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

+816新增论文
+642其中 AI 相关
+95已有论文修订
+95新增 AI 导读

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

统计口径与数据新鲜度

统计 2026-11-20 14:45 — 2026-11-21 14:45(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-21 12:19最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

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

29,158 已收录论文24,181 AI 相关8,992 有修订112 HF 社区发现

浏览论文库

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

找到 24,181 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-21 12:17(北京时间)
新论文 · v1arXiv:2611.24077

AeroRSI: Recursive Self-Improvement for Aerial Vision-Language Navigation

Shiwei Wu, Yuwen Pan, Zhi Li, Yaowen Li, Yan Zhou, Xuemiao Xu, Yu Liu

原文摘要

Vision-language-action (VLA) models enable unmanned aerial vehicles (UAVs) to follow natural-language instructions, yet still fail frequently in closed-loop flight. Correcting these failures requires repairs, i.e., actions that avoid them at the states the policy itself reaches, which human pilots cannot label at scale and rule-based planners cannot supply reliably. In this paper, we present AeroRSI, a recursive self-improvement framework in which an aerial navigation policy iteratively learns from its own failures. Our key insight is that a failing policy should judge its own repairs by flying on from them. Judged in this way, its sampled actions and even rule-based proposals that hurt it when imitated directly become reliable supervision. Moreover, an episode terminates at its first failure, concealing subsequent failures until the earlier ones are repaired. AeroRSI thus renews its supervision in every round, anchoring the failures of the current policy, verifying repairs by its continuation, and distilling them while keeping its behavior elsewhere. On TravelUAV, three rounds raise the success rate of a supervised 2B-parameter policy from 46.4% to 67.5% and reduce collisions by half, surpassing the previous state of the art by 10.6 points from the same initialization without extra human labels or inference cost. Code and models are available at https://github.com/starsholic/AeroRSI.

cs.CV · cs.AI · cs.RO

新论文 · v1arXiv:2611.24076

Meaning Reaches Object Choice but Not Completion: Where Is a VLA Vulnerable to Landmark Synonyms?

Ryoichi Kurihara, Kota Sunamoto

原文摘要

VLA policies are fragile to rewording. Success rates alone cannot say whether a policy fails to choose the object or to complete the task, nor where in the policy to intervene. We study the released GR00T N1.7 on two LIBERO-Spatial tasks differing only in the landmark word telling which of two bowls to take. Synonyms absent from the training instructions send the policy to the bowl they name, unlike made-up words, yet complete 67.6% of one task's closed-loop episodes, against the original word's 100%. Before the connector between the vision-language model and the action generator, replacing only the landmark tokens' vectors with the other landmark's (pasting) sends nearly every episode to the other bowl. After the connector, pasting at the landmark, the remaining instruction, or the image tokens alone rarely moves the choice, but any two do. Hiding the landmark tokens from the action generator after the connector still lowers completion for the synonyms that fall short. Finetuning on diversified landmark words repairs that one task's completion for synonyms absent from finetuning; finetuning on the original instructions does not. Training only the connector on the diversified words suffices, as does copying the connector of a policy fully finetuned on them into one finetuned on the original instructions. An open-loop score on starting frames misses that weights outside the connector also carry part of the repair. For now, on this evidence, robustness to rewording should be judged in closed loop, with actions executed, intervening on both sides of the connector, scoring choice and completion separately.

cs.RO · cs.CL · cs.LG

新论文 · v1arXiv:2611.24075

Recurrent Domain Incremental Adaptation of Time-Series Foundation Models

Sachith Abeywickrama, Peiliang Gong, Emadeldeen Eldele, Min Wu, Chau Yuen

原文摘要

Pretrained Time-Series Foundation Models (TSFMs) provide strong representations for downstream classification tasks. Yet adapting a TSFM under changing operating conditions remains challenging. A model must improve on each newly observed condition without sacrificing conditions it has already learned, often from only a small labeled calibration set and without knowing the operating condition at inference time. We find that this difficulty has an important geometric structure. Under an operating-condition shift, the embedding space of a TSFM does not move uniformly: some local regions change substantially, while others remain nearly unchanged. Consequently, a global adaptation can unnecessarily move embeddings that were already correctly classified. To address this, we propose the novel PACT (Piecewise Adaptation via Chart Transport). PACT partitions the frozen source embedding space into local charts and learns lightweight residual transports for the regions affected by each new condition. An explicit identity path is always retained, and a learned gate softly selects among the identity and accumulated transports for each input. The pretrained TSFM encoder, classifier head, and previously learned transports remain fixed. With only 5% labeled calibration data per arriving condition and no condition identity at test time, PACT maintains an average macro-F1 above 0.95 across a four-condition bearing-fault stream. Under homogeneous tasks and label spaces, similar retention-adaptation behavior is observed across public benchmarks. Code is available at https://github.com/Sachithx/PACT.

cs.AI · cs.LG · eess.SP

版本更新 · v2arXiv:2608.04590

JUDOR: CTDE-Based Multi-Agent Reinforcement Learning for the Co-Optimization of Joint UAV Motion and Decentralized Opportunistic Routing in Delay-Tolerant Network

Xiao Wang, Shun-Ren Yang

原文摘要

The growing deployment of delay-tolerant networks (DTNs) has made store-carry-forward (SCF) communication indispensable for operating under sparse intermittent connectivity. However, intermittent contact opportunities, finite node buffers, and limited message time-to-live (TTL) degrade message delivery performance and induce network congestion. Yet unmanned aerial vehicles (UAVs), with extended radio coverage and controllable trajectories, can bridge ground-node pairs that would otherwise suffer poor connectivity. Nevertheless, conventional routing schemes decouple UAV motion control from message replication, and consequently fail to leverage UAV mobility to proactively create future contact opportunities. This paper presents JUDOR (Joint-UAV-Decentralized-Opportunistic-Routing), a centralized training with decentralized execution (CTDE)-based multi-agent reinforcement learning framework optimized for joint discrete UAV heading control and decentralized opportunistic routing. Specifically, we model the tightly coupled UAV motion-routing co-optimization task as a factored partially observable Markov decision process with a shared team reward. JUDOR adopts decentralized actors for local-node execution and a centralized critic used exclusively in training, with optional multi-horizon congestion-hotspot auxiliary supervision. Through evaluation across four distinct traffic modes, JUDOR yields consistent performance gains over both classical and learning-based DTN baselines, including PRoPHET and MaxProp, whilst preserving contact-limited decentralized execution suitable for real-world DTN deployments.

cs.AI

版本更新 · v2arXiv:2611.09154

Learning When to Think: Dual-Reference Offline Optimization for Adaptive VLM Reasoning

Hongbin Lin, Sizhe Zou, Juangui Xu, Xinyue Xu, Zhuoli Ouyang, Stanislav Kolchin, Chengwei Qin, Zhongxiang Dai 等 9 位作者

原文摘要

Reasoning traces improve vision-language models on complex multimodal tasks, but forcing every input to follow a long reasoning process leads to overthinking and unnecessary inference cost. We study adaptive fast/slow thinking for VLMs, where one model answers directly for simple inputs and reasons step by step when needed. Fast answers and slow traces often come from distinct behavior policies, making a single-reference offline objective misweight one mode. We propose DUAL-REFERENCE BOLT (DR-BOLT), which distinguishes the desired fast/slow mixture reference from the sampling distribution and corrects their mismatch in reward-weighted training. Fast/slow partition competition explains routing in the optimal target; an efficient source-wise correction approximates the exact objective when response formats are well separated. Matched Qwen3-VL comparisons use identical candidate banks to isolate the objective from slow-teacher quality. DR-BOLT improves score-length trade-offs over the evaluated offline and online baselines on ScienceQA and three held-out transfer benchmarks, with a 24.1% cross-benchmark slow ratio and a 2.90-fold end-to-end training speedup over reproduced GRPO. These results support adaptive reasoning in the studied two-reference setting, while broader robustness remains open. Our code is available at https://github.com/HongbinLin3589/DR-BOLT.

cs.AI

新论文 · v1arXiv:2611.24074

SemPhys: Semantic-Physical Co-Learning for Humanoid Motion Generation via Differentiable World Models

Zejian Kang, Yeqiang Wang, Zekai Chen, Boyang Shen, Mingyu Ma, Bingqian Wu, Fan Zhou, Jianjie Zhang 等 12 位作者

原文摘要

In human-robot collaboration, natural language lets people direct a humanoid's whole-body movements without programming joint trajectories. A common approach generates a kinematic reference from an instruction and relies on a tracking controller to execute it. However, as the controller maintains balance under contact constraints, the executed motion can depart from the reference in ways that change the action it conveys. Our central insight is that semantic alignment must be evaluated directly on the complete motion the robot physically executes. Building on this insight, we introduce \textbf{SemPhys}, a framework that trains motion generators using semantic feedback from simulated executions. A differentiable world model and a fixed tracking controller predict a short segment of a recorded execution. SemPhys inserts this prediction at its original position, scores the complete trajectory against the instruction, and backpropagates through the predicted segment. A paired anchoring loss stabilizes adaptation by keeping each generated reference close to its counterpart from the initial generator. On BONES-SEED, SemPhys largely preserves instruction retrieval accuracy from reference to execution and achieves the highest retrieval accuracy on executed motions among the evaluated methods.

cs.RO · cs.CV

新论文 · v1arXiv:2611.24073

3D GameDev Arena: Benchmarking Coding Agents That Plan, Build, and Playtest extreme Long-Horizon Browser Games

Bizhe Bai, Jiliang Jian, Yuqing Yan, Tao Chen, Ying Wen

原文摘要

Coding agents now resolve repository issues and generate compact playable games, but whether they can build a large 3D game from scratch remains open. Such games tightly couple mechanics, scenes, assets, and interaction logic, and their quality rests on player enjoyment, which existing benchmarks rarely assess. Thus, we introduce 3D GameDev Arena, a first extreme long-horizon benchmark of 108 from-scratch game-development tasks across 14 genres. Given only a text specification and UI/UX reference images, a fully automated plan--implement--test--repair pipeline builds each task into a self-contained browser game. Runs average $\sim$1,000 LLM calls ; the longest uses 14,407 calls, 2.13B tokens, and 608 assets. Evaluation spans (1) objective code-level functional acceptance against the game specification, (2) agent-based QA and playtesting of 3D presentation and playability, and (3) blind pairwise Bradley-Terry human preference in a public arena. The preliminary findings indicate that long-horizon 3D game development remains far from solved, and code-level correctness alone does not dictate player preference. The public arena is hosted online.

cs.AI

版本更新 · v2arXiv:2609.02448

Adapting a Foundation Model for Lunar Surface Height Estimation

Patrick Bauer, Marius Schwinning, Melanie Siegel, Andreas Weinmann, Hichem Snoussi

原文摘要

Digital elevation models (DEMs) can provide accurate height information, making it invaluable for analyzing the lunar surface. Due to recent advancements in computer vision, the focus is on generating DEMs using monocular depth estimation via deep learning. Classic lunar crater detection methods rely mainly on 2D image data. Our goal is to address this issue by developing a relative lunar surface height estimator that could provide additional information for crater localization. In this letter, we present a methodology that builds on the zero-shot relative depth estimation model Depth Anything V2 (DAV2). Other works have been using it as a state-of-the-art comparison for their proposed lunar DEM estimation method, but without adaptations to the target domain. Thus, it may underperform. Therefore, we propose a fine-tuning strategy with publicly available SPG-derived DEM data of the lunar surface, and refer to it as DEM-DAV2. Specifically, we introduce new learnable parameters in the Transformer-based encoder with low-rank adaptation (LoRA), and we unfreeze certain layers in the model's decoder. On our test set, the proposed DEM-DAV2 reduces the average MAE and RMSE by $\sim$80\% and $\sim$79\%, respectively, compared to the zero-shot DAV2. This demonstrates a significant improvement in performance compared to the zero-shot model, effectively transforming DAV2 into a reliable relative height estimator of the lunar surface. However, the predictions remain relative and require scale recovery from a low-resolution DEM such as SLDEM.

cs.CV

新论文 · v1arXiv:2611.24071

Measuring and Charging for Retained Context in Long-Context LLM Services

Yi Lu, Qianshi Lin, Yumeng Yang, Hongxia Wang

原文摘要

Long-context large language model (LLM) services charge for repeated context, yet billing weights need not reflect its contribution to answer quality. We present a three-layer framework separating content relevance, observable proxies and auditable positional weights. Context-ablation experiments show that relevance dominates position and that distance-decay families do not outperform a relevance-only baseline. A measured serving stack shows that cache hits reduce repeated-prefix compute cost by over two orders of magnitude at 65,536 tokens, while cold resends constrain sustainable discounts. Cross-provider measurements separate cache value from tokenisation and list-price differences. Calibrated simulations reduce the 99th percentile of session charges by 31-42%, although revenue-matched nonlinear tariffs achieve similar reductions and median charges increase. The framework provides a bounded per-token accounting rule and measured deployment constraints, rather than a validated utility estimator.

cs.PF · cs.CL

新论文 · v1arXiv:2611.24070

FDM+: A Framework for Constrained Control of the 2D Radiative Transport Equation

Hualing Zhong, Shuai Wang, Hui Xie, Heng Yong

原文摘要

Applying the frozen diffusion model (FDM) [29] to the constrained control of the 2D radiative transport equation presents several challenges. These include the dimensional difference between control variables and state functions, the resolution mismatch between traditional numerical solvers and deep learning models, and solution non-uniqueness. To address these obstacles, we propose a novel approach, FDM+, which includes three key improvements. Firstly, a decoupled prediction architecture is introduced to address the dimensional heterogeneity. Secondly, an efficient parameterization strategy resolves the resolution mismatch. Meanwhile, we derive a total error bound to analyze how the parameterization error and the model estimation error jointly affect the performance. Thirdly, the conditional distribution learned by the diffusion model is leveraged to perform design space exploration for multi-solution problems. A series of numerical experiments demonstrate that FDM+ outperforms existing methods in reconstruction accuracy and design flexibility.

cs.LG · physics.comp-ph

新论文 · v1arXiv:2611.24068

Benchmarking Reasoning Reliability,Prior_Evidence Contrast,and Spectral Discrimination in Remote-Sensing Geological Image Interpretation

Jun Wang, Fengpeng Li, Wei Han, Yuxuan Li, Tianjin Huang, Jining Yan, Sheng Wang, Kan Sun 等 9 位作者

原文摘要

Reliable lithology interpretation requires accurate geological recognition and reasoning grounded in visual evidence. Although Multimodal Large Language Models show promise in remote sensing, their reliability in geological interpretation remains unclear. We introduce HyGeoBench, a diagnostic benchmark of 10,000 questions based on hyperspectral imagery across three dimensions: reasoning reliability, prior_evidence contrast, and spectral discrimination. It covers ten task types, with both multiple-choice and open-ended response formats. Integrating hyperspectral satellite imagery with geological annotations and verified evidence, HyGeoBench examines reasoning errors, conflicts between geological priors and visual evidence, and cross-band inconsistencies. Evaluations reveal limitations in fine-grained recognition and evidence-grounded interpretation. To address these challenges, we propose HyGeoAgent, a tool-augmented framework integrating region observation, text knowledge retrieval, same-band image retrieval, and native PCA analysis through adaptive tool use. Experiments demonstrate consistent improvements over the corresponding base models, improving geological image interpretation across diagnostic dimensions.

cs.CV

版本更新 · v2arXiv:2409.08231

Design Optimization of Nuclear Fusion Reactor through Deep Reinforcement Learning

Jinsu Kim, Jaemin Seo

原文摘要

In this research, we investigate deep reinforcement learning (DRL) as a practical tool for constrained, multi-objective optimization for tokamak fusion reactor designs. The task is to optimize the design configurations that minimize building cost while achieving high energy gain and long energy confinement time, subject to the four operational limits required for stable operation. Due to the coupled physics-engineering evaluation, the resulting objectives and constraints are black-box, making it challenging to find optimal solutions with conventional search strategies. To address this, we implement Reward-Constrained Policy Optimization (RCPO) for tokamak fusion reactor design. This method integrates multiple constraints into the reward functions, guiding the policy towards feasible solutions and simplifying the design optimization process. Our numerical results show that the proposed DRL-based approach finds more feasible designs that satisfy operational constraints more efficiently than other data-driven optimization methods, demonstrating the potential of DRL for efficient future reactor design.

physics.plasm-ph · cs.AI · cs.LG

版本更新 · v3arXiv:2611.02658

OctreeSplat: Hierarchical Octree-Guided Feed-Forward Gaussian Splatting

Yibin Zhao, Yihan Pan, Wenli Yang, Jun Nan, Jianjun Yi

原文摘要

3D Gaussian Splatting (3DGS) has shown considerable potential for novel view synthesis (NVS) and 3D reconstruction. Feed-forward methods avoid costly per-scene optimization, but fixed image or spatial resolutions limit their ability to allocate primitives according to local complexity. We present OctreeSplat, a hierarchical octree-guided feed-forward Gaussian splatting. OctreeSplat extracts multi-view features and geometric priors from a frozen visual foundation model (VFM), constructs a hierarchical octree, and predicts densification scores. The resulting representation allocates Gaussian primitives at spatially varying densities according to local complexity. Stochastic solids volume rendering improves depth rendering, strengthens geometric supervision, and supports surface reconstruction. When camera parameters are available, an optional test-time optimization (TTO) module improves agreement between predicted and ground-truth parameters. Training updates only 76 million additional parameters and requires 35 GPU-hours. Experiments on Replica, TUM, Tanks and Temples, and DTU demonstrate competitive RGB view synthesis, depth synthesis, and surface reconstruction. OctreeSplat reduces the number of Gaussian primitives while maintaining high reconstruction quality.

cs.CV

版本更新 · v3arXiv:2606.20118

Pose6DAug: Physically Plausible Multi-view Object Swapping for Robot Data Augmentation

Jonghoon Lee, Seong Hyeon Park, Byungwoo Jeon, Minha Lee, Jinwoo Shin

原文摘要

Vision-language-action (VLA) policies have shown strong potential for general-purpose manipulation, yet they often fail on novel, out-of-distribution objects whose appearance or geometry deviates from the training distribution. The standard remedy is to collect multi-view teleoperation data for every failure case, but this scales poorly in both cost and time. We introduce Pose6DAug, a failure-driven data augmentation framework that turns existing successful episodes into targeted demonstrations for a policy's failure modes, without any new data collection. Our key insight is that each successful episode already encodes a physically valid action trajectory together with calibrated multi-view observations. By swapping only the manipulated object while preserving this trajectory, we obtain new and physically grounded demonstrations. However, the challenge lies in synthesizing object swaps, e.g., naive 2D video editing breaks multi-view consistency and physical plausibility, particularly under heavy occlusion and egocentric viewpoints. Our method instead operates directly in 3D, anchoring the target object with an explicit mesh driven by a temporally coherent 6D pose trajectory. This ensures geometrically consistent renderings across all camera views. Fine-tuning a VLA on data augmented by our method improves success rates by 16.5% relative to the state-of-the-art baseline on novel objects, while preserving in-distribution performance. On a real robot, it more than doubles the success rate of the base policy. These results show that multi-view and physically consistent augmentation is a practical path to scalable VLA generalization.

cs.RO · cs.LG

版本更新 · v2arXiv:2611.21885

Language Models Can Rewrite Their Own Weights

Jiajun Fan, Neil He, Changyi Yang, Andy Mao, Xi Zhu, Chaoran Cheng, Kaiqu Liang, Shuaike Shen 等 11 位作者

原文摘要

A language model is trained once and frozen. Making it better takes another training run, or a prompt, a harness or extra computation at inference that is gone once the task ends. We show a third way: to our knowledge for the first time, a language model rewrites its own weights by its own reasoning and becomes better, with no gradients and no training. We call this self-rewriting: the model is its own optimizer. In earlier work, a frontier model edited a much smaller one; here an open 27B model, given its weights, its graded answers and the ability to run itself, diagnoses its own failure and rewrites the weights behind it. It surpasses itself: its own edit makes it 6.6 points more accurate on problems it never saw. It keeps getting better: it tests its edits on practice problems and submits only what works, each edit improving on the last, and over eight consecutive sessions its accuracy only rises. It repairs a damaged copy of itself, like a broken robot that fixes itself without being rebuilt: the copy solves almost nothing, yet with no retraining and without ever reading the original weights, the model brings it back to over 80% of the original accuracy on image tasks it never saw. And the ability emerges with scale: in one model family, the smaller models make no edit or break themselves, and only the largest improves and repairs itself. Unlike test-time scaling, self-rewriting writes its gain into the weights, at no cost per answer; unlike training, it runs only forward passes, as serving does. The wall between training and inference is a limit of our tools, not a property of intelligence. We believe self-rewriting is a necessary step toward recursive self-improvement: a model that, like a human, does not separate training from inference could keep upgrading itself for life, with no larger model, no harness or prompt engineering, and no pre- or post-training.

cs.AI

版本更新 · v4arXiv:2607.27084

SciFigQual-Bench: Scientific Figure Quality Assessment with Manuscript-Linked Context

Zihan Deng, Chuanzhi Xu, Huiqi Liang, Xiaozhen Zhong, Weidong Cai, Lequan Yu

原文摘要

Scientific figure quality is bound to the manuscript: a crop can be visually clear and still contradict its caption or the paragraph that cites it. Most image quality assessment and chart-understanding methods target a detached visual surface or a question-answering score, rather than alignment among the figure, its caption, and the citing text. A single end-to-end prompt that sees the image and the text together still mixes what is visible with what the caption and the citing paragraph claim, so fluent wording can raise a visual score and missing text can be treated as low quality. To address this gap, we introduce SciFigQual-Bench, a manuscript-linked benchmark for published CS-conference figures that binds each figure to its caption and to index-resolved citing paragraphs, and scores visual clarity, layout, caption consistency, context consistency, and misleading risk, leaving a dimension unevaluated when its evidence is absent. SFQ-Agent reads the image and the text in separate calls and records modality-specific evidence. A cross-modal judge scores caption and citing-text alignment from those reports, and a deterministic runner keeps the visual scores, sets misleading risk by a fixed rule, and applies the written caps, so each dimension follows the rubric instead of a single end-to-end prompt. Experiments comparing this staged judge with single-pass and OCR-sidecar protocols find the closest point-estimate fit to mean human ratings under staged judging, while the gap between protocols remains small. Caption consistency remains the main gap, and agreement with the rater mean answers a different question from agreement among human raters. Project page: https://frankdengai.github.io/SciFigQual-Bench/

cs.CV · cs.AI

新论文 · v1arXiv:2611.24062

JEPA Dots: Dual-Latent World Model for Language-Guided Robot Control

Xianhao Pan, Chi Li, Hanchao Li, Yunxin Liu, Jiankai Sun, Luhui Hu

原文摘要

Predictive world models for language-guided robot control face a fundamental conditioning challenge: instructions should influence action selection without altering the predicted consequences of fixed actions under identical observations. To address this challenge, we introduce JEPA Dots (Dynamics-Oriented Task Separation), a compact dual-latent framework that jointly learns visual dynamics and language-guided control through separate dynamics and task representations. Language-conditioned task features guide a flow-matching policy, while masked attention restricts world prediction to instruction-independent dynamics history and supplied actions. In closed-loop evaluation on 40 trained LIBERO tasks, JEPA Dots achieves 84.9% overall success with approximately 39.1M parameters, compared with the published 78.7% of NanoVLA-S with 161.0M parameters under a different evaluation protocol. In the reference experiment, JEPA Dots also improves overall success from 76.0% for our task-ID-conditioned adaptation baseline to 84.9%, with LIBERO-Goal success increasing from 67.1% to 88.5%. Instruction-switching experiments further confirm that task representations and policy outputs respond to language changes, while dynamics representations and world predictions remain invariant under fixed observations and supplied actions.

cs.RO

版本更新 · v2arXiv:2607.01919

ElephantAgent: Contextual State Continuity in Agentic Systems

Jiankai Jin, Xiangzheng Zhang, Zhao Liu, Wenzhuo Xu, Dongdong Yang, Deyue Zhang, Quanchen Zou

原文摘要

Agentic systems enhance their capabilities by invoking external tools and maintaining persistent memory. However, recent tool and memory poisoning attacks can covertly bias agent behavior. We observe that such attacks mounted out of band share a single enabler: each succeeds only by presenting the agent with a contextual state (e.g., tool state and memory) that is not the latest faithfully evolved one. This observation suggests that a continuity check over the contextual state covers this family of attacks. We present ElephantAgent, a protocol that enforces Contextual State Continuity over the bounded, security-critical subset of the agent's context. Before each planning-and-execution cycle, ElephantAgent recomputes the digest of the local contextual state and verifies it against the latest authorized digest in a linearizable ledger of state transitions, maintained over replicated trusted hardware. The protocol also provides Historical Traceability, a secondary guarantee that keeps state transitions attributable and recoverable. We integrate ElephantAgent into DeepSeek Harness (DSH) and implement ElephantDSH, an agentic system protected by contextual state verification. ElephantDSH withstood all out-of-band attack attempts, from rug pulls and rollbacks to attacks on its own mechanism, and recovered from an in-band prompt injection in all trials. By comparison, existing defenses typically check only what enters the context: of the five we evaluated, all leaked the targeted content under the rollback, and three also under the rug pull. ElephantDSH completed all normal coding tasks with a relative overhead of 1.02-1.07$\times$ over the original DSH with model latency variation removed.

cs.AI · cs.CR

新论文 · v1arXiv:2611.24060

R$^{3}$Sieve: Suppressing Unconditional Information from Attention to Answers for High-Resolution VQA

Yifan Cao, Yawen Zhang, Xiaocui Yang, Faxian Wan, Shi Feng, Daling Wang, Yifei Zhang, Shangshu Yu 等 9 位作者

原文摘要

High-resolution visual question answering (VQA) requires models to identify question-relevant small objects in high-resolution images and select the correct answer based on what they recognize, demanding fine-grained perception. Previous methods focus on obtaining better visual evidence (image crops), but obtaining it alone is insufficient, using it effectively is equally crucial. Visual evidence should therefore be utilized as effectively as possible throughout inference. We argue that small-target evidence is obscured by unconditional information that remains observable in subsequent processing even after the condition on which it should depend is removed. Such information hinders the effective use of small-target evidence. We propose R$^{3}$Sieve (Residual-Region-Response Sieve), a training-free framework. We divide inference into three stages---localization, cropping, and answering---and introduce distinct counterfactual references to suppress interference at each stage. During localization, the Shared-Subspace Canceller (SSC) uses a generic image description to construct a counterfactual reference and obtain question-relevant attention maps. During cropping, the Unique-Contribution Spotter (UCS) constructs counterfactual references from visual and non-visual features to identify distinctive, complementary crop regions. During answering, the Content-Aligned Voter (CAV) constructs an image-free branch to calibrate answer responses. Experiments on three high-resolution VQA benchmarks show that R$^{3}$Sieve achieves the highest overall accuracy among compared methods across diverse settings, improving base models by up to 21.5 percentage points.

cs.CV

新论文 · v1arXiv:2611.24059

Replay, Don't Resample: Off-Policy Consolidation of Long-Horizon RL Experts

Zhenrui Yue, Gongxiaohui Chen, Huimin Zeng, Honglei Zhuang, Jiashuo Sun, Zhen Han, Dong Wang

原文摘要

Frontier post-training follows a common recipe: train one expert per domain with reinforcement learning (RL), then consolidate the experts into one model via multi-teacher on-policy distillation (MOPD). Such consolidation is expensive: the student samples in every environment again, an expert scores every sampled token, and the student inherits the drift of the final experts. Yet we argue that the experts' RL runs have already produced what consolidation needs: rollouts with their sampling probabilities and rewards, in the order the experts learned from them. We propose multi-expert replay policy distillation (MRPD), which replays these training logs through the student in the experts' order. At each step, the policy gradient is recomputed on these scored trajectories, and the final experts' distributions are distilled at the logged prefixes. Following the experts' own trajectory reduces the training-inference mismatch and keeps the student within the clip range, so the replay is off-policy in origin but quasi-on-policy throughout consolidation. MRPD samples nothing and calls no environment or reward model: the trainer is the sole process, and consolidating the experts costs only about a seventh of the compute used to train them. On three long-horizon agentic domains (reasoning, search and SQL), MRPD yields the best student model, substantially outperforming online RL baselines, and surpassing MOPD from the same experts at a fraction of MOPD's cost. Our analysis on the same logs shows that the replayed trajectories carry most of the gain, and that both the policy gradient and the distillation term add to it. Most interestingly, the order of the logs matters more than their amount or rewards, suggesting that expert training logs are a curriculum, not a data mixture.

cs.LG · cs.AI

版本更新 · v2arXiv:2606.26587

SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference

Haoqian Meng, Yilun Luo, Yafei Zhao, Wenyuan Liu, Huaqing Zheng, Xindian Ma, Peng Zhang

原文摘要

Low-bit floating-point formats and semi-structured sparsity are increasingly supported by modern accelerators, yet combining them for LLM activation compression remains challenging: activations contain input-dependent outliers that dominate block scales in FP4 quantization, and directly applying N:M sparsity masks discards moderate values, coupling sparsification loss with quantization error. We introduce SharQ, a training-free inference method that bridges activation sparsity and FP4 quantization through an online sparse--dense decomposition. For each activation tensor, SharQ generates an input-adaptive N:M mask to extract an outlier-dominated sparse backbone, quantizes it to FP4, and defines a dense residual relative to the quantized sparse backbone rather than the unquantized sparse values. A sparse FP4 GEMM processes the backbone while a dense FP4 GEMM compensates for both mask-induced activation loss and sparse-path quantization error. The two paths share a single FP4 weight payload with path-specific scale views, and a fused preparation kernel absorbs mask generation, residual construction, and layer normalization into one operator. SharQ requires no calibration data, retraining, or model-specific tuning. Evaluated on Llama-3.1-8B, Qwen2.5-7B, Qwen3-30B-A3B, and Qwen3-VL-8B, SharQ recovers 43--63% of the NVFP4-to-FP16 accuracy gap across language and vision-language tasks, and generalizes across NVFP4, HiF4, and MXFP4 formats. On an RTX 5090, SharQ delivers 2.2--2.4$\times$ latency reduction over FP16 and 1.2--1.4$\times$ throughput improvement over FP8 in language model serving, and up to 1.58$\times$ speedup on Wan2.2-T2V-A14B video generation when combined with SageAttention. Our code is available at https://github.com/actypedef/SharQ.

cs.LG · cs.AI

新论文 · v1arXiv:2611.24058

MIRAGE: A Framework for Multi-hop Reasoning under Dynamic Knowledge Injection

Heyang Xu, HanWen Hao, Haifeng Wang, Haoyi Zhou, Jianxin Li

原文摘要

Multi-hop Question Answering (MQA) requires large language models (LLMs) to retrieve and synthesize multiple pieces of evidence into a coherent reasoning path, which is essential for complex tasks. However, prior work on knowledge updating for MQA has largely focused on simple fact edits, overlooking a more challenging and realistic setting---\emph{Knowledge Injection}---in which the model must simultaneously accommodate both conflicting updates and entirely novel information. Under this setting, a hop-level answer is often wrong simply because the model trusts the wrong knowledge source: it clings to outdated parametric memory when the injected knowledge conflicts with it (K2), and it cannot answer when the required fact is entirely new (K3). To investigate this, we introduce NEXUS, the first benchmark tailored for multi-hop reasoning under Knowledge Injection, which explicitly covers both conflicting (K2) and incremental (K3) knowledge. We further propose MIRAGE, an efficient, training-free framework whose core component, the Correlation-aware Dual-path Reasoner (CDR), resolves such per-hop source-trust errors via P(True)-based confidence routing, complemented by a lightweight template-guided corrector for residual structural deviations. Extensive experiments across four LLM families and two benchmarks demonstrate that MIRAGE consistently improves both final-answer accuracy and hop-level reasoning accuracy, establishing a new state of the art in this challenging setting.

cs.CL

新论文 · v1arXiv:2611.24057

Faster to the Wrong Answer: The Price of Local Updates in Federated Optimization

Molly Wang

原文摘要

Local computation can improve the conditioning of federated updates but may also introduce bias, causing their fixed point to differ from the global minimizer. We characterize the minimum worst-case bias achievable under a prescribed conditioning bound. We consider strongly convex quadratic client objectives with Hessian eigenvalues in $[μ,L]$ and local updates represented by a shared positive spectral filter. Client heterogeneity is measured by the root-mean-square gradient at the global minimizer. For a uniform surrogate condition number bounded by $χ\in[1,L/μ]$, we establish the exact worst-case bias per unit heterogeneity, $(L-μχ)/(2μL\sqrtχ)$, and show that local gradient descent with sufficiently small step sizes and proximal updates attain this bound. Our analysis reduces populations of arbitrary dimension, including noncommuting Hessians, to distributions over scalar curvatures. This reduction provides optimality certificates and worst-case constructions involving at most three scalar clients. We further extend the sharp proximal bias bound to smooth strongly convex objectives beyond quadratics and quantify the error introduced by inexact local solves.

cs.LG

新论文 · v1arXiv:2611.24055

Grading Executable Legal Rules Against Independent Records: Transitional Provisions and Korean SME Status

Jinhyung Bae

原文摘要

Laws are increasingly turned into programs that decide cases, such as whether a firm was a small or medium enterprise (SME) on a given date, and large language models (LLMs) now write them. The programs are usually graded against answers written by someone who read the same law, so a program can score well by sharing the grader's reading rather than by getting the law right. LexTemporis grades them instead against records of what actually happened: 10,901 firm-years of SME status disclosed by Korean listed companies, with every version of the governing rules. An appendix adds comparison cases from US income tax and German hospital minimum volumes. The date changes outcomes mainly through provisions outside the main statutory text: one transitional provision raises agreement with the disclosures from 83-86% to 89-90%. The disclosures also test the author's reading of a contested provision, favouring it in 130 of 189 firm-years where readings diverge. Four LLMs wrote the rules, given different selections of the law. A pre-registered comparison finds no condition better across all four models. In exploratory analyses, supplying the right provisions looks better graded against the author's reading than against the records: the advantage shrinks for three models (for one, to between a quarter and three-fifths of its size), and per firm-year it is several times larger where the author's reading is wrong, partly by construction, since there reproducing that reading more faithfully must lose. Grading legal programs against the grader's own reading can overstate how well they work.

cs.CL · cs.AI

版本更新 · v2arXiv:2611.21519

Embedding Cosine Measures Text Length Before It Measures Homogenization

Fengming Liu, Mingyu Shu

原文摘要

Mean pairwise cosine similarity between transformer embeddings is a standard measure of whether online discourse is converging under generative AI. When the embeddings are mean-pooled outputs of an encoder never trained for sentence similarity, this measure mostly tracks text length. In 6.06 million Reddit comments, an uncentered DeBERTa-v3 measure had suggested that only conservative discourse broke toward homogeneity after late 2022. Truncating the same comments from 10 to 150 words raises their mean cosine by eight times the reported break. A placebo that swaps every comment for a random one of equal length reproduces 81% of the asymmetry, and a fixed-content simulation reproduces 78%. The leading principal direction carries 99.9% of raw DeBERTa homogeneity, and once length is handled the asymmetry disappears under every encoder. On 575,000 Stack Exchange answers the measure reports a post-ChatGPT homogenization of technical sites that the placebo fully reproduces, although answers grew longer. A census of 31 studies locates the risk where anisotropic encoders meet time-series or length-unequal designs. A reverse control on a well-known experiment shows that centering with renormalization can reverse a genuine effect. We propose a low-cost diagnostic protocol. Corrected measures show that the two political camps converged between 2019 and 2022 and stopped converging before ChatGPT. Cross-camp replies engage their targets more closely than same-camp replies.

cs.CL · cs.CY · cs.SI

版本更新 · v2arXiv:2611.10592

MotionGraphicsBench: Evaluating Motion Graphics Generation with Programmatic Verification

Jiaju Ma, Ashvini Kumar Jindal, Chuhua Wang, Sumithra Bhakthavatsalam

原文摘要

Large language models can generate motion graphics from static images and natural language requests, but it remains difficult to evaluate how well models follow requests that combine spatial, appearance, and temporal requirements. We introduce MotionGraphicsBench, an expert-curated benchmark of 100 instances for animating an input static SVG with a motion edit prompt. By supplying all required visual elements, our editing task reflects a designer-like workflow and isolates motion generation from graphic generation. Derived from real-world Lottie motion designs, our benchmark contains diverse combinations of spatio-temporal animation behaviors. Each benchmark instance pairs its model input with a deterministic, human-readable verification program that accepts different valid implementations and reports which prompt requirements pass or fail, rather than comparing against a reference animation or using a VLM judge. We construct these programs with a multi-agent pipeline using human-labeled animations and synthesized test cases. On a balanced test set of 180 animations, our programs achieve 93.3% raw agreement with human judgments, outperforming frontier VLM judges by 4.4%-12.8% while achieving at least 90% recall for both human approvals and rejections. Across 47 configurations from 32 LLMs and 12 providers, pass@1 ranges from 5.0% to 68.0%. Across 15 models at their highest reasoning effort, additional attempts improve every model, but pass@32 still ranges from 38.0% to 95.0%. Our analysis of verification programs reveals that generated animations often fail to match reference poses, follow requested motion trajectories, or coordinate related events. We fine-tune Qwen3.8-27B with group relative policy optimization (GRPO) using our verification programs as rewards, improving pass@1 from 30.31% to 52.19% on held-out instances. Our project page is at https://motion-graphics-bench.github.io/.

cs.CV · cs.AI

新论文 · v1arXiv:2611.24052

From Data to Valves: A Self-Learning, Production-Grade Edge Stack for Sustainable HVAC Control

Abbas Raza Ali, Muhammad Hamza Rehman, Xu Li, Muhammad Ajmal Siddiqui

原文摘要

Learned HVAC controllers rarely leave the simulator, and those that do are typically trained offline, frozen at deployment and retrained from the cloud. We present a production stack in which three learning components run on one NVIDIA Jetson Orin gateway per building, adapt continuously from the building's own data, and verify their own effect. (a) A compact transformer-encoder forecaster, one per equipment group with learned per-unit embeddings, predicts every air-side unit's next-cycle temperature and learns prequentially, one gradient step per five-minute cycle, behind five independent gates that bound the influence of any corrupt reading. (b) A distributional soft actor-critic agent, one shared policy across unit types, converts the measured state into a chilled-water valve position under a structural safety envelope that no learned component can soften; its second generation learns nightly on the device from recorded transitions alone, with a behaviour anchor and critic calibration terms, is scored by four model-free metrics and promoted by atomic hot-swap. (c) A change-point regression baseline held to ASHRAE Guideline 14 supplies the counterfactual against which savings are certified. On eighteen months of data from a hotel the forecaster converges from random weights within a week and reaches 0.37, 0.39 and 0.11 degree C MAE for air-handling, fresh-air and fan-coil units with no retraining event; on a different hotel the baseline reproduces a 46-day held-out period to within 0.07%. The RL agent has completed its in-simulator campaign and runs in shadow on 104 units; we report measured latency and memory of every service on a shared 15 GB device, the systems lessons of the campaign, and precisely what the evidence does and does not yet establish.

cs.LG · cs.AI

新论文 · v1arXiv:2611.24051

Attention Is Not Relevance: Challenging the Causality of Attention in Autonomous Driving

Luca Lusvarghi, Javier Gozalvez

原文摘要

Attention has become one of the most widely adopted solutions for estimating the relevance of surrounding objects and interpreting an autonomous vehicle's driving decisions. However, its ability to capture the causal impact of individual input elements on an AI model's output has long been discussed in the literature, particularly in the Natural Language Processing (NLP) domain. This work brings this discussion into the autonomous driving domain by analyzing the accuracy of attention in estimating the impact of an input object on an autonomous vehicle's driving, i.e., its relevance. To this end, we compare different attention-based relevance estimation methods against a causal reference of object relevance obtained through counterfactual analysis. We perform our analysis on more than 80,000 synthetic driving scenes using an autonomous vehicle based on PlanT 2.0, a state-of-the-art Transformer-based model for object-level planning. Results show that attention provides a correlational rather than causal indication of object relevance. Attention-based relevance estimation methods do not consistently identify all relevant objects and tend to incorrectly classify as relevant objects with little or no impact on the vehicle's driving. In particular, results show that attention can effectively identify relevant objects in simple single-lane, straight-road scenarios, but its accuracy degrades in more complex settings, such as multi-lane roads and intersections. These findings highlight the limitations of attention as a proxy for object relevance, calling for novel methods able to accurately capture the causal impact of individual objects on the vehicle's driving and distinguish it from correlational associations.

cs.AI · cs.RO

新论文 · v1arXiv:2611.24049

A Hybrid RAG Framework for Deceptive Review Detection: Integrating RoBERTa-LoRA, FAISS-Based Retrieval, and LLM Reasoning

Andrei Kostin, José Ángel Diaz-Garcia, Karel Gutiérrez-Batista

原文摘要

The growth of deceptive online opinions poses a critical risk of misinformation and fraud. This paper proposes a Retrieval-Augmented Generation (RAG) system to automatically distinguish spam reviews. The architecture integrates RoBERTa with Low-Rank Adaptation (LoRA) classification, FAISS semantic vector retrieval, and a Large Language Model (LLM) for final reasoning. The system was evaluated across balanced and severely imbalanced real-world scenarios. Results demonstrate that five epochs of fine-tuning enable robust domain transfer, achieving an F1-score of 0.90 in cross-domain evaluation, outperforming baselines (OpenAI Detector,F1 = 0.82) and demonstrating competitive performance against specialized in-domain SOTA models (fakeRoBERTa,F1 = 0.97). The proposed architecture provides evidence-backed explainability and statistically significant precision gains while maintaining comparable overall detection performance.

cs.CL

版本更新 · v2arXiv:2611.23158

MoE--LoRA Methods for Multilingual Adaptation: A Systematic Empirical Study

Nikit Srivastava, Joel Dag, René Speck, Axel-Cyrille Ngonga Ngomo

原文摘要

Routed low-rank adapters offer a parameter-efficient means to improve the performance of large language models (LLMs) across tasks. Hence, it is assumed that they can also be used to improve the support of low-resource languages in LLMs, which currently serve English far better than the thousands of other languages spoken worldwide. However, this assumption has never been verified empirically under a shared multilingual evaluation protocol. We address exactly this gap by comparing eight low-rank adaptation (LoRA) methods on a frozen base model, using 146 training runs across 200 unique languages and more than 10,000 evaluations. Our results suggest that a reduced-scale search can identify viable configurations of LoRA methods but does not reliably predict their relative performance, as the top-ranking configuration is never best at scale. LoRA reliably improves language-model fit on translation data, consistently degrades reading comprehension, and benefits low-resource languages most. These findings suggest that LoRA methods should be regarded as targeted tools rather than general-purpose means for improving the multilingual capabilities of LLMs.

cs.CL · cs.LG