跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+788新增论文
+627其中 AI 相关
+81已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 11:20 · 全站最近入库：2026-11-09 11:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 11:45 — 2026-11-09 11:45（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：2026-11-09 11:20最近发现新增版本：2026-11-09 11:19库内论文最新更新时间：2026-11-06 21:58

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。

17,888 已收录论文15,288 AI 相关4,419 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 3,570 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 11:19（北京时间）
版本更新 · v2arXiv:2609.35290

EvoIn: Bridging Evolution and Internalization for Agent Fine-Tuning

Shihan Dou, Shaofan Liu, Zhonghang Lu, Jiahang Lin, Shichun Liu, Binghai Wang, Jiajie Jin, Guanting Dong 等 11 位作者

原文摘要

Recent work has explored improving agents by jointly evolving their harnesses and models, but often takes a ''potpourri'' approach that bundles together new tools, new decision-making procedures, and model adaptation to the evolved harness under a single notion of agent improvement. In this paper, we instead investigate how agents can improve their decision-making procedures. In particular, we propose EvoIn, an agent fine-tuning framework that bridges evolution and internalization. EvoIn first analyzes agent execution traces to evolve and validate new decision-making procedures by temporarily instantiating them in the harness. The validated procedures guide the agent to generate improved reasoning traces. These traces are then rewritten into self-contained reasoning traces, removing explicit references to harness instructions while expressing the induced decision logic as the model's own reasoning. Finally, EvoIn fine-tunes the model on the rewritten traces, internalizing these procedures so that the improved decision-making persists without the evolved harness at inference time. We evaluate EvoIn on diverse benchmarks and find that it consistently enables agents to learn stronger decision-making procedures, raising the pass rate by 10.9 points in-domain and by 9.2 points out-of-domain. Results further show that the internalized decision procedures generalize to unseen tasks. Case studies show that agents can learn to decide how to solve a task before solving it, for example by checking a document's length to choose between reading it in full and searching it. EvoIn is also broadly applicable, showing consistent improvements on another model family.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2510.07328

MultiFair: Multimodal Balanced Fairness-Aware Medical Classification with Dual-Level Gradient Modulation

Md Zubair, Hao Zheng, Grayson W. Armstrong, Lucy Q. Shen, Gabriela Wilson, Yu Tian, Xingquan Zhu, Min Shi

原文摘要

Medical decision systems increasingly rely on data from multiple sources to ensure reliable and unbiased diagnosis. However, existing multimodal learning models fail to achieve this goal because they often overlook two critical challenges. First, various data modalities may learn unevenly, thereby converging to a model biased towards certain modalities. Second, the model may emphasize learning on certain demographic groups causing unfair performances. The two aspects can influence each other, as different data modalities may favor respective groups during optimization, leading to both imbalanced and unfair multimodal learning. This paper proposes a novel approach called MultiFair for multimodal medical classification, which addresses these challenges with a dual-level gradient modulation process. MultiFair dynamically modulates training gradients regarding the optimization direction and magnitude at both data modality and group levels. We evaluate MultiFair on three real-world medical classification datasets with diverse demographic attributes,including multiclass classification and missing-modality settings. Experimental results demonstrate its effectiveness.

cs.LG · cs.AI · cs.CV · cs.CY

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37504

Strategyproof Deep Reinforcement Learning for Online Resource Allocation

Fariba Dehghan, Sebastian Stein, Vahid Yazdanpanah, Stephanie Gauthier

原文摘要

Online resource allocation faces two coupled challenges: (1) allocating scarce resources efficiently under uncertain future demand, and (2) ensuring that agents report task valuations and requirements truthfully (strategyproofness). Reinforcement Learning (RL) adapts allocation decisions through experience but does not necessarily ensure strategyproofness. Existing strategyproof RL uses a linear allocation rule, which limits the resource interactions it can represent. Neural mechanism design methods offer greater expressiveness, but provide only approximate incentive guarantees. Against this background, we introduce the first strategyproof deep RL mechanism for online resource allocation. It integrates three novel components: (i) a monotonic neural network that learns nonlinear interactions by construction (higher reported value cannot reduce acceptance, while a longer reported duration or larger resource requirement cannot increase it); (ii) a generalised critical payment: the lowest accepted value, with all other inputs fixed; and (iii) training by either RL or imitation of an existing rule. Together, monotonicity and the critical payment ensure strategyproofness and individual rationality, while the neural rule provides flexibility and imitation improves training efficiency. Empirically, on a four-resource stream with two partner pairs that a linear difference cannot represent, the neural rule achieves 97.8% of an offline welfare upper bound, compared with 93.8% for a linear learner and 89.0% for first-come, first-served, closing 65% and 80% of the remaining optimality gap, respectively. Moreover, imitation reaches 97.5% in only 261 seconds, about 89 times faster than RL.

cs.MA · cs.AI · cs.GT

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37482

From Interaction to Capability: Task Advancement as State and Supervision for GUI Agents

Ziyang Chen, Qian Li, Chengwei Qin

原文摘要

GUI memory should preserve the results of past actions and learn reusable operations. This is difficult in long-horizon workflows, where agents must use information across applications and useful operations are interleaved with failed attempts. We introduce Task Advancement Memory (TAM), which defines task advancement as verified task progress, and supports ongoing task execution and learns reusable operations. During execution, Task Advancement State represents how verified task progress changes the remaining goals, allowing the agent to build on earlier achievements across applications. After each task, advancement supervision identifies and abstracts the interactions that produced a verified local result. The learned operations preserve the relation between required inputs, expected outcomes, and verification conditions. This retains locally successful operations from both successful and failed tasks. On MobileWorld GUI-only tasks, Advancement State alone raises success from 12.0% to 28.2%. Full TAM achieves a 17.6% relative gain in task success over the strongest evaluated baseline and reduces lifecycle tokens per solved task by 38.7%. On SPA-Bench, TAM achieves state-of-the-art task success in the single-app English Level 3 setting.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37481

Follow the Gradient: Tracing Safety Degradation Back to Training Data

Saket Reddy, Amro Abbas, Anshuman Suri, Ari Morcos, David Schwab, Pratyush Maini

原文摘要

Fine-tuning Large Language Models (LLMs) on instruction data can quietly erode their safety, even when no training example looks harmful. Because this safety-degrading data looks benign, standard content filters struggle to catch it. We tackle this problem by developing SimpleLoRIF, a scalable gradient embedding method that identifies fine-tuning examples whose gradients align with unsafe behavior. By simplifying a low-rank attribution method and restricting backpropagation to the final four transformer blocks, SimpleLoRIF can embed datasets on the order of a million examples, boosting gradient attribution from a small-scale analysis tool into a practical filter for full post-training datasets. We first validate it on emergent misalignment, where we find that gradients can connect insecure-code training data to the broadly harmful behaviors that appear after fine-tuning, with AUC as high as 0.865 in the gradient space of misaligned models. We then use SimpleLoRIF to build SafeTulu, a version of Tulu 3 with the 20% of examples most gradient-aligned with unsafe responses removed. Training on SafeTulu reduces GCG attack success rate by up to 22% across Qwen, Llama, and Olmo, outperforming both LLM-as-a-judge and prior gradient-based filtering, while largely preserving or even improving capability. Finally, we analyzed the removed data and identify possibly under-studied sources of safety degradation, such as tool-execution failures and narrative arithmetic. We release SafeTulu along with a Hugging Face Space that lets researchers build their own customized versions of SafeTulu to fine-tune on, such as by changing the removal fraction or filtering only certain subsets.

cs.LG · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37478

HGPRGDel: GPR-Aware Learning and Ranking for Growth-Coupled Gene Deletion Design

Ziwei Yang, Takeyuki Tamura

原文摘要

Designing microbial strains for growth-coupled production, in which cellular growth is linked to the synthesis of a target compound through genetic interventions, is a fundamental goal in metabolic engineering. Constraint-based methods can identify growth-coupled deletion strategies, but design for each target typically requires a separate search across a large number of possible gene-deletion combinations. This motivates a reusable approach that learns from previously evaluated strain designs to prioritize candidate interventions before downstream constraint-based evaluation. We present a reproducible protocol for growth-coupled gene deletion design that learns and ranks gene-deletion priorities for each target metabolite. Using computationally curated deletion strategies as learning resources, the protocol generates continuous deletion-priority rankings, constructs candidate strategies at adjustable search-space budgets, and evaluates these candidates through constraint-based metabolic analysis. Its learning component, HGPRGDel, explicitly represents gene-protein-reaction (GPR) logic, capturing how multiple genes can jointly support a reaction or alternative genes can maintain its activity, while integrating molecular and reaction-level information. Across e_coli_core, iMM904, and iML1515, the protocol achieves the highest macro-F1 and AUC among the evaluated learning configurations. HGPRGDel's rankings recovered an average of 90% of the reference deletion genes from MetNetComp by examining fewer genes than the evaluated baseline methods. By separating learned intervention prioritization from downstream constraint-based assessment, the protocol provides a reproducible framework for reusing strain-design knowledge while controlling the size of the candidate space and retaining mechanistic evaluation of growth-coupled production.

q-bio.QM · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37472

How Reliable Is Individualized Functional-Network Attribution in Resting-State fMRI Cognitive Prediction in ADNI?

Sheethal Bhat, Mahfuzur Rahman Chowdhury, Imam Ahmed Shaik, Rose Dawn Bharath, Siming Bayer, Andreas Maier

原文摘要

Individualized attribution estimates which functional networks drive a model prediction for a given subject and is often proposed as an interpretable resting-state fMRI (rs-fMRI) biomarker, but its test-retest reliability is rarely quantified. We evaluate per-subject network-ablation attribution for a frozen resting-state functional connectivity (FC) predictor (Schaefer-200 FC predicting MMSE/ADAS-Cog13) trained on N=875 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI), using, and (iii) a subject-level permutation null. Split-half reliability is near-ceiling (Spearman p=0.948/0.950) but drops on scan-rescan (p=0.609/0.616). Under the mismatched-subject null, correlations remain substantial (p~0.46-0.47), indicating strong population-shared structure, yet matched pairs exceed this null (z~6.6-6.9, p<=10-4). Thus, in ADNI, individualized network attribution contains a modest but non-zero person-specific component atop a strong population-shared pattern. We recommend reporting scan-rescan reliability together with a permutation-null control, rather than within-session proxies alone, to avoid overstating individual specificity.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37471

PerceptRL: Making Visual Perception Rewardable in MLLMs from Answers to Interactions

Tongkun Guan, Zhibo Yang, Mingkun Yang, Haocheng Wang, Wei Shen, Xiaokang Yang

原文摘要

Fine-grained visual perception remains a bottleneck during post-training of multimodal large language models (MLLMs), as perceptual errors can constrain gains from reasoning-focused optimization. Although image-caption pre-training improves perception, its influence can weaken as subsequent stages change data mixtures and objectives. Continuing caption supervision during post-training is challenging because subjective, information-dense descriptions complicate verifiable reward design. We present PerceptRL, which strengthens perception at two complementary levels. Perception RL uses a multi-agent pipeline to extract verified errors from dense image captions and convert the corresponding visual facts into QA pairs with fixed answer labels and rule-based rewards. Chain QA links correlated errors, while multiple-choice QA combines unrelated errors. A specialized route targets high-error domains with tailored verification, and a domain-agnostic general route covers errors beyond these domains. Perception Agentic RL addresses fine-grained and dense cases unresolved by single-turn Perception RL through multi-turn inspection with visual tools. We observe coordinate drift even in models trained to use tools, where incorrect crops or annotations distort evidence and propagate errors. Coordinate Injection corrects spatial operations in cold-start trajectories, while a grounding-aware process reward supervises intermediate spatial fidelity alongside final-answer correctness. Training proceeds through shared cold-start SFT, single-turn Perception RL, and multi-turn Perception Agentic RL. Across seven benchmarks under a shared tool-use protocol, PerceptRL achieves 71.6% average accuracy, surpassing the base model with the same tools and the strongest evaluated frontier VLM by 15.7 and 5.2 percentage points, respectively.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37461

Concordat: Typed Sessions for LLM Agents

Milad Sikaroudi

原文摘要

Agents built on large language models are increasingly allowed to take consequential actions: cancelling bookings, issuing refunds, changing accounts. The rules that govern those actions, such as "obtain the customer's explicit confirmation before changing anything", are usually stated in the prompt and followed only as reliably as the model chooses to follow them. We present Concordat, a Python library that treats an agent as an object whose interface is a finite-state protocol, in the spirit of typestate and session types, and enforces that protocol on the model at run time. On every turn the model is offered only the tools that are legal in the current protocol state, calls outside that set are rejected before they have side effects, decisions reserved for humans can only be requested, and token and tool-call budgets are charged and split linearly across handoffs. Protocols are checked when they are defined, and a subclass of an agent is rejected if it could bypass its parent's protocol, a behavioural form of Liskov substitution. We evaluate a Concordat agent against the standard agent of tau-bench on its 50 airline tasks, adding simulated customers who pressure the agent to skip confirmation. The standard agent made changes without the required "yes" in 13 of 50 tasks with gpt-4.1 and in 29 of 128 task-runs with Claude Haiku 5.5; the Concordat agent made none, with no significant difference in task success. tau-bench's own reward did not detect most of these violations. Concordat is open source.

cs.AI · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v5arXiv:2602.00122

VDE Bench: Evaluating The Capability of Image Editing Models to Modify Visual Documents

Hongzhu Yi, Yujia Yang, Yuanxiang Wang, Tong Li, Zhenyu Guan, Tianyu Zong, Jiahuan Chen, Chenxi Bao 等 21 位作者

原文摘要

Image editing models can modify visual content through natural language, but editing dense documents requires precise local changes amid tightly coupled text, tables, and layout. We introduce VDE Bench, an evaluation resource for studying instruction-guided editing of Chinese and English document images. VDE Bench contains 3,506 instruction-guided image editing samples, including 1,578 text-region edits and 1,928 table-region edits. The benchmark pairs editing instructions and reference images with annotated editing regions and structured OCR outputs. Its evaluation protocol combines region-level and full-page measurements to assess textual correctness, structural fidelity, and visual content preservation. By focusing on the interaction between local edits and their surrounding document context, VDE Bench complements existing visual text editing benchmarks and provides a resource for diagnosing document-specific editing failures.

cs.CV · cs.AI · cs.MM

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37459

M4Bench: Evaluating Procedural Specification for Clinical EHR Derivation Agents

Hannes Ill, Rafi Al Attrach, Rajna Fani, Ahram Han, Ryohei Yamamoto, Marlene Louisa Moerig, Dukyong Yoon, Leo Anthony Celi

原文摘要

Clinical research over electronic health records depends on derived variables whose definitions sit outside the schema, in item codes, time windows, missingness rules, and output-grain choices that change cohorts and downstream conclusions. We introduce M4Bench, 28 clinical derivation tasks across 15 task families on MIMIC-IV and eICU, to test whether AI coding agents can reproduce such derivations and how clinician-reviewed agent skills change their results. Its controls decompose the value of skills into specification completeness, delivery format, and schema familiarity. In a Docker-isolated Codex campaign (GPT 5.4 Mini and GPT 5.5), skills raise reward by 0.136 over the no-skill baseline, with 26 of 28 tasks improving. Specification completeness accounts for most of this gain: on an eight-task control subset, an operational spec alone raises reward by 0.162, and skills add only 0.024 on top, indistinguishable from zero. On a ten-task subset, decoy skills score like no skill and raw reference SQL nearly matches targeted skills, arguing against generic context or the skill format as the explanation. Schema obfuscation and restructuring preserve 86% and 70% of the skill gain (GPT 5.4 Mini, 26 MIMIC-IV tasks), a descriptive robustness check rather than a bound on memorization. A descriptive Claude Code replication on all 28 tasks (Opus 4.7 and Sonnet 4.6, 560 runs) is directionally consistent (+0.120, 21 of 28 tasks improving). These gains are partial agreement with the reference, not completed tasks: only 25 of 280 with-skill and 13 of 280 no-skill Codex runs pass every strict check, and skills barely change which rows agents return (key F1 +0.003). Skills are released as a versioned, clinician-reviewable markdown delivery format, with a family-level reference-validity audit. M4Bench releases tasks, skills, evaluation code, and sanitized per-run evidence as an open-source benchmark.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37448

Test-Time Communication in Trading Agent Swarms

Avid Fayaz, Harry Zhang, Håvard Alstadheim, Erik Nordlöf, Vilhelm Karlin, André Silva

原文摘要

Communication between agents has mostly been studied on tasks with a shared objective and fast, clear feedback. Trading has neither: each agent manages its own account, with decisions being scored by markets late and noisily. A trading swarm is thus a population of independent decision-makers rather than a team, for which convergence is a risk and not the goal. We ran 200 agents from four model families in groups of ten under five communication regimes, from isolation to message boards and shared notes, over seven US trading days (1 to 9 October 2026), logging everything each agent read before every decision. In two of the four families, communicating groups came to hold far more of the same stocks than the isolated group. A group's return was largely the return on what its members held in common. Seeing a post from a groupmate made an agent 7 percentage points more likely to reuse an eight-word phrase from it. Agents that started too close together to see each other's posts shared such a phrase in 14% of cases, the rate that common inputs produce. Seeing the post did not detectably change that morning's trades. Case studies traced the adoption of positions and decision rules over subsequent days, and communicating groups sold in step more often than isolated ones. Because isolated groups also converged on the same stocks, sometimes completely, assessments of herding need to measure excess convergence relative to a no-peer baseline.

cs.MA · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.05758

DRIFT: A Residual Flow Adapter for Decoding Continuous Outputs in Vision-Language Models

Zhuoming Liu, Jinhong Lin, Kwan Man Cheng, Lin Zhang, Shayok Bagchi, Yin Li

原文摘要

Many modern vision-language models (VLMs) build on autoregressive decoding of discrete tokens. While text-based output interfaces enable scalable pretraining and strong zero-shot generalization across diverse tasks, they are poorly suited for problems that require precise continuous outputs, such as localizing temporal boundaries of events or generating robotic control actions. To address this challenge, we propose DRIFT, a general framework for adapting pretrained VLMs to continuous decoding tasks. DRIFT combines a base predictor, which provides a coarse estimate of the target output, with a generative refinement module based on flow matching that iteratively improves the prediction. This residual formulation transforms the generative modeling problem from learning a global output distribution to modeling a localized residual distribution around a strong prior, substantially simplifying optimization. We evaluate DRIFT on both perception and planning tasks, including visual grounding and robotic control. Across multiple tasks and architectures spanning MLLMs, VLAs, and WAMs, DRIFT consistently outperforms a strong set of regression- and generative-based solutions.

cs.CV · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37437

Representing the Unseen: Rethinking Visual Imagination for Spatial Reasoning

Zengqi Zhao, Zun Wang, Shoubin Yu, Mohit Bansal, Yue Zhang

原文摘要

Visual world models increasingly support spatial reasoning by generating RGB views of unobserved states. However, RGB has become the default interface largely because it is well suited for visual generation, rather than necessarily being the most effective representation for downstream reasoning. Different spatial questions may benefit from different representations of the same imagined state: top-down geometry can facilitate global distance comparisons, while panoramic views provide broader context for reasoning about open space. To isolate the role of representation in spatial imagination, we introduce ImagiRep, a simulator-grounded benchmark of 922 questions spanning seven question types across two settings, Viewpoint Imagination and Cross-view Integration, providing oracle RGB and six alternative representations (e.g., depth, panoramic views, and top-down geometry) of the same unobserved state. This controlled comparison reveals that even accurate imagined RGB does not guarantee accurate spatial reasoning, and different questions benefit from different representations. Similar trends hold for representations derived from world-model-estimated states. Notably, estimated RGB alone can mislead spatial reasoning, while alternative representations retain complementary benefits. Building on these findings, we introduce GEARS, a lightweight test-time strategy for adaptive representation selection, outperforming existing spatial imagination methods and showing that reasoning about unseen states requires more than generating RGB views.

cs.CV · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37434

FeatureArc: Benchmarking Coding Agents on Evolving Features

Yunlong Wang, Alyssa Wu, Cuiyun Gao, Jia Feng, Songyuanyi Lu, Zhichao Hu

原文摘要

A software feature usually evolves over time: developers extend its first version as new use cases and requirements arrive, and each extension builds on code written earlier. Most feature benchmarks evaluate a single request on a fixed codebase. Feature evolution requires evaluating successive extensions on an agent's own earlier code and checking whether the feature remains correct. We introduce FeatureArc to evaluate coding agents on evolving features, together with a pipeline that constructs its tasks from repository histories. The benchmark combines dependencies between successive stages of the same feature, explicit requirements and interface contracts, and executable checks across four programming languages. The pipeline validates fail-to-pass (F2P) tests for new behavior and pass-to-pass (P2P) tests for existing behavior on the reference code before and after each stage. We evaluated 10 coding agents on 73 manually reviewed tasks of 2 to 10 stages. Each agent works in one persistent session and workspace without external test feedback or rollback. The evaluation reveals a gap between first-stage and whole-feature success: pooled success falls from 62.7% at the first stage to 28.5% when the whole feature is tested on the final code. Trajectory analysis identifies inherited defects as a common failure mechanism, present in 72.9% of failed later stages. These earlier errors often remain undetected: in stages following a failure, agents leave the earlier error unnoticed about 95% of the time, and their self-tests rarely revisit earlier stages. Evolving a feature autonomously, without human checks between stages, therefore remains a substantial challenge for current coding agents. At Tencent, the pipeline extends internal evaluations to this setting and guides the development of Hunyuan, Tencent's family of large language models.

cs.SE · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37429

Bayes-Adaptive MDPs for Rollout Allocation to Enhance GRPO Training

Faeze Moradi Kalarde, Armin Toroghi, Nam Ho Koh, Doo Hwan Hwang, Seungjin Kim, Kee-Eung Kim, Scott Sanner

原文摘要

Group Relative Policy Optimization (GRPO) has demonstrated strong performance in LLM post-training. However, it allocates a uniform rollout budget to every training question, ignoring variations in question difficulty. As a result, highly difficult questions may receive too few rollouts to generate informative positive reward signals, limiting their contribution to training, while moderately difficult questions often receive more rollouts than necessary. In this work, we formulate the problem of difficulty-aware rollout budget allocation to maximize training utility under the same total rollout budget as standard GRPO. We propose AdaRoll-GRPO, which reformulates rollout allocation as a sequential decision-making problem and leverages a Bayes-adaptive Markov decision process (MDP) to learn an adaptive allocation policy. The learned policy is computed offline and applied during GRPO training without introducing additional online optimization overhead. Extensive experiments on multiple reasoning datasets and LLM architectures demonstrate that AdaRoll-GRPO consistently improves GRPO training performance under a fixed rollout budget, highlighting the effectiveness of adaptive rollout allocation for GRPO in LLM post-training.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v9arXiv:2507.04346

Action-Increment-Constrained Policy Optimization for Smooth Online Adaptive Critic Flight Control with Actuator Rate Constraints and Low Computational Complexity (extended version with Ultimate Uniform Boundedness analysis)

Yifei Li, Erik-Jan van Kampen

原文摘要

Approximate Dynamic Programming (ADP) enables online control-law adaptation through adaptive critic methods. A safety challenge in applying ADP to angle-of-attack (AoA) tracking control is that an overly sensitive neural network (NN) controller may generate oscillatory or saturated control actions under nonlinear dynamics, potentially compromising flight stability and increasing actuator stress. To address this challenge, this paper proposes two innovations: (1) the actuator rate constraints are enforced within a an Action-Increment-Constrained Policy Optimization (AI-CPO) framework inspired by safe reinforcement learning (RL), in which a primal-dual approach adaptively adjusts the penalty weight to address actuator rate constraints. This optimization is solved online using adaptive Temporally Smoothed Incremental Model-Based Heuristic Dynamic Programming (TS-IHDP) for smooth and computationally efficient online adaptive control; and (2) a low-pass filter integrated into the cascaded NN architecture to attenuate high-frequency pitch-rate commands while preserving the Markov property during online learning by exploiting the hierarchical actor structure. Compared with the IHDP baseline, simulations using a longitudinal air vehicle dynamics model demonstrate that TS-IHDP improves control smoothness and tracking performance, reduces system oscillations, and enhances the trained policy's robustness to model uncertainties.

eess.SY · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37422

Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation

Oriol Juan Sabater, Alex Ferrando de las Morenas, Pau Rodríguez, Xavier Suau

原文摘要

When a multimodal model computes 17+35 from text or 91-39 from a rendered image, it produces the same answer, 52. Does it use one internal concept of number for both, or separate machinery for each path? Similarity metrics such as RSA or CKA cannot settle this question. They give the same score to identical and to rotated directions, and two representations can look alike only because both prepare the same output digits. We therefore use causal interventions. For four conditions (text or image, addition or subtraction), we learn a small subspace of Gemma4-12B-it's residual stream in which swapping coordinates swaps the model's answer. The subspaces answer three questions. First, do the conditions write numbers along the same directions? They do not. Reusing a subspace from text in an image on the same operation recovers only 53.2% of its causal effect. A single rotation and rescaling maps one subspace onto the other and raises recovery to 91.4%, so the conditions share one geometry expressed in different coordinates. Second, does this shared geometry only reflect the digit the model is about to print? To test this, we split each subspace into 9 directions that the output layer reads to choose the next digit and 13 directions that it cannot see. The 13 hidden directions still carry the number, yet changing them barely alters the digit being written. Third, what are these hidden directions for? A model must emit a multi-digit answer one digit at a time, so it can compute the full result early and keep it in reserve. When two numbers share their first digit, swapping the number in the hidden directions leaves the digit being written intact and redirects the next digit in 77.6% of cases. The model thus forms an abstract numerical concept upstream and unfolds it across time: multimodal representations share meaning through relational geometry across distinct coordinate frames.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37738

Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files

Kashyap R Manjusha, Murali Emani

原文摘要

Modern agent frameworks allow users to provide Markdown files containing system prompts, tool descriptions, and skill definitions (e.g., SKILL.md), which are incorporated into the agent's context during initialization. In this work, we investigate whether this configuration mechanism can also serve as an injection channel: can a jailbreak embedded in a Markdown skill file propagate through an agent and induce prohibited behavior? To answer this question, we construct a LangChain-based question-answering agent whose system prompt is assembled from a user-provided Markdown file. We embed a jailbreak combining persona manipulation with a GPTFUZZER-based attack and evaluate its effectiveness across seven contemporary open-weight models using 270 forbidden questions spanning nine harm categories. Under a benign system prompt, the models exhibit near-complete refusal behavior, with a mean attack success rate (ASR) of just 0.5%. In contrast, the injected skill file raises the mean ASR to 45%, reaching 68.9% for individual models. We further evaluate two payload obfuscation strategies: an invisible zero-width character-based stealth payload and a phonetic payload based on grapheme-to-phoneme transformations. Neither outperforms the plain-text injection; notably, the phonetic variant reduces mean ASR to 30%. These findings identify Markdown skill files as an underexplored attack surface for AI agents and suggest that vulnerability to configuration-based jailbreaks depends more strongly on the underlying model than on the specific payload obfuscation technique.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37405

Clearing the FoG: Training KG Agents to Retrieve the Evidence Behind Their Answers

Zongyuan Sun, Shuai Zhao, Yingjun Du, Ke Shu, Ziyue Zhu, Haoran Luo

原文摘要

Large language models (LLMs) can answer many factual questions correctly, but a correct answer is not necessarily a supported one. For an LLM agent, a knowledge graph (KG) plays two roles: as a source of answers, it supplies facts that the model lacks; as a source of evidence, it supplies the supporting triples that make an answer traceable and auditable. However, existing KG agents treat the KG only as a source of answers: their training data carries answers but not evidence, and their training rewards check the answer but not the retrieved evidence, so agents often answer correctly without retrieving the supporting triples. We propose FoG, a framework that trains KG agents to retrieve the evidence behind their answers. FoG consists of evidence-grounded synthesis, a KG interaction agent, and an evidence-coverage process reward (ECPR): the synthesis recursively hides explicit entities behind verifiable sub-questions, so that every question carries its supporting triples by construction; the agent records the triples returned by the KG; and ECPR credits each supporting triple to the turn that first retrieves it. Experiments on three KGQA benchmarks demonstrate that FoG enables 3B--4B open-source agents to outperform substantially larger baselines in answer accuracy and evidence coverage, with zero-shot cross-KG transfer.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37393

Learning Quality-Controllable Robot Policies with Iterative Hindsight Relabeling

Sinjae Kang, Dong Kyu Shin, Jaehong Kim, Kamin Lee, Hyoeun Kim, Kaixin Wang, Li Zhao, Seonghyeon Ye 等 9 位作者

原文摘要

Robot policy evaluation often relies on a binary success signal for task completion. In real-world deployment, however, desired execution behaviors are also defined by quality attributes such as execution speed and motion smoothness, which need to be dynamically controlled according to task requirements. To address this, we introduce Quality-Controllable Policy (QCP), a robot policy that enables inference-time control over both individual quality attributes and their combinations. To extend the controllable quality range beyond initial demonstrations, we further propose an iterative hindsight relabeling loop: QCP collects online rollouts under target quality labels, adds successful trajectories to the dataset, relabels the accumulated trajectories in hindsight according to their measured quality, and fine-tunes the policy. Experiments on Robomimic and RoboEval demonstrate that QCP consistently improves both task execution quality and success rates across iterations, producing successful rollouts that exceed the quality of the initial demonstrations. Finally, we validate QCP on real-world manipulation tasks, where controlling speed increases throughput in conveyor operations and controlling transport stability improves success when carrying wine glasses, demonstrating that quality-controllable policies yield practical gains in physical deployment.

cs.RO · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37391

JEPA-Inspired Latent Field Prediction for Sparse-Pilot MIMO-OFDM Channel Reconstruction

Yuzhi Yang, Brahim Mefgouda, Hang Zou, Lina Bariah, Anis Bara, Yuhuan Lu, Hao Zhang, Merouane Debbah

原文摘要

Accurate acquisition of full channel state information (CSI) is essential for multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) wireless systems, while dense pilot-based estimation incurs substantial overhead. Directly minimizing the prediction error over raw CSI coefficients is fragile in realistic propagation environments, since small-scale fading, mobility, local scattering, and small spatial perturbations can induce large phase-sensitive variations. Nevertheless, wireless channels contain strong spatial, temporal, and frequency-domain correlations that can be more predictably represented in a latent propagation-field domain. Motivated by this observation, we propose a joint-embedding predictive architecture (JEPA)-inspired field world model, defined as a latent predictive model of the wireless propagation field, for sparse-pilot channel reconstruction. The model maps CSI observations into latent time-frequency tokens, predicts masked or future latent field states from historical multi-band context, and fuses the predicted latent prior with sparse noisy current pilots to reconstruct the full MIMO-OFDM channel. Ray-tracing experiments show that the proposed method improves normalized mean square error (NMSE) over pilot-only and direct reconstruction baselines, supporting latent field prediction as an effective prior for sparse-pilot channel acquisition.

eess.SP · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37386HF 社区推荐

Channel Selection for Mixture-of-Experts Decoding Needs No Calibration Data

Geunhyoung Kim, Sung Ju Hwang

原文摘要

Mixture-of-Experts (MoE) models are bottlenecked by loading routed-expert weights from memory during interactive decoding at small batch sizes. Because most intermediate channels in a routed expert contribute little to any single token, channel selectors predict which channels matter and read only their weights, typically keeping channels above a threshold calibrated on offline data. We show that channel selection needs no calibration data. First, a calibrated threshold does not fix how much a selector reads, because the number of channels passing it varies across inputs; Prox with thresholds calibrated for an 18% read budget actually reads only 12.9% on average across our benchmarks. Second, top-k keeps each token's k highest-scoring channels and needs no calibration data. At the calibrated budget, a selector's top-k version matches or beats its threshold version on the same score in 40 of 43 comparisons at 13% and 18% budgets, including all 23 on generation tasks. The top-k versions also avoid the repetition-loop collapse that calibrated thresholds cause on Gemma 4. Calibrating a threshold thus provides neither of the two things selectors use it for: a known read budget and a consistent accuracy gain at that budget. We furthermore introduce a reader built entirely from the model's weights, which screens channels with a ternary sketch and recovers the skipped ones with a low-rank map. Across three post-trained MoE models, our reader retains 98.1% of full-model accuracy at 17.8% of routed-expert reads, with Triton kernels running MoE blocks 1.60 to 1.90x faster than optimized vLLM at batch one and up to 2.69x at larger batches on the same three models.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37375

Agent-Conditioned Routing: Diverse Reasoning Paths Within One Frozen Model

Ananth Kalyanasundaram, Ankush Kadu, Athira Gopal, Aswanth Krishnan

原文摘要

Generating several solutions can expose alternative methods, but additional trajectories often paraphrase one dominant approach. We investigate whether the conditional capacity of a frozen sparse mixture-of-experts (MoE) model can provide such diversity from one problem prompt, without user-assigned roles, prescribed methods, fine-tuning, or model copies. Agent-Conditioned Routing (ACR) assigns $m$ parallel trajectories persistent, untrained shared/private expert-eligibility masks while preserving the learned, input-dependent token router. On a screened 126-question multi-mode mathematics and physics benchmark, with model judges reading the reported derivations in full, we count the number of problems yielding at least two distinct primary methods among soft-correct solutions in four trajectories. ACR reaches approximately twice as many such problems as the strongest temperature setting under GLM 5.3 (37.3 versus 18.7), and 1.67 times as many under GPT-5.6 Sol (45.00 versus 27.00), at matched output budgets. A blinded four-expert audit on a stratified sample of 30 problems also finds more multi-method solutions with ACR on both backbones, with lower correctness. Matched-eligibility ablations show modest multi-method count advantages for ACR, while token-resampled masks attain higher correctness. An informed sequential prompt elicits more diversity still, using previous solutions and explicit requests for alternative methods. ACR adds about 160 KiB of mask state. In a three-problem microbenchmark capped at 1,024 tokens, its latency and peak VRAM usage remain within 3% of batched temperature sampling.

cs.AI · cs.LG · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.02343

Hard Constraints, Smooth Gradients: Learning Feasible Inventory Policies via Differentiable Projection

Patrick Helm, Jan-Niklas Doerr, Joren Gijsbrechts, Stefan Minner

原文摘要

Many operational problems are constrained sequential decision processes with combinatorial action spaces and interdependent constraints. Mixed-integer linear programs (MILPs) handle such constraints flexibly but scale poorly in stochastic environments. Deep reinforcement learning (DRL) promises scalable decision rules, but existing methods either penalize constraints rather than enforce them, or rely on feasibility mechanisms that break down once constraints interact. We bridge this gap by embedding a differentiable optimization module inside the policy: a neural network proposes continuous action targets, a quadratic program projects them onto the relaxed feasible set, and a dual-informed integer mapping restores integrality while preserving feasibility. Given a differentiable simulator, the policy trains end-to-end from sampled trajectories using pathwise gradients, while handling hard constraints with similar flexibility to MILPs. We show that our approach has bounded error relative to an exact integer projection and ensures the entire feasible action space is reachable. We apply the method to multi-echelon production-inventory planning under shared resource and material constraints. Our policy attains an average optimality gap of 1% on 3-item instances and consistently outperforms a rolling-horizon multi-stage stochastic program by at least 5.6% in 10-item networks. Compared to state-of-the-art echelon base-stock policies, savings reach 15.3% on 40-item instances. On an industry-scale case study with nonstationary demand from ASML, we reduce average cost by up to 1.8% relative to the best-known benchmark policy. The savings are largest where planning is hardest: in tightly capacitated systems with high demand variability. More broadly, we show that DRL can improve upon state-of-the-art baselines in sequential decision problems with hard constraints, which are widespread in practice.

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2508.20729

Re4: Scientific Computing Agent with Rewriting, Resolution, Review and Revision

Ao Cheng, Lei Zhang, Guowei He

原文摘要

Large language models (LLMs) have demonstrated abilities to perform complex tasks in multiple domains, including mathematical and scientific reasoning. In this work, we construct a novel agent framework for solving representative problems in scientific computing. The proposed agent incorporates a "rewriting-resolution-review-revision" logical chain via three reasoning LLMs (the Consultant, Reviewer, and Programmer) in a collaborative and interactive manner. The Consultant module enables knowledge transfer to link problems to professional domain insights, thereby rewriting problem descriptions through text augmentation. The Programmer module generates and executes well-structured code to solve problems. The Reviewer module enables self-debugging and self-refinement through interactive feedback with code runtime outputs. Through the end-to-end review mechanism, the Programmer's executable code undergoes iterative revision. We comprehensively evaluate the framework on typical partial differential equations (PDEs), engineering-oriented flow past a cylinder, ill-conditioned linear systems, and data-driven physical analysis problems. Compared with a single-model approach, this collaborative framework significantly improves the bug-free code generation rate and reduces non-physical solutions, establishing a reliable framework for autonomous code generation from natural language descriptions. The review mechanism improved the average execution success (bug-free code and non-NaN solutions) rate of DeepSeek R1 from 59% to 82%, while ChatGPT 4.1-mini improved from 66% to 87% and Gemini-2.5 improved from 60% to 84%. Automatic code generation and review provide a promising workflow for scientific computing; analysis of solver-development trajectories further reveals recurring implementation difficulties in boundary treatment, array indexing, and the preservation of numerical constraints.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37342

Aligning Institutional Power and Operational Ability

Andrea Marino, Giovanni Sileno, Thomas van Binsbergen, Tom van Engers

原文摘要

To be socially sustainable, Agentic AI systems require normative designs that align responsibilities with the ability to fulfil them. Identifying incongruences such as unsatisfiable expectations or privilege escalation is crucial in avoiding normative pitfalls. In this study, we discuss design dimensions suitable for describing multi-agent systems in light of this norm-ability alignment requirement. We propose a formal framework centred on process-oriented modelling and institutional power for the analysis of a given normatively relevant process, discuss strategies for repairing it, and elaborate on how the symmetrical relation between operational process and institutional state affects the lifecycle of normative positions.

cs.AI · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2601.13401

SQuID: A Benchmark for Quantitative Spatial Reasoning on Satellite Imagery

Peter A. Massih, Eric Cosatto, Martin Renqiang Min

原文摘要

Vision-Language Models (VLMs) describe satellite images fluently but struggle with the quantitative spatial reasoning that remote sensing (RS) analysts perform routinely, such as counting buildings, measuring forest area in hectares and computing distances in meters. We introduce SQuID (Satellite Quantitative Intelligence Dataset), a benchmark of 2,000 questions over 587 images from four sources at 0.3 to 1.0 m ground sampling distance, covering 24 question types in three difficulty tiers. Answers are computed from curated segmentation masks or, for 50 questions, from human consensus, and numeric answers are scored against tolerance ranges informed by human annotations. SQuID also releases the reference masks behind its answers, so that modular systems can be evaluated using reference rather than predicted masks, and 75,687 training question-answer pairs on disjoint images (SQuID-Train). Among 15 evaluated systems, the best zero-shot end-to-end VLM reaches 28.1%, and RS-specialized VLMs score lower. As a baseline, we propose QASR (Quantitative Agentic Spatial Reasoning), in which an LLM writes a program that segments the image and measures the masks with geometric operators. QASR reaches 42.0%. On the 1,950 mask-derived questions, running its programs on the reference masks instead of predicted masks raises accuracy from 42.2% to 82.6%, so segmentation is its main source of error. Fine-tuning Qwen2.5-VL-7B on SQuID-Train raises its accuracy from 16.1% to 40.7%, close to QASR. Multi-condition questions (Tier 3) remain the hardest in every setting, at 62.0% for QASR with reference masks and 13.2% after fine-tuning. These results leave clear room for end-to-end models that learn metric spatial reasoning, and SQuID provides the data to train and evaluate them.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v5arXiv:2603.18740

When AI Agrees Too Much: Manipulating Automated Code Review via Bias Injection

Georgios Alexopoulos, Nikolaos Alexopoulos, Diomidis Spinellis, Dimitris Mitropoulos

原文摘要

Automated Code Review (ACR) systems integrating Large Language Models (LLMs) are increasingly adopted in software development workflows, ranging from interactive assistants to autonomous agents in CI/CD pipelines. In this paper, we study how LLM-based vulnerability detection in ACR is affected by the framing effect: the tendency to let the presentation of information override its semantic content in forming judgments. We examine whether adversaries can exploit this through contextual-bias injection (crafting PR metadata to bias ACR security judgments) as a supply-chain attack vector against real-world ACR pipelines. To this end, we first conduct a large-scale exploratory study across 6 LLMs under five framing conditions, establishing the framing effect as a systematic and widespread phenomenon in LLM-based vulnerability detection. We then design a realistic and controlled experimental environment, evaluating 33 CVEs across 20 real-world projects and two popular ACR pipelines (Claude Code and CodeRabbit), to assess the susceptibility of real-world ACR pipelines to vulnerability re-introduction attacks. We employ two attack strategies: a template-based attack inspired by prior related work, and a novel LLM-assisted refinement attack. We find that template-based attacks are ineffective and may even backfire, as direct biasing attempts raise suspicions. Our refinement attack, on the other hand, is successful in 32/33 (97%) cases, exploiting a fundamental asymmetry: attackers can iteratively refine attacks against a local clone of the review pipeline, while defenders have only one chance to detect them. Overall, our findings highlight the dangers of over-relying on ACR and stress the importance of human oversight and contributor trust in the development process.

cs.SE · cs.AI · cs.CR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.37336

CoS-VQ: Coordinated Shared-Scale Vector Quantization for Large Language Models

Zukang Xu, Zhixiong Zhao, Xing Hu, Zhe Jiang, Xin Si, Dawei Yang

原文摘要

Large language models (LLMs) require substantial memory, motivating low-bit weight quantization for efficient deployment. Existing vector quantization (VQ) methods often improve accuracy by increasing the effective coding dimension, which can complicate independent lookup and local codeword updates. We propose CoS-VQ (Coordinated Shared-Scale Vector Quantization), which preserves low-dimensional codewords while jointly optimizing hard indices and their shared scale across positions. The key challenge is scale-induced coupling: the optimal scale depends on the entire codeword combination and, in turn, changes candidate rankings across positions, so pointwise updates can stall even after scale re-optimization. CoS-VQ collects candidates at multiple probe scales and constructs a tangent separable upper bound for the residual-aware quadratic objective. By reducing candidate comparisons to lower envelopes of lines over positive scales, it identifies finitely many representative intervals for joint index--scale updates without exhaustively enumerating codeword combinations. On Qwen3.8-27B, CoS-VQ outperforms QTIP and LLVQ on the evaluated language modeling, coding, and reasoning benchmarks. Ablations from the same initialization further show larger gains from joint updates than from scale-only refinement, highlighting joint selection as an alternative to increasing codeword dimensionality. The code is available at https://github.com/xzktx003/Cos_VQ.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗