让下一篇重要论文，不再错过。

版本更新 · v2arXiv:2607.09153

更新于 2026-11-04

KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling

Peng Kuang, Haibo Jin, Xiaoyu Han, Yanli Wang, Xiaopeng Yuan, Ye Yu, Kaidi Xu, Haohan Wang

原文摘要

Process Reward Models (PRMs) have been proven to be highly effective in guiding test-time scaling (TTS) methods, which significantly boost the capabilities of LLM-based multi-agent systems. However, existing PRMs are text-based: they re-encode the entire trajectory text from scratch. In long multi-agent rollouts, the scoring cost, growing quadratically with respect to sequence length L, creates a severe computational bottleneck, severely limiting PRMs' application in long-context scenarios. To resolve this, we introduce KV-PRM, a highly efficient process reward model that eliminates the heavy text re-encoding by directly reading the KV cache produced naturally during the LLM's generation phase. By processing a single "verify token" against the pre-existing KV cache, KV-PRM reduces the scoring cost from O(L^2) to O(L). We formally prove that the KV cache contains strictly greater information capacity than text, and is more efficient for downstream reward modeling. Empirically, across the MATH, GSM8K, and AIME benchmarks, KV-PRM matches or strictly outperforms text-PRMs under various TTS methods such as Beam Search, MCTS, and Weighted Voting, with up to a 5,000x reduction in scoring FLOPs, a 37x reduction in latency, and a 34x reduction in per-sequence memory footprint compared to text-based PRMs.

cs.AI

新论文 · v1arXiv:2611.05231

更新于 2026-11-04

组合诊断：通过组合已学放射学征象检测和定位未见疾病

Diagnosis by Composition: Detecting and Localizing Unseen Diseases by Composing Learned Radiological Findings

Kaihua Hou, Wei Liu, Yi Huang, Hao Zhang, Haibo Jin, Yifei Sun, Yuan gao, Zuoou Li 等 13 位作者

AI 速读基于摘要

提出诊断即组合（DbC）框架，通过学习放射学征象词汇并依据指南组合征象图来诊断和定位未见疾病，在肝脏CT上优于对比方法。

cs.CV

查看原文摘要

Deep models for radiological diagnosis are typically trained to predict disease labels directly from images, which limits them to diseases seen during training and makes their decisions hard to interpret in radiologists' terms. Radiology, however, proceeds from visual findings to disease through published guidelines that make this decision rule explicit, interpretable, and reproducible. We introduce diagnosis by composition (DbC), a representation learning framework that encodes this findings-first logic. DbC learns a vocabulary of radiological findings by aligning image features with the text embeddings of the findings, each finding represented as a dense map over the scan. A disease is defined by a set of required findings derived from its guidelines, and its diagnosis map is a fixed, parameter-free composition of the corresponding finding maps. Because the finding-to-disease mapping is derived from guidelines rather than learned, DbC is interpretable and auditable by design, can extend to new diseases by specifying their findings without retraining, and can diagnose unseen diseases by composing their findings. On multi-phase liver CT, DbC detects and localizes unseen diseases, rare diseases from a different institution, and an external cohort significantly better than all comparators, including models that align images with disease names. DbC also improves detection and localization in the fully supervised setting and remains superior across training-set sizes. Although demonstrated on liver CT, DbC applies wherever radiological guidelines define diseases through findings.

版本更新 · v2arXiv:2610.03335

更新于 2026-11-02

Randomization Beyond Deterministic Adaptivity in Parallel Sampling

Da Li, Bo Jin, Xiaopeng Wei

原文摘要

Parallel samplers, including masked diffusion models, generate several coordinates per round from conditional marginals, and their accuracy depends on which coordinates are revealed together. A schedule can be improved in two natural ways: adaptively, by choosing later batches from the values already generated, or by randomization, by drawing the schedule at random. We show that randomization is strictly more powerful. For every $r \geq 1$ we construct a full-support distribution on $9r$ binary coordinates such that every deterministic three-round sampler, even with value-dependent batches and arbitrary product kernels, has forward Kullback-Leibler divergence at least $γr - \log 4$ with $γ\approx 0.0966$, whereas drawing one of four fixed schedules with two fair bits drives the divergence below $18 r β^r$ with $β\approx 0.966$. Both samplers query conditionals of the same target. The separation holds for every fixed number of rounds $T \geq 2$, a strict advantage survives perturbations of the target and errors in the conditional predictor, and exact rational arithmetic certifies the gap at nine coordinates, where trained neural predictors preserve it. The mechanism is a hidden mixture: each schedule preserves the dependence pattern of one component, while no deterministic first batch can serve them all.

cs.IT

版本更新 · v2arXiv:2606.06252

更新于 2026-11-02

Can Test-Time Reconstruction Close the Loop on Latent Reasoning?

Xiaopeng Yuan, Haibo Jin, Ye Yu, Peng Kuang, Lijun Yu, Yushun Dong, Haohan Wang

原文摘要

Recent work moves intermediate reasoning from natural-language traces into latent or cache-level representations to reduce token overhead and avoid a discrete communication bottleneck. However, this shift also removes a key advantage of textual reasoning: intermediate states are no longer inspectable, making it difficult to determine whether a latent state still preserves the constraints of the original query. As a result, latent reasoning typically operates in an open loop, where a latent state is produced and consumed without an input-anchored fidelity check. We propose ReLAT (Reconstruction-Guided Latent Reasoning At Test Time), a self-supervised test-time training method that closes this loop using the query itself as the reference. Our key observation is that if a latent state faithfully represents a query, the query should be recoverable from it; if the query cannot be recovered, the latent state has lost task-relevant information. ReLAT operationalizes this principle by constructing a differentiable Question -> Latent Thought -> Question cycle and optimizing query reconstruction loss through the latent thought before answer generation. This anchors opaque latent computation to the problem specification it is supposed to represent. Across mathematical reasoning, knowledge QA, and code generation benchmarks on the Qwen family, ReLAT consistently improves over single-model inference, text-based collaboration, open-loop latent collaboration, and alternative test-time training objectives. On Qwen3-8B, ReLAT raises AIME 2024 accuracy from 56.7% to 73.3%, a 16.6-point gain over the strongest open-loop latent baseline.

cs.AI

版本更新 · v2arXiv:2610.09401

更新于 2026-11-02

Can the Library Speak? Self-Advertised Method Selection for Formal Proving

Xiaopeng Yuan, Suijin Wang, Yanli Wang, Haibo Jin, Peng Kuang, Jerry Wang, Lijun Yu, Haohan Wang

原文摘要

LLM-based formal provers can retrieve relevant lemmas and prior proofs, but relevance alone does not say whether a mathematical method can be used on the current theorem. A method has prerequisites, a target, an intended action, and obligations that its use leaves to prove. Methods that look equally related to a theorem may therefore differ substantially in whether they offer a plausible next step. We formulate this as an applicability-aware method-selection problem and introduce self-advertisement: before candidates are ranked, a model generates a problem-specific proposal for each one, stating what part of the goal it targets, what action it would take, and what conditions that action requires. We organize 82 reusable methods from Putnam 2000-2014 as Method Contracts, which pair applicability descriptions with Mathlib anchors, a checked example or scaffold, and expected proof obligations. A single batched call elicits proposals across the library; vague or unsupported proposals are demoted, yielding a ranked shortlist accompanied by inspectable claims about each candidate's use. We analyze when similarity-based representations cannot distinguish methods with different applicability, how errors in applicability estimates affect shortlist quality, and what a checked scaffold guarantees under its stated assumptions. Against lexical, embedding, and embedding-plus-LLM reranking baselines, self-advertisement achieves 95.0% hit@5 on Putnam 2015-2025, compared with 84.2% for the strongest reranker. On IMO ProofBench, it achieves 91.7% compared with 88.3%. These results indicate improved coverage of annotated methods in the retrieved shortlists, particularly on Putnam.

cs.AI

新论文 · v1arXiv:2610.40276

更新于 2026-10-29

TaskFrame: Connecting Task-Level Planning with Frame-Level Control for Game Agents

Guangzhi Wang, Junfu Pu, Yiran Wang, Xingyilang Yin, Honghao Chen, Cheng Cheng, Lingen Li, Yinan Zhou 等 13 位作者

原文摘要

Building general game agents requires both high-level task reasoning and low-latency action control in rapidly changing visual environments. These capabilities favor models with substantially different computational profiles, making them difficult to realize within a single model. In this work, we present TaskFrame, a hierarchical planner-controller framework that connects task-level planning with frame-level keyboard-and-mouse control. A vision-language task planner decomposes a high-level goal into an ordered sequence of atomic instructions and is optimized through supervised learning and dependency-aware GRPO. The hidden states of special goal tokens that terminate each atomic instruction provide continuous goal embeddings that are cached and reused across control steps. A causal Transformer controller combines the embedding of the active goal with recent observations to predict discrete keyboard and mouse-button actions and continuous camera motion at every game step. We train the system in three stages: planner training, controller pretraining, and joint alignment. Experiments in Minecraft show that TaskFrame is competitive with general-purpose vision-language models on instruction decomposition, and that in closed-loop evaluation the agent achieves a higher combat success rate than the evaluated baselines.

cs.CV · cs.AI

版本更新 · v2arXiv:2610.31066

更新于 2026-10-28

Dual Exam: Controlled Evaluation of Multimodal Agents and World Models

Zhefan Rao, Meng Chu, Xuan Billy Zhang, Jingfan Chen, Guohao Cai, Runyi Li, Shiyi Du, Xu Huang 等 92 位作者

原文摘要

World Models (WMs) can support Multimodal Agents (MAs) through simulated experience, while Agent trajectories can inform World Model refinement. Evaluating this interaction requires separate evidence about action selection and state evolution. We introduce Dual Exam, which tests Agents under fixed environment response mechanisms and World Models under fixed action interventions across physical, digital, social, and scientific domains. The Agent track compares 10 configurations on 1,088 items in 15 reporting categories. The World Model track contains 400 cases across eight scenarios; 15 configurations have full coverage and one covers six scenarios. GPT 6 Astra leads the case-weighted Agent Index at 31.3, and GPT-Image-2 leads the full-coverage WM Overall at 21.7. These scores use different tasks and metrics and are not directly comparable. A 150-case pilot of one coupled pairing records zero task successes and motivates component-level diagnosis. Results reveal uneven task performance and failures to preserve task-relevant state. Their value for predicting coupled success or sustained reciprocal learning remains to be established.

cs.AI

版本更新 · v2arXiv:2605.21948

更新于 2026-10-28

SCI-Defense: Detecting Persuasion-Structured Manipulation of LLM-Based Ranking in Generative Engines

Xucheng Yu, Haibo Jin, Huimin Zeng, Haohan Wang

原文摘要

Generative engine optimization (GEO) has reshaped how sellers compete for visibility in LLM-generated rankings. By embedding persuasive arguments in the content they control, sellers can influence how generative engines rank products and web pages and push their own items to the top. Existing defenses either detect harmful intent or unnatural language, which fluent GEO content often lacks, or rewrite all candidates, which alters legitimate listings without identifying the manipulated ones. However, these approaches overlook the persuasive structure through which GEO attacks influence LLM-based rankings. To address this gap, we present SCI-Defense, a framework for detecting persuasion-structured ranking manipulation, in which an LLM judge scores five persuasion devices grounded in research on influence and deception: authority, social proof, scarcity and urgency, comparative claims, and narrative purposiveness. Two lightweight gates for unnatural strings and inflated relevance complement the judge, and flagged candidates are demoted rather than rewritten, each with a record of the candidate and the criterion that triggered the flag. On 1,152 optimized CORE attacks across six Amazon product categories and six MS MARCO web domains, SCI-Defense detects 89.6% of reasoning attacks and 96.2% of review attacks at a 1.2% false positive rate and blocks 96.1% of the attacks that pushed their target into the top three. On a 120-query subset, paraphrasing and truncation block a similar share of attacks but push 4.0--4.5% of honest items out of the top three, compared with 0.6% for SCI-Defense.

cs.LG

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1822ACL Anthology · ACL 2026

来源日期 2026-07-01（仅到月份）

On-policy Reinforcement Fine-tuning with Offline reward for Multi-step Embodied Planning

Di Wu, Jiaxin Fan tongji university, Chloe Gu, Guanbo Wang, Wei Yin, Wenhao Li Tongji University, Bo Jin Tongji University

原文摘要

Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and verbal goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle in interactive environments. Reinforcement learning (RL) offers a natural way to address this limitation, yet online RL approaches suffer from costly interaction and sparse rewards in embodied settings. This paper introduces ORBIT, an On-policy Reinforcement fine-tuning (RFT) framework with offline rewards for EmBodIed Task Planning, that preserves the generalization benefits of RFT while addressing the challenges of costly interaction and sparse rewards, supported by solid theoretical guarantees. Our approach is evaluated on EmbodiedBench, a recent benchmark for interactive embodied tasks, covering both in-domain and out-of-domain scenarios. Experimental results show that ORBIT achieves SOTA performance on EB-ALFRED, outperforming all closed-source and online-RL-based methods, while being substantially more efficient in training speed and computational cost, remaining robust to sub-optimal expert trajectories, and exhibiting strong generalization to unseen environments. We released all code and data at https://github.com/mail-taii/Reinforced-Reasoning-for-Embodied-Planning.

cs.CL · cs.CV

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1903ACL Anthology · ACL 2026

来源日期 2026-07-01（仅到月份）

Learning to Conceal Risk: Controllable Multi-turn Red Teaming for LLMs in the Financial Domain

Gang Cheng Bloomberg, Haibo Jin University of Illinois at Urbana-Champaign, Wenbin Zhang Florida International University, Haohan Wang University of Illinois at Urbana-Champaign, Jun Zhuang Boise State University and Indiana University Purdue University Indianapolis

原文摘要

Large Language Models (LLMs) are increasingly deployed in finance, where unsafe behavior can lead to serious regulatory risks. However, most red-teaming research focuses on overtly harmful content and overlooks attacks that appear legitimate on the surface yet induce regulatory-violating responses. We address this gap by introducing a controllable black-box multi-turn risk-concealed redteaming framework (CoRT) that progressively conceals surface-level risk while exploiting regulatory-violating behaviors. CoRT contains two key components: (i) a Risk Concealment Attacker (RCA) that generates multiturn prompts via iterative refinement, and (ii) a Risk Concealment Controller (RCC) that predicts a turn-level Risk Concealment Score (RCS) to steer RCA’s follow-up style. We also build a domain-specific benchmark, FinRisk-Bench, with 522 instructions spanning six financial risk categories. Experiments on nine widely used LLMs show that CoRT (RCA) achieves 93.19% average attack success rate (ASR), and CoRT (RCA+RCC) further improves the average ASR to 95.00%. Our code and FinRisk-Bench are available at https://github.com/gcheng128/CoRT.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.2005ACL Anthology · ACL 2026

来源日期 2026-07-01（仅到月份）

SiLP: Enhancing Non-Dominant Language Capabilities with a Selective Bidirectional Language Projection Framework

Junpeng Liu, Jiuyi Li Dalian University of Technology, Kaiyu Huang Beijing Jiaotong University, Bo Jin Dalian University of Technology, Degen Huang Fuyao University of Science and Technology, Hui Xiong Hong Kong University of Science and Technology (Guangzhou)

原文摘要

Current large language models (LLMs) often exhibit performance imbalances between dominant languages (e.g., English) and non-dominant ones due to the skewed distribution of pretraining data. A common strategy to address this issue is to enhance cross-lingual alignment, thereby facilitating non-dominant language processing. However, existing methods typically rely on additional training objectives or language-specific parameters, which increase training complexity and cost. In this work, we propose a selective bidirectional language projection framework that enables efficient multilingual alignment and language shift using the intrinsic parameters. Specifically, we first identify the layers most sensitive to language projection between non-dominant and dominant languages through neuron activation analysis. We then perform sequential language projection within the selected layers by mapping non-dominant representations into the dominant language space and reverting them before generation. The bidirectional projection benefits the subsequent instruction tuning in non-dominant languages. Experiments on seven benchmarks demonstrate that our method remarkably enhances the performance of non-dominant languages. Further analyses indicate that our method learns better internal representations and exhibits strong generalization capabilities.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.953ACL Anthology · ACL 2026

来源日期 2026-07-01（仅到月份）

VideoCuRL: Video Curriculum Reinforcement Learning with Orthogonal Difficulty Decomposition

Hongbo Jin, Kuanwei Lin, Wenhao Zhang, Yichen Jin Peking University, Ge Li Peking University

原文摘要

Reinforcement Learning (RL) is crucial for empowering Video-LLMs with complex spatiotemporal reasoning. However, current RL paradigms predominantly rely on random data shuffling or naive curriculum strategies based on scalar difficulty metrics. We argue that scalar metrics fail to disentangle two orthogonal challenges in video understanding: Visual-Temporal Perception Load and Cognitive Reasoning Depth. To address this, we propose VideoCuRL, a novel framework that decomposes difficulty into these two axes. We employ efficient, training-free proxies—optical flow/keyframe entropy for visual complexity and Calibrated Surprisal for cognitive complexity—to map data onto a 2D curriculum grid. A competence-aware Diagonal Wavefront strategy then schedules training from base alignment to complex reasoning. Furthermore, we introduce Dynamic Sparse KL and Structured Revisiting to stabilize training against reward collapse and catastrophic forgetting. Extensive experiments show that VideoCuRL surpasses strong RL baselines on reasoning (+2.5% on VSI-Bench) and perception (+2.9% on VideoMME) tasks. Notably, VideoCuRL eliminates the prohibitive inference overhead of generation-based curricula, offering a scalable solution for robust video post-training.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表