Overall, we present an analytical framework for PD and demonstrate its practical viability.

cs.AI

新论文 · v1arXiv:2611.19448

CLEW: Log-Anchored Backward Tracing for Root-Cause Code Localization

Wenji Bai, Muhammad Waseem, Jaakko Peltonen, Pekka Abrahamsson

原文摘要

Failures in industrial software systems may involve multiple components and be difficult to reproduce outside the production environment. Root-cause code localization therefore often relies on issue descriptions and runtime logs. However, log evidence may be noisy or incomplete, and the code reporting a symptom may differ from the code containing the underlying defect. These conditions make it challenging to identify relevant log evidence and determine which functions could have caused the reported failure. To address this gap, we introduce CLEW, an agentic approach that combines rule-based log-to-code anchoring with LLM-guided backward code tracing. CLEW organizes selected observations into diagnostic states, guides source exploration from anchored symptom functions, and returns ranked candidate functions, an affected module, and an explanation comprising a proposed mechanism path and associated log events. For evaluation, we construct LogLoc, a benchmark comprising 192 real-world issue and fault-injection cases. Against three baselines, CLEW records the highest observed Acc@1 (77.3%) and MRR (81.3%) on LogLoc-Issues and matches the leading baselines at 97.6% Acc@1 on LogLoc-Airflow. Averaged over all 192 cases, CLEW's wall-clock time is approximately 73% lower than LocAgent's at a similar average inference cost. No aggregate degradation is observed under the tested perturbations, with modest changes in Acc@1 and MRR across the evaluated seeds and base models. These results suggest that CLEW can combine function-level localization with evidence links and runtime efficiency under the evaluated conditions.

cs.SE · cs.AI

版本更新 · v2arXiv:2602.12083

Differentiable Modal Logic for Multi-Agent Diagnosis, Orchestration and Communication

Antonin Sulc

原文摘要

As multi-agent AI systems evolve from simple chatbots to autonomous swarms, debugging semantic failures requires reasoning about knowledge, belief, causality, and obligation, precisely what modal logic was designed to formalize. However, traditional modal logic requires manual specification of relationship structures that are unknown or dynamic in real systems. This tutorial demonstrates differentiable modal logic (DML), implemented via Modal Logic Neural Networks (MLNNs), enabling systems to learn trust networks, temporal orders, and regulatory boundaries from behavioral data. We present a unified neurosymbolic debugging framework through four modalities: epistemic (who to trust), temporal (what can follow what), deontic (what actions are permitted), and doxastic (who believes what). Each modality is demonstrated on concrete multi-agent scenarios (from exposing an attacker in a drone swarm to deciding when to consult another LLM code reviewer), with complete implementations showing how logical contradictions become learnable optimization objectives. Key contributions for the neurosymbolic community: (1) interpretable learned structures where trust and temporal order are explicit parameters, not opaque embeddings; (2) knowledge injection via differentiable axioms that guide learning with sparse data; (3) compositional multi-modal reasoning that combines epistemic and doxastic constraints; and (4) practical deployment patterns for monitoring, active control and communication of multi-agent systems. All code is provided as executable Python scripts.

cs.AI · cs.LO

版本更新 · v4arXiv:2609.08788

Adaptive Anisotropic Attention for Axis-Structured Signals

Mahir Jain, Parshva Runwal, Aditya Ray Mishra, Raj Singh Bani, Arvasu Kulkarni, Sandeep Singh, Siddharth Panwar

原文摘要

Dense self-attention treats all token pairs as equally plausible before learning, an interaction-isotropic prior that can be mismatched to structured signals. For structured, low signal-to-noise ratio (SNR) signals such as EEG, dependencies are organized along the electrode and time axes, and this uniform prior exposes each token to many irrelevant interactions. We introduce Adaptive Anisotropic Attention (AAA), which splits attention into two paths: a temporal path, where each token attends to the tokens of its own electrode across time, and a spatial path, where it attends to the tokens of the other electrodes at the same time step. A small gate predicts, for every token, a convex combination of the two path outputs: two non-negative weights that sum to one. On six EEG downstream tasks, the resulting model, AXON (AXis-factorized Operator Network), improves mean balanced accuracy over a dense baseline under both linear probing and full fine-tuning. We show that both paths (temporal and spatial) are necessary and that the weighted sum beats a hard choice of one path; most of the benefit comes from the gate learning a different temporal/spatial balance at each layer of the network. Controlled audio spectrogram experiments show that axis factorization transfers beyond EEG. These results suggest that aligning attention with the natural axes of structured signals provides a useful inductive bias.

cs.LG · cs.AI

版本更新 · v5arXiv:2303.17707

Stop Using Plausibility as the Criterion for Explainable AI

Weina Jin, Xiaoxiao Li, Ghassan Hamarneh

原文摘要

Explainable artificial intelligence (XAI) is motivated by the problem of making AI predictions understandable, transparent, and responsible, as AI becomes increasingly impactful in society and high-stakes domains. The evaluation and optimization criteria of XAI are gatekeepers for XAI algorithms to achieve their expected goals and should withstand rigorous inspection. To improve the scientific rigor of XAI, we conduct the first comprehensive and critical examination of a common XAI criterion: plausibility. Plausibility assesses how convincing the AI explanation is to humans, and is usually quantified by metrics of feature localization or feature correlation. Our examination shows that plausibility is an invalid measure of explainability, and human explanations are not ground truth for XAI, because both ignore the necessary assumptions of explanation. One fundamental assumption is that AI explanations are supposed to behave adversarially, analogous to auditors' or doctors' roles of identifying potential problems. Evaluating or optimizing XAI to generate more plausible explanations is analogous to incentivizing (i.e., bribing) auditors/doctors to generate more positive reports, which is different from encouraging AI models to learn more plausible features (analogous to encouraging companies/patients to improve their conduct/health). Our examination further reveals the consequences of using plausibility as the XAI criterion, including increasing misleading explanations that manipulate users, deteriorating users' trust in the AI system, undermining human autonomy, being unable to achieve complementary human-AI task performance, and abandoning other possible approaches of enhancing understandability. Due to the invalidity of measurements and the unethical issues, this paper argues that the AI community should stop using plausibility as the criterion to evaluate and optimize XAI algorithms.

cs.AI · cs.HC

新论文 · v1arXiv:2611.19439

Do Rule-Based Tools Help LLMs Translate Endangered Uralic Languages in the Prompt or over MCP?

Iaroslav Chelombitko, Mika Hämäläinen

原文摘要

Komi-Zyrian, Moksha, Erzya and Udmurt have mature rule-based resources, among them finite-state analysers, constraint grammars and bilingual dictionaries, which language models do not use. We ask whether giving a model access to these resources improves machine translation from the four languages into English, and whether it matters how the access is provided: as analyses computed in advance and written into the prompt, or as tools that the model calls itself through the Model Context Protocol (MCP). We evaluate manually: an expert linguist read every English output of the strongest model under three conditions, 816 sentences in all, accepted or rejected each one, and named the main error in every rejected translation. With the analyses in the prompt, the number of accepted translations rises from 13 to 43 out of 272; with the tools attached it rises to 55. The gain comes from the bilingual dictionary. For Udmurt, whose dictionary has no English side, the prompt-based condition brings no gain at all, while for the other three languages an English gloss reaches the prompt for 74 to 80% of the source words. Tool calling is also much harder to get working than prompt injection. Of 31 open-weight models, 14 never call a tool, 11 of them because they cannot produce a usable call, and half of the calls that are made name a language the toolkit does not support. We release the data set, every model output and all tool-call logs.

cs.CL · cs.AI

版本更新 · v5arXiv:2601.00821

Fidelity Before Structure: Verbatim Chunks Beat Lossy Artifact Extraction in Long-Conversation LLM Memory

Tao An

原文摘要

A growing class of conversational-memory systems compresses dialogue history into structured artifacts (extracted facts, decisions, or events). The premise is that distilled structure retrieves better than raw text. We test this premise with a controlled ablation. Within one fixed retrieval--rerank--reasoning pipeline, we swap only the stored representation (LLM-extracted typed artifacts versus verbatim conversation chunks) and hold the model, retriever, reranker, and judge constant. Verbatim chunks win by 15.9 points on LoCoMo (43.9% vs. 28.0%) and by 22.0 points on LongMemEval-S (67.4% vs. 45.4%). A semantic graph does not recover the gap, and six confound controls reproduce the effect. The mechanism is lossy distillation rather than structure per se. Accuracy tracks how much source text survives in the store, and the extracted-artifact pipeline does not beat naive RAG in overall accuracy (chunks do abstain worse; see Limitations). For the extraction designs we test, structured memory should augment verbatim text rather than replace it. Adding artifacts alongside chunks costs no detected accuracy, while substituting them forfeits the gap. Code and data are available at https://github.com/tao-hpu/cog-canvas.

cs.AI · cs.CL · cs.IR

版本更新 · v2arXiv:2610.28907

JaxMARL-BC: Multi-Agent Reinforcement Learning for Macroeconomic Modelling, in Seconds

Daniele Montagnani, Andrea Coletta, Antonietta Mira, Aldo Glielmo

原文摘要

Multi-agent reinforcement learning (MARL) offers a principled way of populating simulated economies with many heterogeneous decision makers. Its adoption, pioneered in finance, has nonetheless been slow in macroeconomics because of two main limitations: the lack of a shared frame of reference connecting MARL systems to dynamic stochastic general equilibrium (DSGE) models the profession already trusts, and the very long training times. The first limitation was recently addressed by the Multi-Agent Reinforcement Learning Business Cycle (MARL-BC), which places deep RL households in a real business cycle economy and reduces to classic DSGE models in specific limits. Here we address the second limitation. We present JaxMARL-BC, a JAX-accelerated simulator that trains MARL-BC economies in seconds (between 75x and 724x faster than the reference implementation), and scales to 200,000 independent learning households. We verify its correctness against the known DSGE limits, characterise its scaling through a simple cost law, and demonstrate its value by a proof-of-concept calibration of an extended model with heterogeneous returns on capital, in a two-hour experiment that would previously have taken over a month. We release JaxMARL-BC as open source.

cs.MA · cs.AI · cs.LG · econ.GN

新论文 · v1arXiv:2611.19429

Prepare Lighter and Faster for Open-Vocabulary Queries: A Query-Wise 3D Segmenter for Gaussian Splatting

Nhat-Quynh Le-Pham, Khoi Nguyen, Binh-Son Hua

原文摘要

Open-vocabulary 3D understanding aims to localize and segment arbitrary objects beyond predefined categories. However, existing approaches based on Gaussian Splatting often rely on expensive preprocessing and lengthy training, limiting their applicability when fast and lightweight solutions are desired. We propose a lightweight, training-free framework for query-driven 3D instance segmentation on Gaussian Splatting. Instead of conventional mask-wise CLIP embeddings, we use dense pixel-level CLIP features to preserve fine-grained semantic information and enable flexible text-driven localization without relying on predefined instance proposals. Given a text query, we identify a high-quality target region in a carefully selected pivot view and propagate it across views using a video segmentor, providing consistent and comprehensive coverage of the target instance. We then lift the resulting 2D tracklets to 3D Gaussians through a novel visibility voting strategy, yielding accurate 3D segmentation masks with clean object boundaries. Extensive experiments demonstrate state-of-the-art performance on LERF-OVS, ScanNet, and ScanNet++, competitive results on 3D-OVS, and significant speedups over existing open-vocabulary 3D segmentation methods. These results highlight the effectiveness of our framework for efficient open-vocabulary 3D understanding without per-scene training.

cs.CV · cs.AI · cs.GR

新论文 · v1arXiv:2611.19419

Do Pretraining Optimizer Gains Transfer to KL Distillation?

Lavinia Ghita, Dhruv Desai

原文摘要

Matrix-structured optimizers can improve language-model pretraining, but it is unclear whether these gains transfer to short-horizon KL distillation from an instruction-tuned initialization. We compare AdamW and diagonal AdaGrad with two tractable Shampoo reductions, SOAP, and All-2D Muon across four dense Qwen student-teacher pairs under a common forward-KL objective and fixed training budget. Across scales, none of the evaluated matrix-structured recipes consistently improves the quality versus trainer-time trade-off over the elementwise references: AdamW generally attains the lowest validation KL, while AdaGrad can be slightly faster at larger scale. SOAP often tracks AdamW in KL but incurs substantial trainer-time overhead, and the two Shampoo reductions reverse ordering across model pairs, showing that conclusions depend on the feasible recipe. These results suggest that pretraining optimizer gains do not transfer automatically to KL distillation and should be re-established for the target objective and systems regime.

cs.LG · cs.AI · cs.CL

版本更新 · v5arXiv:2510.13939

Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers

Tuhin Chakrabarty, Jane C. Ginsburg, Paramveer Dhillon

原文摘要

The use of copyrighted books for training AI has sparked lawsuits from authors concerned about AI generating derivative content. Yet whether these models can produce high-quality literary text emulating authors' voices remains unclear. We conducted a preregistered study comparing MFA-trained writers with three frontier models (ChatGPT, Claude, Gemini) writing up to 450-word excerpts emulating 50 award-winning authors' styles. In blind pairwise evaluations by 28 MFA-trained readers and 516 college-educated general readers, AI text from in-context prompting was strongly disfavored by MFA readers for stylistic fidelity (OR=0.16) and quality (OR=0.13), while general readers showed no fidelity preference (OR=1.06) but favored AI for quality (OR=1.82). Fine-tuning ChatGPT on authors' complete works reversed these results: MFA readers favored AI for fidelity (OR=8.16) and quality (OR=1.87), with general readers showing even stronger preference (fidelity OR=16.65; quality OR=5.42). Both groups preferred fine-tuned AI, but the writer-type X reader-type interaction remained significant (p=0.021 for fidelity; p<10^-4 for quality), indicating general readers favored AI by a wider margin. Effects are robust under cluster-robust inference and generalize across authors in heterogeneity analyses. Fine-tuned outputs were rarely flagged as AI-generated (3% vs. 97% for prompting) by leading detectors. Mediation analysis shows fine-tuning eliminates detectable AI quirks that penalize in-context outputs, altering the nexus between detectability and preference. While not accounting for effort to transform AI output into publishable prose, the median fine-tuning cost of $81 per author represents a 99.7% reduction versus typical writer compensation. Author-specific fine-tuning enables non-verbatim AI writing preferred over expert human writing, providing evidence relevant to copyright's fourth fair-use factor.

cs.CL · cs.AI · cs.CY

新论文 · v1arXiv:2611.19413

PMR: MoE Routing with Shared Prototypes and Congestion Pricing

Wending Wang, Binqi Huang, Ju'an Li, Yu Guo, Lei Duan, Yuqi Ouyang, Dengchun Li

原文摘要

Conventional linear routers in sparse mixture-of-experts models select experts directly from token-expert matching scores, coupling input matching with expert allocation. We propose Prototype-Mediated Routing (PMR), a two-stage router with shared prototypes. PMR first computes each token's distribution over prototypes, then uses it to mix prototype-conditioned expert distributions according to the law of total probability. This formulation models input matching and expert allocation separately, explicitly represents many-to-many associations between prototypes and experts, and provides separate interfaces for context adaptation and load regulation. Within this framework, short-term states summarize preceding tokens to adapt the prototypes, while congestion prices updated from dispatch feedback achieve loss-free load balancing by adjusting the conditional expert distributions. We use a two-pass blockwise approximation to accelerate short-term state processing, achieving over 20 times the training throughput of the eager tokenwise PMR implementation while maintaining high agreement in routing decisions. Across three pretraining configurations with 0.43B, 1.18B, and 0.48B total parameters, PMR improves mean zero-shot accuracy on seven benchmarks over linear routing baselines with matched expert architectures by 1.54, 2.65, and 1.45 percentage points (pp), respectively. The corresponding gains on ARC-Easy are 6.71, 5.80, and 3.92 pp. In parameter-efficient fine-tuning of 7B-8B backbones, PMR's improvements over MixLoRA vary across backbones, with gains of up to 1.76 pp in mean accuracy across four tasks.

cs.AI · cs.LG

版本更新 · v2arXiv:2608.28327

How Diverse Are Layered Safeguards? Estimands and Measurement for Ensembles of LLM Jailbreak Defenses

Abrar Alotaibi, Moataz Ahmed

原文摘要

Practitioners protect large language models (LLMs) by stacking safeguards: a likelihood filter, a guard classifier, hidden-state probes, a decoding-time intervention. Such a stack is an ensemble of learned detectors, and ensemble theory states when combination pays: when members fail on different inputs. Stacking is recommended throughout the LLM security literature, but this condition has not been measured for any pair of safeguards. The condition is three quantities, not one, and conflating them is why it has resisted measurement. Separately optimized attacks license only a claim about which \emph{behaviors} are jointly vulnerable; the identity that governs a deployed stack is written over shared \emph{inputs}; and neither predicts what an adversary attacking the assembled pipeline achieves. We define the three estimands, give an estimator for each, and supply two controls: the correlation between two independent runs of the same attack, which sets the level a pair must clear, and a multilevel model separating detector diversity from behavior difficulty. On a seven-layer stack and two open-weight targets, no pair of safeguards is distinguishable from the correlation that two independent runs of the attack produce with no safeguard present, and behavior difficulty explains much of what remains. A second attack family, an attacker model rewriting its own prompts, reaches that verdict independently on a thirtieth of the budget and overturns the practical one: the only member that blocked the first adversary admits 0.19 under the second, against 0.03. Diversity between neural safeguards is measurable, but it and any member's strength are properties of a (stack, adversary, target) triple.

cs.CR · cs.AI · cs.CL

新论文 · v1arXiv:2611.19403

Confidence Without Cause: Predicting and Localizing Physics-Shortcut Sensitivity in Language Model Hidden States

Shubham Mishra, Atharv Sharma, Dhruv Kumar

原文摘要

A language model's confidence on a physics problem deserves trust only if it ignores details irrelevant to the answer. We introduce DECOY (Distractor-Embedded Certified cOunterfactuals for phYsics): 155 families of four prompts differing in one irrelevant detail. A symbolic solver certifies that each edit preserves the answer. Families are labelled by the largest drop in the model's log-probability of the certified answer, using a threshold fixed before probe training. Qwen2.5-7B-Instruct is sensitive on 50.0% of the 140 families yielding labels, and Llama-3.1-8B-Instruct on 22.9%. Effects are largest when distractor values lie numerically close to the answer. A linear probe on Qwen's layer-18 hidden state, averaged over four renderings, predicts sensitivity before generation at AUROC 0.760 with whole families held out. A separately tuned single-rendering probe reaches 0.747. A frozen layer-16 probe reaches 0.820 pooled AUROC on 15 families built after training, though within-cue-type AUROC is lower (0.667-0.6875). Under a paired family bootstrap, the probe performs on par with a fitted token-entropy baseline (AUROC 0.727) and achieves the lowest Brier score among tested predictors (0.204 versus 0.214). Knocking out four attention heads at layer 16 cancels sensitivity on a 20-family panel, localising it to a narrow mechanism. These results show that Qwen's confidence can track irrelevant details, that this failure is detectable before generation, and that targeted intervention can suppress it. DECOY provides certified ground truth and paired-tested baselines for evaluating hidden-state monitors.

cs.AI · cs.LG

版本更新 · v3arXiv:2511.15090

SciEGQA: A Dataset for Scientific Evidence-Grounded Question Answering and Reasoning

Wenhan Yu, Zhaoxi Zhang, Wang Chen, Guanqiang Qi, Weikang Li, Lei Sha, Deguo Xia, Jizhou Huang

原文摘要

Scientific documents contain complex multimodal structures, which makes evidence localization and scientific reasoning in Document Visual Question Answering particularly challenging. However, most existing benchmarks evaluate models only at the page level without explicitly annotating the evidence regions that support the answer, which limits both interpretability and the reliability of evaluation. To address this limitation, we introduce SciEGQA, a scientific document question answering and reasoning dataset with semantic evidence grounding, where supporting evidence is represented as semantically coherent document regions annotated with bounding boxes. SciEGQA consists of two components: a **human-annotated fine-grained benchmark** containing 1,623 high-quality question--answer pairs, and a **large-scale automatically constructed training set** with over 30K QA pairs generated through an automated data construction pipeline. Extensive experiments on a wide range of Vision-Language Models (VLMs) show that existing models still struggle with evidence localization and evidence-based question answering in scientific documents. Training on the proposed dataset significantly improves the scientific reasoning capabilities of VLMs. The project page is available at https://yuwenhan07.github.io/SciEGQA-project/.

cs.DB · cs.AI · cs.CV

版本更新 · v4arXiv:2511.07979

Benchmarking Multi-Step Legal Reasoning and Analyzing Chain-of-Thought Effects in Large Language Models

Wenhan Yu, Xinbo Lin, Lanxin Ni, Jinhua Cheng, Lei Sha

原文摘要

Large language models (LLMs) have demonstrated strong reasoning abilities across specialized domains, motivating research into their application to legal reasoning. However, existing legal benchmarks often conflate factual recall with genuine inference, fragment the reasoning process, and overlook the quality of reasoning. To address these limitations, we introduce MSLR, the first Chinese multi-step legal reasoning dataset grounded in real-world judicial decision making. MSLR adopts the IRAC framework (Issue, Rule, Application, Conclusion) to model structured expert reasoning from official legal documents. In addition, we design a scalable Human-LLM collaborative annotation pipeline that efficiently produces fine-grained step-level reasoning annotations and provides a reusable methodological framework for multi-step reasoning datasets. Evaluation of multiple LLMs on MSLR shows only moderate performance, highlighting the challenges of adapting to complex legal reasoning. Further experiments demonstrate that Self-Initiated Chain-of-Thought prompts generated by models autonomously improve reasoning coherence and quality, outperforming human-designed prompts. MSLR contributes to advancing LLM reasoning and Chain-of-Thought strategies and offers open resources for future research. The dataset and code are available at https://github.com/yuwenhan07/MSLR-Bench and https://law.sjtu.edu.cn/flszyjzx/index.html.

cs.AI

版本更新 · v2arXiv:2304.04308

Ensemble Modeling for Time Series Forecasting: an Adaptive Robust Optimization Approach

Léonard Boussioux, Henry Mao, Dimitris Bertsimas

原文摘要

Accurate time-series forecasting is critical for many problems involving temporal data. Ensemble modeling is a well-established technique that leverages multiple predictive models to improve accuracy and robustness, since a single predictor's performance can vary widely due to shifts in the underlying data distribution. This paper proposes a new methodology for building robust ensembles of time series forecasting models. Our approach uses Adaptive Robust Optimization (ARO) to construct a linear regression ensemble whose model weights adapt over time. We demonstrate the effectiveness of our method through synthetic experiments covering eight disturbance scenarios, from no drift to abrupt regime shifts, and real-world applications, including air pollution management, energy consumption forecasting, and tropical cyclone intensity forecasting. In the real-world applications, our adaptive ensembles outperform the best ensemble member in hindsight by 16-26% in root mean square error (RMSE) and 14-28% in conditional value at risk at the 15% level and improve over the other ensemble methods we compare on most metrics. In the synthetic experiments, they have lower mean RMSE than a rolling-ridge comparator in seven of the eight scenarios, significantly in four, while an abrupt regime shift that first appears after model selection remains a limitation.

cs.LG · cs.AI · math.OC

版本更新 · v2arXiv:2512.16795

From Evidence to Judgment: Trace-Augmented Reasoning under Conflicting Evidence

Shubham Mishra, Atharv Sharma, Dhruv Kumar, Harsh Sharma

原文摘要

Retrieved documents often disagree, contain outdated information, or fail to support any answer, and retrieval-augmented generation (RAG) systems have no reliable way to decide which documents to trust. We treat this as a reasoning problem and test it separately from answer generation. We release 943 training and validation examples and a disjoint 736-example benchmark, and fine-tune four language models to write a structured think trace before answering: the model judges each document, infers the conflict pattern, and writes a grounded answer or a justified refusal. Our evaluation framework, CATS (Conflict-Aware Trust Score), scores grounded refusal, behavioural adherence, factual grounding, and single-truth recall; a human study validates factual grounding and single-truth recall (kappa=0.78 and 0.79). The think trace improves the answer/refuse decision: under all three prompts, untuned models score at most 0.429 Refusal F1 (F1 with refusal as the positive class) and trace-tuned models at least 0.826. On the answerable examples a model answers, factual grounding under the 40-word prompt is at most 0.11 for untuned and 0.86-0.88 for trace-tuned models. Cutting the prompt from 1,965 words to 40 leaves trace-tuned scores within the benchmark's measurement noise (+/-0.034). Answer-only fine-tuning trails trace supervision on every metric under the minimal prompt.

cs.CL · cs.AI · cs.CY · cs.IR

版本更新 · v2arXiv:2611.00911

S1CogBench: What System One Decision Models Can and Cannot Do on Cognitive Experiments

Dengzhe Hou, Kazunori D Yamada

原文摘要

Decision-making in humans and AI involves multiple cognitive abilities. System One decision models produce fast, structured choices. Their cognitive strengths and limitations, however, remain poorly understood. We introduce S1CogBench, a benchmark of 24 established cognitive tasks and 27 evaluation modes, evaluating seven open models and the API model Jev. It combines human-response prediction, autonomous execution, and controlled task manipulations to reveal strengths and failures that aggregate scores miss. We find distinct behavioral profiles across the evaluated models. In these models, strong performance on some learning tasks coexists with failures to gather information or adapt to feedback, while high accuracy does not guarantee human-like responses to changing task demands. Shared readout controls show that model rankings depend partly on how choices are extracted. Separate latency comparisons with generative language models show that long responses are much slower than direct decisions, whereas matched one-token generation adds little overhead. To our knowledge, S1CogBench is the first benchmark of System One decision models on established cognitive experiments. It provides a common empirical basis for identifying their capabilities and limitations before integrating them into automated decision systems.

cs.CL · cs.AI · q-bio.NC

新论文 · v1arXiv:2611.19389

Pixel-KWS: Rendered-Glyph Representations for Orthographically Robust Open-Vocabulary Keyword Spotting

Adarsh Arigala, Ramesh Gundluru, Srinivasan Umesh, Ludovick Lepauloux, Hakim Hacid

原文摘要

Open-vocabulary keyword spotting (KWS) allows users to enroll arbitrary keywords at inference time. However, character and phoneme-based text encoders rely on fixed symbol vocabularies learned during training, limiting their ability to handle unseen or visually similar symbols such as homoglyphs, l33tspeak substitutions, and accented characters. We propose PIXEL-KWS, an open-vocabulary KWS framework that represents characters through rendered glyphs rather than learned embedding lookups. We render each character as a grayscale image and encode it with a lightweight CNN, enabling the text encoder to generalize from visual form rather than a fixed vocabulary. On LibriPhrase, PIXEL-KWS and its matched character-based counterpart achieve 9.57% and 9.55% EER on LibriPhrase-hard, 0.43% and 0.41% EER on LibriPhrase-easy, outperforming all published baselines without pre-training or synthetic data. PIXEL-KWS provides substantially greater robustness to both homoglyph and l33tspeak symbol-level perturbations. These results highlight glyph-based representations as a simple and effective approach for robust open-vocabulary KWS in practical settings.

eess.AS · cs.AI · cs.CV · cs.SD

新论文 · v1arXiv:2611.19388

SARGO: Agreement-Gated Reinforcement and Suppression via On-Policy Self-Guidance for Visual Hallucination Mitigation

Jinyu Li, Zhiming Zhang, Hao Fang, Hongyao Yu, Bin Chen, Albert Ma, Shu-Tao Xia

原文摘要

Vision-language models remain susceptible to visual hallucinations, often mixing grounded information with unsupported claims in the same response. Response-level reinforcement learning assigns a shared advantage to all tokens in a response, causing negative updates to suppress correct information alongside unsupported claims. To reduce hallucination while preserving useful visual information, we introduce SARGO (Self-guided Advantage-Routed Grounding Optimization), a post-training framework for selective reinforcement and suppression through on-policy self-guidance. Direct and reciprocal weighting gives better-supported tokens stronger reinforcement on positive updates and weaker penalties on negative updates. Agreement between overall and hallucination-specific advantages selects when to apply this guidance. At the response level, a reference-calibrated reward couples factuality with the retention and expansion of grounded information, discouraging lower hallucination scores obtained through omission. Experiments on Qwen3-VL-4B, Qwen3-VL-8B, and Gemma3-4B demonstrate that SARGO outperforms the evaluated mitigation baselines across Obj-Hal, AMBER, THRONE, and SHR evaluations. SARGO achieves these gains while preserving useful visual information and general multimodal capabilities.

cs.AI

新论文 · v1arXiv:2611.19385

Does a completed simulation run mean a usable urban wind result for AI agents?

Zhongke Qu, Yunwei Zhang, Junwei Su, Yabin Zhang, Chengwei Li, Qingxiang Shui, Hailong Qu, Zhaolin Gu

原文摘要

AI agents based on large language models (LLMs) can now configure and run computational fluid dynamics (CFD) solvers, but a completed run does not ensure that delivered metrics can be recomputed from raw outputs. This gap was examined for urban wind simulation, where multi-case comparisons must hold non-target conditions fixed and reported values must be traceable to raw samples. UrbanWind-AgentBench, comprising 200 tasks and 367 cases in 10 engineering task families, was constructed. Public task contracts specify variable factors, fixed conditions, numerical thresholds and deliverable requirements, and two-layer scoring separates a structural score for deliverables, run provenance and raw evidence from a numerical score obtained by independent recomputation against references. WindPilot was then proposed as an execution framework that incorporates contract parsing, execution provenance, deterministic verification and bounded acceptance review into the LLM execution loop. On 30 OpenFOAM tasks with independent references, at least the structural maximum of 60 points was reached by a base agent on 28 tasks, but numerical credit was obtained on only 6. With the same model name and budget, WindPilot increased the number of tasks with numerical credit to 17, full-score tasks from 0 to 3 and the mean task score from 62.5 to 72.5. Mean action steps fell from 130.4 to 104.9, and mean token consumption decreased by 12.9%. The mean gain for atmospheric boundary layer (ABL) tasks was 18.5 points. A completed simulation run is therefore not sufficient for a usable urban wind result; usability depends on checks that tie reported metrics to raw evidence during execution.

cs.AI

版本更新 · v2arXiv:2605.05166

The First Token Knows: Single-Decode Confidence for Hallucination Detection

Mina Gabriel, Pei Wang

原文摘要

Sampling-based uncertainty methods detect hallucinations by generating multiple answers and measuring agreement, often with additional semantic clustering. We study whether uncertainty is already visible in the first content-bearing answer token of a single greedy decode. We define first-token confidence, phi_first, as one minus the normalized entropy of the top-K first-token distribution. Across PopQA and TriviaQA with three 7-8B instruction-tuned models, phi_first matches or modestly exceeds semantic self-consistency while requiring only one decode instead of one greedy decode plus ten sampled generations and NLI clustering. A subsumption analysis shows that phi_first is moderately to strongly correlated with semantic agreement, and combining both signals yields only small additional AUROC gains. We position phi_first as a cheap deployment-oriented baseline for factual QA uncertainty estimation, while noting its limits for high-confidence falsehoods and long-form generation.

cs.CL · cs.AI

版本更新 · v2arXiv:2609.34960

ProofLoom: Proof-Obligation-Driven Theory Construction for Autoformalizing Research-Level Stochastic Optimization

Feiming Wang, Daibo Li, Kun Yuan

原文摘要

Formalizing research-level stochastic optimization in Lean requires both an algorithm model and domain theory connecting foundational libraries to convergence proofs. Revising a model to restore provability can change the mathematical claim. We introduce ProofLoom, a fully automated LLM-agent system for Proof-Obligation-Driven Theory Construction. Given a published algorithm, target theorem, and source proof, ProofLoom autonomously constructs the Lean model and supporting theory. Open proof obligations drive the development of definitions, interfaces, lemmas, and proof plans. Signature contracts record evidence and obligations for model revisions; an independent Judge rejects unsupported assumptions and weakened conclusions. Planner expands the published argument into intermediate claims, and Audit checks whether the Lean proof follows it. Across tasks, SOptLib accumulates verified mathematics and construction experience: reusable results are extracted, generalized, and verified, while modeling decisions and failed proof routes are recorded. Later tasks retrieve these results and records and contribute new developments, forming a cycle of construction, accumulation, and reuse. On fifteen textbook and research-paper tasks, ProofLoom obtains mean human ratings of 6.3/7 and 6.4/7, compared with 4.9/7 and 5.0/7 for the strongest of six baselines. Across 33 developments, it produces 490,693 lines of algorithm-local Lean code with no sorry. The formalizations also expose 28 incorrect formulas, proof gaps, and algorithm-analysis mismatches in published sources across 22 developments, each with checked evidence.

cs.AI · math.OC