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

+0新增论文
+0其中 AI 相关
+2已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-15 07:45 · 全站最近入库：2026-11-14 15:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-14 07:58 — 2026-11-15 07:58（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：2026-11-15 07:45最近发现新增版本：2026-11-14 15:09库内论文最新更新时间：2026-11-13 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。

23,348 已收录论文19,580 AI 相关6,729 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

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 3,930 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-15 07:44（北京时间）
新论文 · v1arXiv:2611.15225

Can MLLMs Reason over Surround-View Driving Videos? A Multi-View Autonomous Driving QA Benchmark

Shunming Liu, Jianyi Peng, Cheng Li, Lu Hou, Pei Huang, Yajing Sun, Huaxiao Liu

原文摘要

Surround-view video understanding in autonomous driving is a crucial capability for Multimodal Large Language Models (MLLMs) to achieve reliable driving reasoning. Real-world driving often requires information beyond the front-view camera. Existing MLLMs primarily focus on front-view information, while their spatio-temporal understanding and reasoning abilities remain unknown. To address this gap, we propose MVDriveBench, a high-information-density video QA benchmark for autonomous driving, characterized by synchronized surround-view inputs, spatio-temporal understanding, and reasoning-chain-oriented evaluation. We design 12 categories of QA tasks that cover the complete reasoning pipeline from low-level visual perception to high-level driving decision-making, enabling a systematic evaluation of MLLMs surround-view spatio-temporal understanding and reasoning capabilities in complex driving environments. We conduct a comprehensive evaluation of 23 open-source, closed-source, and expert MLLMs on MVDriveBench. The results show that current MLLMs still exhibit insufficient surround-view spatio-temporal reasoning ability. Meanwhile, closed-loop evaluations of specialist models in AlpaSim reveal a correlation between surround-view spatio-temporal reasoning capabilities and closed-loop driving performance, highlighting the importance of comprehensive multi-view understanding for reliable vehicle control.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.13675

OpticalRAG: Pixel-Space Compression for Token-Efficient Retrieval-Augmented Generation

Senhao Liu, Yuheng Zhang, Chunyu Wei, Yueguo Chen, Xinran Zhang

原文摘要

Retrieval-augmented generation (RAG) is bottlenecked by the token budget of large language models: feeding long documents into the context window is expensive, while aggressive textual compression discards fine-grained information that is irreversibly lost in a one-dimensional token sequence. We argue that this bottleneck is not fundamental to information density but to the choice of compression \emph{modality}. Modern visual encoders can pack the contents of a rendered text page into roughly a hundred visual tokens with little semantic loss, opening a new compression-fidelity tradeoff that purely textual methods cannot reach. We present \textbf{OpticalRAG}, a framework that renders documents as page images, encodes them with a frozen vision encoder, and distills the retrieved visual tokens into as few as $16$ injected tokens before passing them to the LLM. Two ingredients make this work without expensive cross-modal training: (i) \emph{Encoder-Bridged Retrieval}, which projects text-query states into the visual feature space of the frozen encoder, through a projection shared with the compressor, to retrieve visual tokens from a text query, with no contrastive training; and (ii) \emph{Query-Driven Distillation}, a lightweight cross-attention compressor that condenses the retrieved visual tokens into a small set of query-conditioned prefix tokens. On LongBench-v2, BAMBOO, and LooGLE-v2, OpticalRAG achieves the best compression-aware accuracy (CAP) on all three benchmarks while injecting only $16$ tokens, and improves the closed-book Qwen2.5-7B-Instruct backbone by $5.17$ points on LongBench-v2. Code is available at https://github.com/brighterr/OpticalRAG.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.09364

SafeOrchestrate: Runtime Checked Intent-Driven Orchestration for Cyber-Physical Systems

Muhammad Ashfaq, Ahmed R. Sadik, Ewan Warburton, Pyry Kotilainen, Niko Mäkitalo, Tommi Mikkonen

原文摘要

Modern cyber physical systems (CPS) integrate heterogeneous sensors, actuators, controllers, and services, composed at runtime rather than at design time. Orchestrating such systems through natural language (NL) is attractive, but unconstrained large language model (LLM) tool use is risky because actions must respect device capabilities, runtime state, and safety constraints. We present Safe Orchestrate, a reference architecture for runtime checked, intent driven CPS orchestration. Given an NL intent, it derives atomic goals, generates a runtime checkable specification for each, grounds the specifications to device affordances, and verifies every candidate action against runtime state and safety constraints before execution. In a Web of Things (WoT) instantiation, the architecture produces Linear Temporal Logic (LTL) style specifications from the NL intent, grounds them to Thing Description (TD) affordances exposed through the Model Context Protocol (MCP), and checks candidate actions against satisfiability modulo theories (SMT) backed guardrails. We evaluate this instantiation on simulated smart home, greenhouse, and industrial automation tasks against a direct LLM to MCP baseline. Safe Orchestrate maintains 100% required action accuracy on every task and 100% MCP Execution Similarity (MCP ES) on all but one, where MCP ES is 92%. By contrast, the baseline degrades on multi device, event triggered, and ordered tasks, with MCP ES falling to 13%. These results suggest that combining LLM based specification generation with role aware grounding and runtime constraint checking improves reliability for open CPS orchestration without enriched TD annotations or planning domain expertise.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15197

Anytime-Valid Empirical Likelihood Inference from LLM Annotations with Adaptive Human Auditing

Hongfu Gao, Jiang Hu, Pengfei Liu, Wang Zhou

原文摘要

Large language models (LLMs) are increasingly used to annotate unlabeled data as a fast, low-cost alternative to human annotation. However, their outputs can be biased and miscalibrated relative to human labels, potentially leading to misleading analysis and inference. We propose AAELI (Adaptive Human Auditing and Anytime-Valid Empirical Likelihood Inference), an online framework for estimating the population mean of human labels under a limited auditing budget. AAELI uses LLM outputs to guide adaptive human auditing and combines them with audited labels through predictable empirical-likelihood $e$-processes to construct anytime-valid confidence sequences. Our main theoretical result shows that a predictably learned efficient projection preserves the first-order profile empirical-likelihood statistic under adaptive auditing, without requiring a prescribed convergence rate for the projection estimator. This yields an exactly anytime-valid betting construction whose continuous-mixture form incurs a scalar rather than vector-dimensional mixing penalty. Supporting results establish adaptive empirical-likelihood asymptotics and show that a learned audit rule attains the oracle variance within the stated class of augmented estimating-equation designs under the same audit-probability floor and long-run expected budget. Experiments on synthetic data and three real-world datasets demonstrate that AAELI maintains empirical anytime coverage while requiring fewer human audits than the strongest competing baseline to reach a target precision.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15168

MAST: Medoid-Anchored Subspace Steering for Visual Grounding in Multimodal LLMs

Zhitao Chen, Zhenxin Cai, Jingyuan Li, Jin Sha

原文摘要

Existing activation steering methods for multimodal large language models (MLLMs) use the global activation mean as the steering vector, which tends to cause cross-task interference and unstable performance. We trace this failure to the multi-centered geometry of task activations. The global mean lands in the empty region between clusters and is faithful to no single task. Moreover, the steering signal in each task-pure local cluster is low-rank, and different query tasks benefit from emphasizing different families of local-basis directions under the shared routing protocol. To address this, we propose MAST (Medoid-Anchored Subspace Steering), a training-free, input-adaptive activation steering framework. MAST groups samples by task, performs multi-center clustering within each group, and selects one representative medoid per cluster. At inference time, each query is routed to multiple representative medoids. For each matched cluster, a low-rank local steering direction is extracted and aggregated across clusters to produce the final steering vector. Building on MAST, MAST-PR (Probe-Reweighted) uses a lightweight task-aware probe to reweight task-specialized local-direction families, yielding a further average improvement of +8.09 points over MAST.

cs.CV · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15154

SciSelect-Gym: An Environment and Benchmark for Active Learning and Bayesian Optimization Configuration Selection

Ying-Peng Tang, Tianle Zhang, Wanlong Fang, Ng Jing Xiang, Alvin Chan

原文摘要

In many scientific tasks, scientists repeatedly choose a small set of candidates to test, learn from the results, and use the observations to guide the next selection. Methods based on active learning~(AL) or Bayesian optimization~(BO) have been proposed for these iterative screening campaigns with a surrogate model. Yet, selecting the surrogate model and acquisition function for a new dataset remains challenging due to limited empirical evidence, while the effectiveness of large language models (LLMs) in making such decisions remains unclear. To address these gaps, we introduce SciSelect-Gym, an environment for simulating AL/BO data selection campaigns and evaluating configuration choices. We conducted simulations of AL/BO data selection campaigns across several settings and more than 100 molecular datasets. Using completed simulation campaigns as context improves LLM recommendation shortlists on held-out datasets more consistently than it improves the quality of their first choice. SciSelect-Gym provides a simulation platform and recommendation protocols for learning how to choose configurations before an acquisition campaign. An implemented agent use case connects these campaign records and simulation tools to LLM-generated configuration-selection policies. The code is provided as supplementary material.

cs.CE

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2610.26185HF 社区推荐

MoveBench: A Rubric-Based Benchmark for Evaluating Large Language Models in Sports and Health Coaching

Yitong Li, Jiang Bian, Jingling Wang, Pan Ding, Chengying Gu, Xing Huang, Hana Song, Dong Zhang 等 26 位作者

原文摘要

Despite rapid advances in large language models (LLMs), their deployment in the sports and health domain remains high-risk due to the absence of domain-specific evaluation benchmarks. Existing general-purpose benchmarks (e.g., MMLU and HELM) primarily assess broad knowledge and reasoning capabilities, but fail to capture critical domain requirements such as exercise safety, physiological appropriateness, personalization, and actionable coaching quality. This gap is particularly concerning in fitness and health applications, where incorrect or unsafe recommendations can lead to physical harm. To address this limitation, we introduce MoveBench, the first comprehensive benchmarking framework specifically designed for evaluating LLMs in sports and health coaching. The proposed benchmark integrates heterogeneous task formats, including multiple-choice questions (MCQs) and open-ended question answering (OpenQA), grounded in real-world user scenarios and paired with structured rubrics emphasizing safety compliance, scientific validity, and training logic. Through large-scale evaluation of 15 diverse LLMs on 809 questions with 5,012 rubric-level criteria, our results reveal substantial performance variability, particularly in safety awareness, structured program design, and personalization capability, highlighting a critical gap between knowing and doing. Our work establishes a foundational evaluation standard for sports and health LLM applications and provides a reproducible framework for advancing safe, reliable, and context-aware AI-driven coaching systems. Open resources are available at https://github.com/SatoSakula/MoveBench.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.17002

Governing Copyright Responsibly in the Era of Generative AI

D. Kochetkov

原文摘要

Large language models are trained on collections that may include scholarly works, while authors generally lack work-level information about whether and how their works were used. This paper asks how copyright governance frames and operationalises the protection of such authors, and what a responsible alternative would require. Forty legislative, judicial and soft-law instruments issued between 2019 and 2026 were coded deductively against a framework derived from Responsible Research and Innovation theory, yielding 1,337 category assignments across 835 segments. The European Union and the United States are the primary cases; the United Kingdom, China and international bodies serve as comparators. Normative orientation and the distribution of responsibility dominate the corpus, while reflexivity is the least represented category and reaches its minimum in the US sub-corpus. Both primary jurisdictions frame protection procedurally rather than as an entitlement, and no instrument in the corpus creates a general author-facing attribution duty for the use of works in AI training. From the coded gaps, the paper derives a six-instrument governance design and a capacity-based allocation of responsibility among developers, publishers, collecting societies, funders, institutions, regulators, and individual authors.

cs.CY · cs.AI · cs.DL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2506.11148

LLM-to-Phy3D: Physically Conform Online 3D Object Generation with LLMs

Melvin Wong, Yueming Lyu, Thiago Rios, Stefan Menzel, Yew-Soon Ong

原文摘要

The emergence of generative artificial intelligence (GenAI) and large language models (LLMs) has revolutionized the landscape of digital content creation in different modalities. However, its potential use in Physical AI for engineering design, where the production of physically viable artifacts is paramount, remains vastly underexplored. The absence of physical knowledge in existing LLM-to-3D models often results in outputs detached from real-world physical constraints. To address this gap, we introduce LLM-to-Phy3D, a physically conform online 3D object generation that enables existing LLM-to-3D models to produce physically conforming 3D objects on the fly. LLM-to-Phy3D introduces a novel online black-box refinement loop that empowers large language models (LLMs) through synergistic visual and physics-based evaluations. By delivering directional feedback in an iterative refinement process, LLM-to-Phy3D actively drives the discovery of prompts that yield 3D artifacts with enhanced physical performance and greater geometric novelty relative to reference objects, marking a substantial contribution to AI-driven generative design. Systematic evaluations of LLM-to-Phy3D, supported by ablation studies in vehicle design optimization, reveal various LLM improvements gained by 4.5% to 106.7% in producing physically conform target domain 3D designs over conventional LLM-to-3D models. The encouraging results suggest the potential general use of LLM-to-Phy3D in Physical AI for scientific and engineering applications.

cs.CV · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2608.18504

UMER: Unifying Embedding and Ranking via Pair-Aware Discriminative Reasoning for Universal Multimodal Retrieval

Libiao Chen, Xiyang Liu, Yanheng Wei, Zhenyu Tang

原文摘要

Modern Web information access increasingly involves heterogeneous multimodal content, including images, videos and visually rich documents. Users may seek not only globally similar content but also candidates that satisfy fine-grained attributes, relations or answer evidence under complex instructions. Supporting such access requires both efficient corpus-scale retrieval and fine-grained relevance reasoning. Multimodal large language model (MLLM)-based embedding methods derive representations from MLLM hidden states, and recent work further uses Chain-of-Thought (CoT) reasoning to improve embedding capability by injecting intermediate semantic evidence into the representation. However, existing CoT methods typically use item-wise reasoning over queries and candidates in isolation, providing no explicit evidence to distinguish a positive from a semantically confusable hard negative. Moreover, contrastive embeddings capture global similarity but struggle with meta-tasks requiring answer verification, category judgment or fine-grained reasoning. In this paper, we propose UMER, a Unified Multimodal Embedding and Ranking framework for universal multimodal retrieval. UMER replaces item-wise reflection with PairAware Discriminative Reasoning, which compares query-candidate pairs to identify instruction-relevant matching and discrepancy evidence. UMER jointly learns contrastive embeddings for efficient global matching and discriminative ranking for explicit pairwise relevance judgment within a single MLLM. A complementary mutual distillation strategy further transfers reliable pairwise preferences between the embedding and ranking functions. On MMEB-V2, UMER-R scores 65.5/70.6 with Qwen2-VL-2B/7B, outperforming the strongest included matched-backbone baselines by 1.4/2.0 points. UMER also supports budget-adjustable inference from efficient embedding retrieval to optional pair-aware reranking.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15069

DiFiT: Distribution-Fidelity-guided Visual Token Pruning for Multimodal Large Language Models

Jingyuan Zhang, Nihao Niu, Xinyang Lv, Beibei Han, Xiaoyun Wang, Yanning Hou, Ke Xu

原文摘要

Visual token pruning reduces the inference cost of multimodal large language models, but answer accuracy only partly reflects its effect. Pruning can change the probabilities of alternative response tokens even when greedy answers remain correct and unchanged. We compare full and pruned response distributions under shared reference prefixes. Average predictive drift grows with compression on fixed correctly answered samples. Separate analyses associate larger drift with more confidence-decision disagreements and sampling failures. We propose DiFiT (Distribution-Fidelity-guided Token pruning) to learn discrete selection from reference answers and full-token response distributions. Dual-View Distillation matches probabilities across the vocabulary, while answer supervision supports task quality. A learned scorer and differentiable Top-$K$ learn which visual evidence to retain. A multi-budget mask objective supports different retention ratios through one shared ranking over a frozen model. On Qwen2.5-VL-7B, DiFiT reduces sampling-induced negative flips by 30.3% relative to FastV at 10% retention on four-task greedy-correct controls. It preserves 98.74% of baseline performance on MME at 30% retention.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15058

FIBER: Bayesian Fusion of Fast-DetectGPT and IRT-Informed Detector Evidence for LLM-Generated Text Detection

Yan Li, Guangsheng Bao, Zhipeng Chen, Guandong Zhang, Yanbin Zhao

原文摘要

Detecting large language model (LLM)-generated text remains challenging under domain and generator shifts. Supervised detectors may generalize poorly, while zero-shot methods often rely on individual statistical signals without leveraging historical detector behavior. We propose FIBER (Fast-DetectGPT with IRT-informed Bayesian Evidence Recalibration), which integrates continuous Fast-DetectGPT scores with auxiliary detector outputs through Bayesian evidence fusion. Item response theory (IRT) models historical detector-correctness patterns, providing class-conditional detector-output likelihoods through marginalization over latent text representations. These likelihoods are combined with Fast-DetectGPT score densities and a historical class prior for prediction. When historical Fast-DetectGPT scores are unavailable, a variant called FIBER-Stream estimates class-specific score distributions from an unlabeled batch. On a 32,000-text multi-domain benchmark, FIBER achieves the highest accuracy (0.620) and improves mean AUC over Fast-DetectGPT (0.664 versus 0.600) under five-fold cross-validation. FIBER also consistently improves AUC across four individual domains. Under out-of-distribution transfer, FIBER-Stream achieved an AUC of 0.911, compared with 0.889 for FIBER and 0.922 for Fast-DetectGPT alone. These results demonstrate the potential of integrating complementary detector evidence and historical information to improve detection performance across heterogeneous settings.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.10555

YRC-Bench: Benchmarking Culturally Aware Machine Translation for Historical Heritage via Tri-Level Semantic Bridging

Yan Huang, Aobo Li, Jiayi Wang, Huidong Zhu

原文摘要

Translating historical cultural heritage poses profound challenges for machine translation systems due to the acute semantic asymmetry across eras and cultures, where terse classical syntax frequently obscures domain-specific cultural knowledge. While Large Language Models have demonstrated remarkable prowess in generic translation benchmarks, they struggle to negotiate Culturally-Specific Items and diachronic shifts in domain-rich historical texts, such as Yellow River cultural records. To bridge this critical gap, we introduce YRC-Bench, the first human-curated, multi-parallel benchmark dedicated to historical heritage translation across Classical Chinese, Modern Chinese, and nine target foreign languages. YRC-Bench establishes a tri-level parallel corpus incorporating modern explanatory glosses as explicit semantic bridges and operationalizes translation strategies through paired conserved versus substituted CSI variants categorized by cultural translatability. Systematic evaluations of nine leading LLMs across bidirectional translation directions reveal that leveraging modern Chinese glosses - especially human-curated ones - dramatically elevates both lexical fidelity and semantic adequacy. Furthermore, multidimensional human preference audits by native experts unveil distinct cross-cultural preferences between Sinographic-tradition and non-Sinographic language groups.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.11975

AutoQS: Automated Quantization Strategy Design with Large Language Models

Zhen Liu, Zhiqin Yang, Yonggang Zhang, Jianyi Liu, Jinjun Wang, Jingwen Fu

原文摘要

Quantization reduces LLM deployment costs, but effective strategies must balance model quality and storage. LLM-driven search can automate strategy design through repeated proposal and evaluation. This process must identify useful strategies within a limited evaluation budget. The key challenge is that \textit{actual quantization is expensive, while inexpensive proxies may not reflect quantized-model performance.} To address this challenge, we present AutoQS, an \textbf{Auto}mated \textbf{Q}uantization \textbf{S}trategy Design with LLMs. AutoQS keeps weight precision fixed and searches group-size allocations across layers and operators under deployment constraints. Evaluation history guides coordinated LLM proposals, while deterministic validation ensures compatible allocations. An inexpensive quality proxy evaluates candidates, and an candidate repository retains alternative allocations to inform subsequent proposals. Periodic actual quantization complements these proxy evaluations with measured quality and checkpoint storage, providing feedback for further search within the evaluation budget. Experiments on Qwen2.5-7B-Instruct show that AutoQS improves seven-task average accuracy over uniform-group GPTQ and AWQ by 0.70 and 1.12 percentage points, respectively, at four-bit weight precision.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2511.23115

Robust Image Emotion Classification via Multi-Level Affective Captioning

Zibo Zhou, Zhengjun Zhai, Huimin Chen

原文摘要

Image emotion classification (IEC) has been studied for many years and has recently attracted increasing attention with the rapid development of deep learning. Although recent studies have benefited from knowledge embedded in pre-trained visual models, the performance is still limited by the "affective gap" between low-level visual features and high-level affective semantics. Psychological studies suggest that language can convey fine-grained and emotion-rich information. Inspired by this observation, we attempt to narrow the "affective gap" by introducing affective text. Specifically, we design a multi-level contrastive loss to detect emotional concepts from images and introduce an emotional-attribute-guided large language model approach to generate affective sentences. Then, a pre-trained language model is employed to combine emotional concepts with affective sentences for IEC. In addition, we introduce a contrastive loss with a semantic-similarity-based sampling strategy to address large intra-class differences and limited inter-class differences in affective datasets. Moreover, unlike many previous studies, we explicitly consider images containing embedded text. Extensive experiments on multiple benchmarks show that the proposed method helps narrow the "affective gap" and achieves competitive performance.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15029

Mask-Pattern Position Augmentation: Training-Free Long-Context Extrapolation for Diffusion LLMs

Gaurav Sarkar, Syed Affan Daimi

原文摘要

Diffusion large language models (dLLMs) generate by iteratively unmasking tokens with bidirectional attention, but rotary position embeddings (RoPE) still confine them to a fixed pretraining window. Existing training-free extrapolation, notably LongLLaDA, copies NTK-based RoPE scaling from autoregressive (AR) models and applies one static, length-only transform at every position and denoising step. That transplant ignores a signal unique to dLLMs: the evolving mask pattern of the partially decoded sequence. We propose Mask-Pattern Position Augmentation (MPPA). MPPA linearly interpolates indices into the pretrained window, then adds a per-position, per-step offset from cumulative mask density. The offset restores rotary phase resolution that interpolation collapses, and it shrinks as tokens are committed, so late steps see near-clean interpolated geometry. On LLaDA-8B-Instruct at twice its 4096-token window, MPPA reduces perplexity from 19.51 under standard decoding to 3.12 at 8192 tokens, below the in-window figure, with 100% passkey retrieval and unchanged needle-in-a-haystack accuracy. These preliminary $2\times$ results motivate mask-conditioned position engineering as a diffusion-native path to long-context generation.

cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2606.09547

Streaming Interventions: Can Video Large Language Models Correct Mistakes as They Occur?

Apratim Bhattacharyya, Shweta Mahajan, Sanjay Haresh, Rajeev Yasarla, Reza Pourreza, Litian Liu, Risheek Garrepalli, Roland Memisevic

原文摘要

Learning everyday skills, like cooking a dish, relies increasingly on instructional media such as online videos. This opens the door to the use of video (and multimodal) large language models (LLMs) as task guidance assistants. A crucial capability for the real-world success of a prospective task guidance assistant is it's ability to intervene proactively as soon as a mistake is apparent in order to guide the user. To evaluate this crucial capability, we introduce Ego-MC-Bench (Mistake Corrections), a benchmark for evaluating reactive, step-by-step task guidance in realistic cooking scenarios. Extensive experiments show that Ego-MC-Bench is highly challenging for state-of-the-art video LLMs. We argue that a key reason is the limited availability of training data for fine-tuning models on this task. Although there exists a wide range of cooking video datasets, existing datasets lack examples of mistakes along with appropriately timed interventions. To help address this data limitation, we also introduce Ego-CoMist, a counterfactual synthetic dataset created by transforming non -interactive cooking videos into supervised training examples showing proactive interventions. We show that fine-tuning on Ego-CoMist yields performance gains especially for smaller and more efficient video LLMs that are well suited for delivering assistance on edge devices.

cs.CV · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15013

Protecting Sensitive Knowledge Under LLM Unlearning: The Proxy Threat and a Superposition-Aware Defense

Jiayang Li, Tianqing Zhu, Longxiang Gao, Huajie Chen, Dayong Ye, Shang Wang, Huan Huo, Wanlei Zhou

原文摘要

Machine unlearning lets the owner of large language models (LLMs) remove specific entities from a trained model without retraining from scratch. Existing machine unlearning methods focus on removing the requested entities while maintaining utility. However, we observe that the representations of distinct entities in an LLM may overlap in shared parameter subspaces, a phenomenon known as knowledge superposition. Removing one entity may then affect other entities. This exposes a new security threat: legitimate unlearning requests can be exploited to remove sensitive entities that were never requested. This threat is unlike data poisoning or backdoor attacks. We define the attack as the Proxy Unlearning Attack (PUA), which selects removable proxy entities superposed with a sensitive target. The attack can even occur unintentionally, when benign users accidentally remove entities superposed with a sensitive one. In response, we propose Superposition-Aware Unlearning Protection (SAUP), a general defense framework that can be applied to unlearning requests to ensure a correct removal. SAUP performs in two stages. Before unlearning, it uses superposition detection to estimate which sensitive entities the request puts at risk. After unlearning, it decouples the shared representations and restores the affected entities, while keeping the requested entities suppressed. Our findings reveal that legitimate unlearning requests can be maliciously exploited through knowledge superposition. When dealing with unlearning requests, LLM providers need representation-level detection and recovery mechanisms to mitigate this threat and protect sensitive knowledge.

cs.CR

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2506.15253

Specification-based Security Evaluation of Real-World Agents Tool-Calling under Long-Horizon Context Tasks

Yuchuan Fu, Minghe Wang, Xiaohan Yuan, Chenlin Zou, Kaitong Jia, Jinfeng Li, Dongxia Wang

原文摘要

As large language model (LLM) agents are increasingly deployed in long-horizon unattended scenarios, security evaluation of them before deployment becomes increasingly essential and challenging. Existing evaluation methods mainly follow the paradigms of LLM-as-a-Judge or Agent-as-a-Judge. However, such judges are themselves often vulnerable to hallucination and can even be affected by perturbations. Furthermore, because capturing full environment states at every step is often prohibitively expensive in open-ended environments, existing evaluations typically rely on textual logs and the final environment state. As a result, they may miss transient violations that are overwritten before the trajectory ends or never reflected in textual logs. We present RAS-Eval, a specification-based security evaluation framework for LLM agents. RAS-Eval uniformly expresses utility and security objectives as Linear Temporal Logic (LTL) specifications over tool-call trajectories and dynamic environment observations, and verifies them with an off-the-shelf model checker. This design addresses both limitations above. First, unlike LLM-as-a-Judge and Agent-as-a-Judge, RAS-Eval derives deterministic verdicts from formally defined properties, eliminating evaluator-side uncertainty and remaining reliable even for long-horizon trajectories. Second, RAS-Eval observes exactly the states on which the verdict depends, making it affordable to monitor the environment throughout execution and detect transient violations that leave little trace in the logs or final environment state. Moreover, we design a multi-agent-based dataset construction pipeline. Using it, we construct a benchmark (https://github.com/lanzer-tree/RAS-Eval) of 200 normal tasks and 800 attack tasks spanning four attack surfaces: indirect prompt injection, tool poisoning, skill code injection, and tool code injection.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.15006

EvoX: Rethinking the Agent Harness as a Runtime Policy

Haoyang Zhang, Qumeng Sun, Xiao Zhang, Xiang Liu, Zhidong Fan, Junjie Wang

原文摘要

Large language model (LLM) agents increasingly interact with tools and environments, yet plausible model decisions do not always translate into successful execution due to tool-call mismatches, interrupted workflows, and unsupported completion claims. We present EvoX, a Rust-based, multi-model agent runtime guided by The Second Policy, which views the harness as an Operational Policy complementing the foundation model's Semantic Policy. EvoX combines model-facing guidance with deterministic runtime control through three complementary mechanisms: Action Realization, Execution Continuity, and Outcome Grounding. To evaluate harnesses without suppressing their native behaviors, we introduce Contrast, a controlled and auditable framework integrating explicit fairness constraints, provider-bound request verification, and evidence-based adjudication. Across five benchmarks, six foundation models, and six agent harnesses, EvoX achieves the highest numerical success rate of 56.2% over 691 common task--model pairs, exceeding the average of the other five harnesses by approximately 3.0 percentage points. Its larger advantages on GLM-5.3 (+10.7 points) and Kimi K3 (+7.9 points) highlight the model-dependent effectiveness of harness design and the potential for cross-provider adaptability. Achieved without memory, evolution, or extensions, these results establish EvoX as a competitive multi-model agent runtime and underscore the importance of harness-level operational control in realizing foundation-model capabilities.

cs.AI · cs.SE

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14997

Same Task, Different Statement: When and How Reformulation Helps LLM Problem Solving

Wentao Zhang, Xiaoyu Zhang, Chengwei Liu, Yutao Hu, Tao Zhang, Junsheng Wu, Hongping Gan, Yang Liu

原文摘要

When people face a hard problem, they often restate it in a clearer way before solving it. We study whether the same habit helps large language models by rewriting the statement of a problem. A legal rewrite must keep the meaning, the constraints, and the correct answer unchanged. We test three kinds of rewrites. Decomposition makes hidden subgoals explicit. Typed Reformulation casts the task into a more suitable solving frame. Concept Reduction removes surface detail and exposes the task structure. We evaluate them on eight math, logic, and code benchmarks with eight models, all under the same chain-of-thought solving prompt. The effect is large but depends on the problem type. An oracle that picks the best rewrite after the fact raises accuracy by +27.6% on average. Yet no single rewrite helps every problem, and a mismatched rewrite can hurt. Still, the effect follows clear patterns. Each kind of rewrite fits certain problem types. A simple fixed rule learned from one set of problems transfers part of the gain to new problems. It gains +11.0% on the problems it chooses to rewrite, while blind rewriting loses 5.6% on a similar share of problems. Rewriting is therefore not a universal way to solve problems. The question is when and how to rewrite.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v5arXiv:2605.12374

Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models

Yanting Miao, Yutao Sun, Dexin Wang, Mengyu Zhou, Pascal Poupart, Lei Lv, Qi Zhao, Li Wang 等 11 位作者

原文摘要

Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens, avoiding external tools or image generators. However, existing methods usually follow an output-as-input latent paradigm and yield unstable gains. We identify a feature-space mismatch behind this instability: dominant visual-latent models build on pre-norm MLLMs and reuse decoder hidden states as predicted latent inputs, even though these states occupy a substantially different norm regime from the input embeddings the model was trained to consume. This mismatch makes direct latent feedback unreliable. Motivated by this diagnosis, we propose GAP, a Granular Alignment Paradigm for visual latent modeling. GAP aligns visual latent reasoning at three levels: data-level alignment grounds latent targets with inspectable auxiliary visual supervision; feature-level alignment maps decoder outputs into input-compatible visual latents through a lightweight PCA-aligned latent head; and model-level alignment assigns latent supervision selectively to examples where the base MLLM struggles. On Qwen2.5-VL 7B, GAP achieves the best aggregate perception and reasoning scores among all compared methods and is the only evaluated latent method that improves both over the base model. Inference-time interventions further show that generated latents carry task-relevant visual signal rather than merely adding token slots. Code is available at https://github.com/Qwen-Applications/GAP.

cs.CV · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14965

Early Signals, Strong Decisions: Prefix-Guided Sampling for Parallel Test-Time Scaling

Jie Ren, Jonathan Rosenfeld, Neil Thompson

原文摘要

Parallel test-time scaling (TTS) has been shown to enhance reasoning in large language models by generating and aggregating multiple independent reasoning paths. Internal confidence scores can further improve this process by identifying paths of varying quality. However, existing methods remain computationally inefficient, as low-quality paths are fully expanded before being evaluated. In this work, we aim to improve the use of internal confidence signals for better model performance and computational efficiency. We begin by offering a new perspective on the role of confidence in aggregation, showing that it is most impactful for questions near the decision boundary. Notably, when amplified by multi-sample aggregation, even small confidence signals derived from short prefixes can meaningfully influence outcomes in these boundary cases. Building on this insight, we introduce Prefix-Guided Sampling (PreG), a method that reallocates computation by first generating short prefixes, ranking them using internal confidence scores, and then completing only the most promising candidates. This strategy reduces token usage while maintaining gains in path quality. We provide theoretical analysis demonstrating that, under a fixed compute budget, PreG strengthens the effective signal by widening the decision boundary margin. Empirically, our method consistently outperforms standard parallel TTS approaches across benchmarks, while significantly improving token efficiency and reducing overall computation.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14949

RSI-PDEbench: A Benchmark for Self-Improving Agents in PDE Solving

Binghang Lu, Lang Cao, Changhong Mou, Yige Sun, Cindy Xiangrui Kong, Zheyuan Deng, Tian Yuan, Bing Hu 等 11 位作者

原文摘要

Identifying and solving partial differential equations (PDEs) from sampled solution data often requires substantial expert effort in equation discovery, coefficient estimation, and numerical solver configuration, especially under noise and distribution shifts. Can large language model (LLM) agents automate this workflow accurately and efficiently? To investigate this question, we introduce \bench{}, a benchmark of 1,001 cases spanning 42 equation templates, including both classical and fractional PDEs. It provides three evaluation settings: Full evaluates performance on noise-free solution data; Noisy tests robustness to Gaussian noise with an RMS equal to 50\% of the clean-signal RMS; and Hard-OOD tests challenging out-of-distribution generalization to unseen equation templates and shifted coefficient ranges. We further introduce RSI-PDE, a recursively self-improving agent harness that shortlists candidate equations by residual loss, uses relative $L^2$ feedback to refine coefficients and numerical solvers, and applies a validation gate to policy and memory updates. We compare RSI-PDE with single-run baselines and native agent harnesses (Claude Code or Codex) across four frontier models. Across the three settings, RSI-PDE improves the pass rate averaged across the four models by 106.5--181.9\% relative to single-run baselines and by 3.0--19.6\% relative to native harnesses, while reducing total recorded model calls by 46.9\% across the 12 model--setting pairs. These results highlight the potential of validated recursive self-improvement in LLM agents to advance scientific computing.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14946

LOHRE: Execution-Grounded Type Recovery via LLM-Orchestrated Hybrid Analysis

Hao Jiang, Yujie Yang, Tao Li, Chenhui Cui, Liangfeng Xu, Chengjie Liu, Weiping Wen, Rubing Huang 等 9 位作者

原文摘要

Documenting stripped binary functions for security analysis requires recovering function interfaces and user-defined type (UDT) layouts from machine code. Existing large language model (LLM)-based methods face two limitations: they lack execution-based validation to detect inferred UDTs inconsistent with actual memory layouts, and may overlook evidence about shared UDTs across functions without direct call relationships. We present LLM-Orchestrated Hybrid Reverse Engineering (LOHRE), a framework that treats inferred UDTs as falsifiable hypotheses and tests them against program execution. LOHRE uses a hypothesize-validate-refine loop to test candidate layouts through paired dynamic slice emulation of baseline and mutated inputs. To reduce the cost of processing lengthy execution histories, the Execution-Depth Heuristic (EDH) compresses paired traces into an execution-depth difference and a summary of the tested field's observed uses. A cross-function knowledge base aggregates compatible UDT hypotheses and field-level confidence scores to guide testing and hypothesis refinement. We evaluate LOHRE on RealType at four optimization levels and a Windows remote procedure call (RPC) dataset. LOHRE achieves an average UDT Structural accuracy of 57.85% across RealType's four optimization levels and 35.6% on Windows RPC, outperforming the evaluated static baselines. Disabling dynamic validation reduces accuracy to 19.28% on RealType and 9.2% on Windows RPC. Removing the cross-function knowledge base reduces accuracy to 41.03% on RealType and 10.0% on Windows RPC. Replacing EDH's compressed feedback with full execution traces reduces recovery accuracy while increasing token use by 3.7 times and runtime by 3.0 times. These results indicate that execution-based validation, compact feedback, and cross-function evidence contribute to accurate and efficient UDT recovery.

cs.SE

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2602.08294

Not All Gap Correction Helps: A Geometric View of External Information in LLM Inference

Dingzirui Wang, Xuanliang Zhang, Keyan Xu, Qingfu Zhu, Wanxiang Che, Yang Deng

原文摘要

Providing external information beyond the query is a common way to improve large language model (LLM) inference, as seen in in-context learning (ICL), retrieval-augmented generation (RAG), and memory-based methods (Mem). Existing work increasingly suggests that useful external information should compensate for the gap between LLMs and user queries, implicitly assuming that stronger gap-correction can lead to better performance. In this paper, we challenge this assumption and show that more gap-corrective external information can instead harm performance. We analyze this phenomenon in Transformers through the lens of reasoning error, defined as the difference between the predicted answer vector and the ground-truth answer vector. Our analysis reveals that external information can be not only under-corrective but also over-corrective, where excessively strong correction increases reasoning error. We further show that the induced correction vector is jointly determined by attention and the complementarity between the external information and the query, and derive conditions under which information is properly corrective in both direction and magnitude. Experiments on four mainstream LLMs and seven reasoning benchmarks across ICL, RAG, and Mem validate our theory and support a theory-guided information selection method.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v6arXiv:2610.03228

Reference-Compatible Parallel Execution for Quantized LLM Inference

Chengzhan Li

原文摘要

Batching a quantized large language model (LLM) can change full-vocabulary outputs and the key-value (KV) cache even when token positions and causal masks are unchanged. We introduce reference-derived phase plans: per-call execution descriptions that preserve a fixed deployment's producer, physical attention range, arithmetic and state ownership while exposing parallel work. Whole-graph plans split the complete computation at compatibility boundaries; local-attention plans split only attention and retain wider projection and feed-forward work. Independent arithmetic groups can also be packed into one operation without changing each group's reduction. We implement and qualify these paths on CUDA and Vulkan for Qwen, Llama and SmolLM2. On Linux/RTX A5000, compatible speculative execution reduces resident-service time by 17.21 percent, 25.54 percent and 11.52 percent relative to the fastest measured fixed target-only control for Qwen3-8B, Qwen3-14B and Llama-8B. Balanced experiments isolate an increment of 0.80 to 1.64 percent from Q8 projection packing and smaller, geometry-dependent gains from local cuts. For Qwen3-32B with mixed Q4_K/Q6_K weights, independent-channel grouping reduces ordinary service time by 5.61 percent against a graph-level scalar-column loop. Equal-work tests favor eight-query execution in aggregate, whereas complete live requests favor shorter proposals because longer proposals increase target work. The results show how to exploit known-query parallelism within an established numerical reference and how complete-request costs determine the useful execution strategy.

cs.PF

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14935

Do Large Language Models Show Racial and Sex Bias in Pain Management? A Counterfactual Audit of GPT, Claude, Gemini, Llama, DeepSeek and Qwen

Richard J. Young, Alice M. Matthews

原文摘要

Racial and ethnic minority patients receive less pain treatment than non-Hispanic White patients, and large language models (LLMs) used as medical AI could reproduce that bias. Prior audits tested a few proprietary models, stated race explicitly and rarely measured the models' own response instability. In exploratory, non-preregistered analyses, Study 1 had four open-weight LLMs assess six acute-pain vignettes in every combination of four races and two sexes, with identity stated or cued only by a name (two names per group), comparing each identity with a White male reference on the same case and sampling seed. In Study 2, 24 configurations of 17 models, including OpenAI's GPT-4.1 mini and gpt-oss, Anthropic's Claude Haiku 4.5, Google's Gemini 2.5 Flash, Meta's Llama, DeepSeek and Qwen, with reasoning on and off and at two precisions, answered the same vignettes and the 50 clinician-written Q-Pain cases. In this single-prompt, JSON-constrained task, no difference survived correction in Study 1, none of 1,386 comparisons survived the false discovery rate in Study 2, and every pain-severity difference was equivalent to zero within one point on a 0-10 scale (within a quarter point for 237 of 238 Q-Pain comparisons). Opioid decisions were bounded less precisely (Study 1 could not exclude differences of about 20 percentage points), so they show no detected difference, not proof of none. The few differences surviving correction within a model favoured Black patients. Models differed far more from one another and across their own settings than between patients (the reference patient received the high opioid dose in 6% to 93% of responses), although serving software, provider and prompt format also differed. Fairness audits of clinical AI should use several names per group, pair comparisons on the seed, report the noise floor and test each deployed configuration.

cs.CL · cs.AI · cs.CY

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14920

Janus: One Agent, Two Roles for Digital Twin Construction and Network Orchestration

Peijun Hou, Yue Zhang, Yi Wei, Nan Cen

原文摘要

Next-generation wireless networks are increasingly large-scale, heterogeneous, and dynamic, posing significant challenges to network modeling and control. Digital twins offer a promising solution, but existing approaches typically rely on static models and decouple twin construction from network orchestration, limiting adaptability. Meanwhile, large language models (LLMs) provide strong reasoning capabilities but struggle with graph-structured data and are highly sensitive to prompt design, making inference unreliable. In this paper, we propose Janus, an agentic LLM-based framework that unifies digital twin construction and network orchestration within a closed-loop system. At the core of Janus is a novel NetTwin agent, which introduces a new one-agent, two-role paradigm, enabling a single agent to simultaneously construct the digital twin from partial observations while performing task-driven network decision-making. To realize this, we design a modular architecture with three functional roles, including planner, modeler, and coder, along with a newly designed unified prompt module that further enhances graph reasoning robustness. Extensive evaluations on wireless networks with up to 200 nodes and two representative tasks, including user association and SAGIN routing, show that Janus achieves over 90% digital twin accuracy and near-optimal performance with gaps within 10.53%, while maintaining moderate computational overhead. These results demonstrate the effectiveness, scalability, and generalizability of Janus, paving the way for AI-native autonomous wireless network orchestration.

cs.NI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.14912

Rethinking "RL Generalizes, SFT Memorizes": The Role of SFT Data

Yunlong Hou, Fengzhuo Zhang, Yuan Cheng, Jiachun Pan, Xingyao Li, Zhuoran Yang

原文摘要

Large language models trained with Reinforcement Learning (RL) with verifiable rewards exhibit strong reasoning ability and broad generalization, whereas models trained with Supervised Fine-Tuning (SFT) are often viewed as more prone to memorization and limited transfer. This paper rethinks this distinction through the lens of SFT training data on widely used Qwen2.5 and Qwen3 series models. First, we study the role of data source and show that it is critical: a carefully mixed SFT dataset substantially outperforms data generated solely by a larger model. Second, we study the role of data scale and show that matching the number of correct rollouts between SFT and RL greatly improves SFT generalization, while matching the total rollout budget enables SFT to generalize as well as RL. Combining these two factors further enables SFT to generalize even better than RL. Third, by using LLM annotations to characterize the solution methods in training rollouts, we show that larger datasets cover more tail methods and that these tail methods provide generalizable reasoning signals. Finally, we support these empirical findings theoretically by analyzing the training dynamics of shallow transformers under both RL and SFT.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗