大模型推理能力 - arXivDaily 专题

2409.11538 2025-03-28 cs.CL 83%

Chain-of-Thought Prompting for Speech Translation

Ke Hu, Zhehuai Chen, Chao-Han Huck Yang, Piotr Żelasko, Oleksii Hrinchuk, Vitaly Lavrukhin, Jagadeesh Balam, Boris Ginsburg

专题命中其他推理：chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

2501.11273 2025-01-22 cs.CL 83%

Multi-round, Chain-of-thought Post-editing for Unfaithful Summaries

Yi-Hui Lee, Xiangci Li, Jessica Ouyang

专题命中其他推理：chain-of-thought(title,abstract)；reasoning(abstract)；分类 cs.CL

2410.16130 2025-01-03 eess.AS cs.CL cs.SD 83%

Can Large Audio-Language Models Truly Hear? Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoning

Chun-Yi Kuan, Hung-yi Lee

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL

Comments Accepted to ICASSP 2025. Project Website: https://github.com/kuan2jiu99/audio-hallucination

2412.01604 2024-12-17 cs.AI cs.AR 83%

Agentic-HLS: An agentic reasoning based high-level synthesis system using large language models (AI for EDA workshop 2024)

Ali Emre Oztas, Mahdi Jelodari

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.AI

Comments AI4EDA co-located with 38th Conference on Neural Information Processing Systems (NeurIPS 2024)

2412.03944 2024-12-06 cs.AI 83%

Chain-of-Thought in Large Language Models: Decoding, Projection, and Activation

Hao Yang, Qianghua Zhao, Lei Li

专题命中其他推理：chain-of-thought(title,abstract)；reasoning(abstract)；分类 cs.AI

2409.11376 2024-12-05 cs.LG 83%

Towards Time Series Reasoning with LLMs

Winnie Chow, Lauren Gardiner, Haraldur T. Hallgrímsson, Maxwell A. Xu, Shirley You Ren

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.LG

Comments Oral Presentation at 2024 NeurIPS Workshop on Time Series in the Age of Large Models

2411.12591 2024-11-20 cs.CV cs.AI 83%

Thinking Before Looking: Improving Multimodal LLM Reasoning via Mitigating Visual Hallucination

Haojie Zheng, Tianyang Xu, Hanchi Sun, Shu Pu, Ruoxi Chen, Lichao Sun

专题命中其他推理：reasoning(title,abstract)；CoT(abstract)；分类 cs.AI

2304.03087 2024-10-18 cs.CL 83%

Investigating Chain-of-thought with ChatGPT for Stance Detection on Social Media

Bowen Zhang, Xianghua Fu, Daijun Ding, Hu Huang, Genan Dai, Nan Yin, Yangyang Li, Liwen Jing

专题命中其他推理：chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

Comments arXiv admin note: text overlap with arXiv:2212.14548

2410.11588 2024-10-16 cs.CL 83%

Causal Reasoning in Large Language Models: A Knowledge Graph Approach

Yejin Kim, Eojin Kang, Juae Kim, H. Howie Huang

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL

Comments Accepted at NeurIPS 2024 Workshop on Causality and Large Models (CaLM)

2409.02519 2024-09-05 cs.CL cs.SI 83%

Language is Scary when Over-Analyzed: Unpacking Implied Misogynistic Reasoning with Argumentation Theory-Driven Prompts

Arianna Muti, Federico Ruggeri, Khalid Al-Khatib, Alberto Barrón-Cedeño, Tommaso Caselli

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL

2408.14053 2024-08-28 cs.CL 83%

Enhancing Depression Diagnosis with Chain-of-Thought Prompting

Elysia Shi, Adithri Manda, London Chowdhury, Runeema Arun, Kevin Zhu, Michael Lam

专题命中其他推理：chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

2407.17636 2024-07-26 cs.CL 83%

IgnitionInnovators at "Discharge Me!": Chain-of-Thought Instruction Finetuning Large Language Models for Discharge Summaries

An Quang Tang, Xiuzhen Zhang, Minh Ngoc Dinh

专题命中其他推理：chain-of-thought(title)；reasoning(abstract)；CoT(abstract)；分类 cs.CL

Comments Accepted by BioNLP2024 Workshop

2304.01904 2024-02-06 cs.CL 83%

REFINER: Reasoning Feedback on Intermediate Representations

Debjit Paul, Mete Ismayilzada, Maxime Peyrard, Beatriz Borges, Antoine Bosselut, Robert West, Boi Faltings

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL

Comments Accepted at EACL 2024

2401.04398 2024-01-22 cs.CL 83%

Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding

Zilong Wang, Hao Zhang, Chun-Liang Li, Julian Martin Eisenschlos, Vincent Perot, Zifeng Wang, Lesly Miculicich, Yasuhisa Fujii, Jingbo Shang, Chen-Yu Lee, Tomas Pfister

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL

Comments Accepted to ICLR 2024

2306.14122 2023-08-24 cs.CL cs.CV 83%

Chain-of-Thought Prompt Distillation for Multimodal Named Entity Recognition and Multimodal Relation Extraction

Feng Chen, Yujian Feng

专题命中其他推理：chain-of-thought(title)；reasoning(abstract)；CoT(abstract)；分类 cs.CL

Comments modification

2605.28919 2026-05-29 cs.LG cs.AI cs.CL 83%

CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models

CosmicFish-HRM：紧凑语言模型中基于层次循环机制的适应性推理

Venkat Akhil Lakkapragada

机构 * Mistyoz AI Hyderabad, India（Mistyoz AI 德里, 印度）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结提出一种紧凑语言模型CosmicFish-HRM，通过层次推理模块动态分配推理深度，在保持较小参数量的同时实现适应性推理。

Comments 17 pages, 4 figures. Exploratory study of adaptive reasoning depth in compact autoregressive language models. Code available at https://github.com/MistyozAI/CosmicFish-HRM

2306.01009 2023-06-05 cs.CL cs.AI cs.LG 83%

Examining the Emergence of Deductive Reasoning in Generative Language Models

Peter Belcak, Luca A. Lanzendörfer, Roger Wattenhofer

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted to the 1st Natural Language Reasoning and Structured Explanations Workshop (NLRSE@ACL'23). 8 pages, 4 figures, 3 tables

2607.26448 2026-07-30 cs.CL cs.AI 新提交 82%

Mergeable Model-Side Aggregation States for Long-Context Language Models

面向长上下文语言模型的可合并模型侧聚合状态

Dachuan Song, Junyu Yin, Zechen Hu, Xuan Wang

专题命中其他推理：CoT(abstract,abstract_cn)；reasoning(abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI

AI总结针对长上下文语言模型在非加性集合聚合任务中的性能缺陷，提出带HLL草图状态的模型侧聚合接口，该接口可合并跨段状态，在多任务上较基线方法实现显著性能提升。

2506.12349 2026-07-21 cs.CY cs.AI cs.CL 82%

Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek

大语言模型中的信息压制：对DeepSeek的信息审查、量化与特征化

Peiran Qiu, Siyi Zhou, Emilio Ferrara

机构 * Thomas Lord Department of Computer Science, University of Southern California, USA（汤姆斯·劳德计算机科学系，南加州大学）； Information Sciences Institute, University of Southern California, USA（信息科学研究所，南加州大学）； Annenberg School of Communication, University of Southern California, USA（安纳伯格传播学院，南加州大学）

专题命中其他推理：CoT(abstract,abstract_cn)；reasoning(abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI

AI总结本研究通过审计框架揭示DeepSeek在处理政治敏感提示时的信息压制现象，指出模型在内部推理中保留敏感内容，但在最终输出中进行过滤或改写，强调了对AI模型中信息压制机制进行系统审查的重要性。

Journal ref Inf. Sci. 724, C (Jan 2026)

2601.22954 2026-06-15 cs.CL cs.AI 版本更新 82%

Residual Context Diffusion Language Models

残差上下文扩散语言模型

Yuezhou Hu, Harman Singh, Monishwaran Maheswaran, Haocheng Xi, Coleman Hooper, Jintao Zhang, Aditya Tomar, Michael W. Mahoney, Sewon Min, Mehrdad Farajtabar, Kurt Keutzer, Amir Gholami, Chenfeng Xu

机构 * University of California, Berkeley（加州大学伯克利分校）

专题命中其他推理：CoT(summary_cn,abstract)；reasoning(abstract)；分类 cs.CL、cs.AI

AI总结提出残差上下文扩散（RCD）模块，通过回收丢弃令牌的上下文残差提高扩散语言模型的解码效率，在长/短CoT任务上以极少额外计算提升准确率4-11个百分点。

2403.04780 2026-05-26 cs.CL cs.AI 82%

Graph-oriented Instruction Tuning of Large Language Models for Generic Graph Mining

面向通用图挖掘的大语言模型图导向指令微调

Yanchao Tan, Hang Lv, Pengxiang Zhan, Shiping Wang, Carl Yang

机构 * Engineering Research Center of Big Data Intelligence, Ministry of Education（教育部大数据智能工程研究中心）； Fujian Key Laboratory of Network Computing and Intelligent Information Processing（福建省网络计算与智能信息处理重点实验室）； College of Computer and Data Science, Fuzhou University（福州大学计算机与数据科学学院）； Department of Computer Science, Emory University（埃默里大学计算机科学系）

专题命中其他推理：CoT(abstract,abstract_cn)；reasoning(abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI

AI总结提出MuseGraph框架，通过紧凑图描述、基于思维链的指令生成和图感知指令微调，将GNN与LLM结合，实现跨任务和数据集的高效图挖掘。

Comments Accepted by TPAMI 2025

Journal ref IEEE Trans. Pattern Anal. Mach. Intell., vol. 48, no. 1, pp. 155-169, Jan. 2026

2608.13787 2026-08-17 cs.AI cs.CL cs.LG cs.MA 新提交 82%

From Passive Delegates to Strategic Negotiators: Reinforcing Social Reasoning in Small Language Models with SocialRL

从被动代理到策略谈判者：用SocialRL强化小型语言模型的社会推理能力

Wenyue Hua, Zachary Huang, Tyler Payne, Safoora Yousefi, Saleema Amershi, Asli Celikyilmaz

机构 * Microsoft Research（微软研究院）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结该研究提出SocialRL方法训练4B语言模型的社会推理能力，经多领域实验，其跨领域整合后模型平均效用达0.627，优于多数GPT系列模型，心智理论蒸馏可提升效用与泛化性。

Comments 25 pages, 3 figures

2606.28186 2026-08-11 cs.CL cs.AI cs.CY cs.LG 版本更新 82%

Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

LLM推理轨迹中的认知片段实现可解释的人类项目难度预测

Chenguang Wang, Ming Li, Xinyue Zeng, Zhuochun Li, Hong Jiao, Tianyi Zhou, Dawei Zhou

机构 * Virginia Tech（弗吉尼亚理工大学）； University of Maryland（马里兰大学）； MBZUAI（穆桑大学人工智能研究所）； University of Pittsburgh（匹兹堡大学）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结提出Epi2Diff框架，将大型推理模型的推理轨迹映射为认知片段序列，通过推理规模、努力分配和状态转换建模项目难度，在真实数据集上优于基线方法。

Comments 32 pages, 8 figures, 10 tables

2602.05547 2026-08-06 cs.CL cs.AI cs.LG 版本更新 82%

Multi-Task GRPO: Reliable LLM Reasoning Across Tasks

多任务GRPO：跨任务的可靠大语言模型推理

Shyam Sundhar Ramesh, Xiaotong Ji, Matthieu Zimmer, Sangwoong Yoon, Zhiyong Wang, Haitham Bou Ammar, Aurelien Lucchi, Ilija Bogunovic

机构 * UCL Department of EEE（伦敦大学学院电子工程系）； UCL Centre for AI（伦敦大学学院人工智能中心）； Huawei Noah’s Ark Lab（华为诺亚实验室）； UNIST Graduate School of AI（延世大学人工智能研究生院）； University of Edinburgh（爱丁堡大学）； University of Basel（巴塞尔大学）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结本文提出MT-GRPO算法，通过动态调整任务权重和比例保持采样器，提升多任务场景下大语言模型的可靠推理性能。

Comments Accepted at ICML 2026

2607.19226 2026-07-22 cs.CL cs.AI cs.LG 新提交 82%

The Price of Reasoning: Cost-Quality Tradeoffs in Reinforcement Learning for Neural Machine Translation

推理的代价：神经机器翻译强化学习中的成本-质量权衡

Michael Jungo, Aixiu An

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究神经机器翻译强化学习中推理痕迹对翻译质量的影响，通过在训练或推理阶段省略推理痕迹进行实验，发现推理尤其是推理阶段能提升质量，还研究了推理导致计算需求增加与翻译质量提升间的成本-质量权衡。

Journal ref Proceedings of the 2026 ACM Symposium on Document Engineering (DocEng '26)

2607.19181 2026-07-22 cs.CL cs.AI cs.LG 新提交 82%

Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

翻译前推理：用结构化推理增强法律机器翻译

Aixiu An, Michael Jungo, Eloi Eynard, Mark Drenhaus, Andreas Fischer, Jean Hennebert, Sébastien Rumley

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究法律领域神经机器翻译难题，通过比较多种方法，评估小型语言模型在不同再训练范式下的表现，以瑞士法律系统为测试平台，发现强化学习效果好，增强小型模型接近前沿推理模型，再训练范式随模型规模收益递减。

Comments Code available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL

2512.13478 2026-07-16 cs.CL cs.AI cs.LG 82%

NRR-Core: Non-Resolution Reasoning as a Computational Framework for Contextual Identity and Ambiguity Preservation

NRR-Core：非解析推理作为情境身份和歧义保留的计算框架

Kei Saito

机构 * Independent Researcher, Japan（日本独立研究员）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结本文提出NRR框架，通过多向量嵌入、非坍缩注意力和情境身份追踪，解决人工智能系统过早解析歧义的问题，保持歧义保留以提升推理灵活性。

Comments 14 pages, 2 figures, 2 tables. Replacement synced to the current GitHub repository snapshot. Series hub: https://github.com/kei-saito-research/nrr-series-hub

2607.10836 2026-07-14 cs.AI cs.CL cs.LG 新提交 82%

Route, Communicate, and Reason: Gated Routing and Adaptive Depth for Efficient Multi-Agent Reasoning

路由、通信与推理：用于高效多智能体推理的门控路由与自适应深度

Sudipto Ghosh, Tanmoy Chakraborty

机构 * Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi（印度理工学院德里分校雅迪人工智能学院）； Department of Electrical Engineering, Indian Institute of Technology Delhi（印度理工学院德里分校电气工程系）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究针对多智能体推理中未解决的问题，提出GRADE分层多智能体系统，用四个轻量级学习门控及CoGRPO训练方法，智能体模型可热插拔。该系统在多个任务上优于基线，消融实验明确关键因素，证明校准对热插拔必要。

2607.07690 2026-07-09 cs.LG cs.AI cs.CL 新提交 82%

Agon: Competitive Cross-Model RL with Implicit Rival Grading of Reasoning

Agon：具有隐式推理对手评分的竞争性跨模型强化学习

Vladislav Beliaev

机构 * Independent Researcher（独立研究者）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究针对可验证奖励强化学习只评最终答案的问题，提出Agon方法，让两个竞争模型相互评分，通过轮流扮演角色隐式判断推理，在难题上提升了模型表现，且该排序在多种场景和模型家族中可复制。

Comments 15 pages, 7 figures, 8 tables

2604.21495 2026-07-01 cs.LG cs.AI cs.CL 版本更新 82%

Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning

通过操作草图和自监督学习推广表格数据中的数值推理

Hanjun Cho, Gahyun Yoo, Hanseong Kim, Jay-Yoon Lee

机构 * Seoul National University（首尔国立大学）； Soongsil University（顺天大学）

专题命中其他推理：reasoning(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结本文提出TaNOS框架，通过操作草图和自监督学习提升表格数据中数值推理的泛化能力，实验显示其在FinQA数据集上表现优异，且在领域转移中鲁棒性更强。

Comments Accepted to TACL. This is a pre-MIT Press publication version

1. 其他推理 5768 篇

Chain-of-Thought Prompting for Speech Translation

Multi-round, Chain-of-thought Post-editing for Unfaithful Summaries

Can Large Audio-Language Models Truly Hear? Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoning

Agentic-HLS: An agentic reasoning based high-level synthesis system using large language models (AI for EDA workshop 2024)

Chain-of-Thought in Large Language Models: Decoding, Projection, and Activation

Towards Time Series Reasoning with LLMs

Thinking Before Looking: Improving Multimodal LLM Reasoning via Mitigating Visual Hallucination

Investigating Chain-of-thought with ChatGPT for Stance Detection on Social Media

Causal Reasoning in Large Language Models: A Knowledge Graph Approach

Language is Scary when Over-Analyzed: Unpacking Implied Misogynistic Reasoning with Argumentation Theory-Driven Prompts

Enhancing Depression Diagnosis with Chain-of-Thought Prompting

IgnitionInnovators at "Discharge Me!": Chain-of-Thought Instruction Finetuning Large Language Models for Discharge Summaries

REFINER: Reasoning Feedback on Intermediate Representations

Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding

Chain-of-Thought Prompt Distillation for Multimodal Named Entity Recognition and Multimodal Relation Extraction

CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models

Examining the Emergence of Deductive Reasoning in Generative Language Models

Mergeable Model-Side Aggregation States for Long-Context Language Models

Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek

Residual Context Diffusion Language Models

Graph-oriented Instruction Tuning of Large Language Models for Generic Graph Mining

From Passive Delegates to Strategic Negotiators: Reinforcing Social Reasoning in Small Language Models with SocialRL

Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

Multi-Task GRPO: Reliable LLM Reasoning Across Tasks

The Price of Reasoning: Cost-Quality Tradeoffs in Reinforcement Learning for Neural Machine Translation

Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

NRR-Core: Non-Resolution Reasoning as a Computational Framework for Contextual Identity and Ambiguity Preservation

Route, Communicate, and Reason: Gated Routing and Adaptive Depth for Efficient Multi-Agent Reasoning

Agon: Competitive Cross-Model RL with Implicit Rival Grading of Reasoning

Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning