大模型推理能力

Model-First Reasoning LLM Agents: Reducing Hallucinations through Explicit Problem Modeling

基于模型的推理语言模型代理：通过显式问题建模减少幻觉

Annu Rana, Gaurav Kumar

机构 * Stanford AI Professional Program（斯坦福AI专业项目） ； IESE EMBA Program（IESE EMBA项目） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；planning(abstract)；分类 cs.AI

AI总结 本文提出基于模型的推理方法，通过显式问题建模减少LLM在复杂规划任务中的幻觉和不一致性。

Intention Chain-of-Thought Prompting with Dynamic Routing for Code Generation

具有动态路由的意图链式思维提示法用于代码生成

Shen Li, Li Huang, Shaoxiong Zhan, Weifeng Sun, Tao Yin, Zhongxin Liu, Meng Yan

专题命中 逻辑推理 ：chain-of-thought(title,abstract)；reasoning(abstract)；CoT(abstract)；分类 cs.AI

AI总结 RoutingGen通过动态路由策略优化代码生成，结合少样本提示与结构化推理策略ICoT，提升生成效率与质量。

Comments Accepted at AAAI-2026

Cognitive Mirrors: Exploring the Diverse Functional Roles of Attention Heads in LLM Reasoning

认知镜像：探索大语言模型推理中注意力头的多样化功能角色

Xueqi Ma, Jun Wang, Yanbei Jiang, Sarah Monazam Erfani, Tongliang Liu, James Bailey

机构 * The University of Melbourne（墨尔本大学） ； The University of Sydney（悉尼大学） ； Amazon（亚马逊公司） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.AI

AI总结 本文提出认知镜像框架，通过CogQA数据集分析LLM中注意力头的功能专业化，揭示其在推理任务中的关键作用。

Comments Accepted to NeurIPS 2025

MiRAGE: Misconception Detection with Retrieval-Guided Multi-Stage Reasoning and Ensemble Fusion

Cuong Van Duc, Thai Tran Quoc, Minh Nguyen Dinh Tuan, Tam Vu Duc, Son Nguyen Van, Hanh Nguyen Thi

机构 * Hanoi University of Science and Technology（河内科学技术大学） ； School of Computing, Phenikaa University（现象学大学计算机学院） ； Faculty of Interdisciplinary Digital Technology, Phenikaa University（现象学大学跨学科数字技术学院） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.AI

When Thinking Drifts: Evidential Grounding for Robust Video Reasoning

Mi Luo, Zihui Xue, Alex Dimakis, Kristen Grauman

机构 * The University of Texas at Austin（德克萨斯大学奥斯汀分校） ； UC Berkeley（伯克利大学） ； Bespoke Labs（Bespoke实验室） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.AI

Comments Accepted by NeurIPS 2025, Project page: https://vision.cs.utexas.edu/projects/video-ver/

Fine-Tuning LLMs to Analyze Multiple Dimensions of Code Review: A Maximum Entropy Regulated Long Chain-of-Thought Approach

Yongda Yu, Guohao Shi, Xianwei Wu, Haochuan He, XueMing Gu, Qianqian Zhao, Kui Liu, Qiushi Wang, Zhao Tian, Haifeng Shen, Guoping Rong

机构 * Nanjing University（南京大学） ； University of Waterloo（滑铁卢大学） ； Northeastern University（东北大学） ； Huawei Technologies Co., Ltd.（华为技术有限公司） ； Southern Cross University（南方十字大学） 纠错

专题命中 逻辑推理 ：chain-of-thought(title,abstract)；reasoning(abstract)；CoT(abstract)；分类 cs.AI

Comments 22 pages

Skip-Thinking: Chunk-wise Chain-of-Thought Distillation Enable Smaller Language Models to Reason Better and Faster

Xiao Chen, Sihang Zhou, Ke Liang, Xiaoyu Sun, Xinwang Liu

机构 * National University of Defense Technology（国防科技大学） 纠错

专题命中 逻辑推理 ：chain-of-thought(title,abstract)；reasoning(abstract)；CoT(abstract)；分类 cs.CL

Enhancing Chain of Thought Prompting in Large Language Models via Reasoning Patterns

Yufeng Zhang, Xuepeng Wang, Lingxiang Wu, Jinqiao Wang

专题命中 逻辑推理 ：reasoning(title,abstract)；CoT(abstract)；logical reasoning(abstract)；分类 cs.CL

Nature's Insight: A Novel Framework and Comprehensive Analysis of Agentic Reasoning Through the Lens of Neuroscience

Zinan Liu, Haoran Li, Jingyi Lu, Gaoyuan Ma, Xu Hong, Giovanni Iacca, Arvind Kumar, Shaojun Tang, Lin Wang

机构 * School of EEE, Nanyang Technological University (NTU)（南洋理工大学电子工程系） ； School of Cyber Science and Technology, University of Science and Technology of China（中国科学技术大学网络科学与技术学院） ； School of Integrated Circuits and Electronics, Beijing Institute of Technology（北京理工大学集成电路与电子学院） ； School of Social Sciences, Nanyang Technological University（南洋理工大学社会科学学院） ； Department of Information Engineering and Computer Science, University of Trento（特伦托大学信息工程与计算机科学系） ； Division of Computational Science and Technology, KTH Royal Institute of Technology（皇家理工学院计算科学与技术系） ； Bioscience and Biomedical Engineering, Hong Kong University of Science & Technology（香港科技大学生物科学与生物医学工程系） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.LG

Comments 39 pages, 17 figures

Reasoning Capabilities and Invariability of Large Language Models

Alessandro Raganato, Rafael Peñaloza, Marco Viviani, Gabriella Pasi

机构 * Department of Informatics, Systems, and Communication (DISCo) University of Milano-Bicocca（信息学、系统与通信系（DISCo）米兰-比科卡大学） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL

Comments Accepted for publication in the Proceedings of the 23rd IEEE/WIC International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT 2024)

Logic-of-Thought: Injecting Logic into Contexts for Full Reasoning in Large Language Models

Tongxuan Liu, Wenjiang Xu, Weizhe Huang, Yuting Zeng, Jiaxing Wang, Xingyu Wang, Hailong Yang, Jing Li

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL

Comments 18 pages, Accepted to NAACL2025

"Let's Argue Both Sides": Argument Generation Can Force Small Models to Utilize Previously Inaccessible Reasoning Capabilities

Kaveh Eskandari Miandoab, Vasanth Sarathy

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL

Comments Accepted to Workshop on Customizable NLP: Progress and Challenges in Customizing NLP for a Domain, Application, Group, or Individual at EMNLP 2024

Large Language Models Can Learn Temporal Reasoning

Siheng Xiong, Ali Payani, Ramana Kompella, Faramarz Fekri

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL

Comments ACL24 (main)

Wrong-of-Thought: An Integrated Reasoning Framework with Multi-Perspective Verification and Wrong Information

Yongheng Zhang, Qiguang Chen, Jingxuan Zhou, Peng Wang, Jiasheng Si, Jin Wang, Wenpeng Lu, Libo Qin

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL

Comments Accepted by EMNLP 2024 Findings

Meta-Reasoning: Semantics-Symbol Deconstruction for Large Language Models

Yiming Wang, Zhuosheng Zhang, Pei Zhang, Baosong Yang, Rui Wang

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL

Comments Accepted by ACL 2024 Findings

ThinkSum: Probabilistic reasoning over sets using large language models

Batu Ozturkler, Nikolay Malkin, Zhen Wang, Nebojsa Jojic

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL

Comments ACL 2023

Correct Chains, Wrong Answers: Dissociating Reasoning from Output in LLM Logic

正确推理，错误答案：分离LLM中的推理与输出

Abinav Rao, Sujan Rachuri, Nikhil Vemuri

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI；logical reasoning(comments)

AI总结 本文提出新型操作测试基准，区分操作逻辑与名称，揭示LLM在深度推理中存在策略与内容两类失败模式，验证名称不决定推理能力。

Comments 9 pages, 4 figures. ICLR 2026 Workshop on Logical Reasoning of LLMs

Don't Let It Hallucinate: Premise Verification via Retrieval-Augmented Logical Reasoning

不要让它幻觉：通过检索增强的逻辑推理进行前提验证

Yuehan Qin, Shawn Li, Yi Nian, Xinyan Velocity Yu, Yue Zhao, Xuezhe Ma

机构 * University of Southern California（南加州大学） 纠错

专题命中 逻辑推理 ：reasoning(title)；logical reasoning(title)；分类 cs.CL、cs.AI

AI总结 本文提出一种基于检索增强的逻辑推理方法，用于在生成前验证用户查询中的前提，从而减少幻觉并提高事实准确性。

Comments TMLR 2026

Meaningless is better: hashing bias-inducing words in LLM prompts improves performance in logical reasoning and statistical learning

Milena Chadimová, Eduard Jurášek, Tomáš Kliegr

专题命中 逻辑推理 ：reasoning(title)；logical reasoning(title)；分类 cs.CL、cs.AI

NeuralLog: Natural Language Inference with Joint Neural and Logical Reasoning

Zeming Chen, Qiyue Gao, Lawrence S. Moss

专题命中 逻辑推理 ：reasoning(title)；logical reasoning(title)；分类 cs.CL、cs.AI

Comments 8 pages, 4 figures, The 10th Joint Conference on Lexical and Computational Semantics (*SEM2021) @ ACL2021

Reasoning Consensus: Structural Ensembling of LLM Reasoning via Weighted DAG Aggregation

推理共识：通过加权有向无环图聚合实现大语言模型推理的结构集成

Amruta Parulekar, Jinu Lee, Dilek Hakkani-Tür, Hari Sundaram

机构 * University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结 该研究提出通过加权DAG聚合集成LLM推理结构的框架，在六个基准上优于多数投票基线，可提供可检查的共识推理图，还能分析不同推理视角。

Reinforcing Video Reasoning Segmentation to Think Before It Segments

强化视频推理分割以在分割前思考

Sitong Gong, Lu Zhang, Yunzhi Zhuge, Xu Jia, Pingping Zhang, Huchuan Lu

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)

AI总结 Veason-R1通过组相对策略优化和链式思维初始化，提升视频推理分割的结构化推理能力，实现更准确的时空定位和鲁棒性。

Comments 12 pages

Journal ref CVPR 2026

DrugR: Optimizing Molecular Drugs through LLM-based Explicit Reasoning

DrugR: 通过基于大语言模型的显式推理优化分子药物

Haoran Liu, Zheni Zeng, Yukun Yan, Yuxuan Chen, Yunduo Xiao

专题命中 逻辑推理 ：reasoning(title,abstract)；logical reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结 DrugR通过基于大语言模型的显式推理方法，优化分子药物的ADMET性质，提升药理效果并提供可解释的优化依据。

Evaluating and Enhancing the Vulnerability Reasoning Capabilities of Large Language Models

评估和增强大型语言模型的漏洞推理能力

Li Lu, Yanjie Zhao, Hongzhou Rao, Kechi Zhang, Haoyu Wang

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)

AI总结 DAGVul通过建模漏洞推理为有向无环图生成任务，提升大型语言模型在漏洞检测中的推理能力，实验显示其推理F1分数平均提高18.9%。

SwimBird: Eliciting Switchable Reasoning Mode in Hybrid Autoregressive MLLMs

SwimBird: 在混合自回归大语言模型中实现可切换的推理模式

Jintao Tong, Shilin Yan, Hongwei Xue, Xiaojun Tang, Kunyu Shi, Guannan Zhang, Ruixuan Li, Yixiong Zou

机构 * Huazhong University of Science and Technology（华中科技大学） ； Accio Team, Alibaba Group（阿里集团Accio团队） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；CoT(abstract)；logical reasoning(abstract)

AI总结 SwimBird通过动态切换三种推理模式，提升多模态大语言模型在视觉密集任务中的性能，同时保持文本推理能力。

Comments Project Page: https://accio-lab.github.io/SwimBird

TiViBench: Benchmarking Think-in-Video Reasoning for Video Generative Models

TiViBench: 用于视频生成模型的视频推理基准测试

Harold Haodong Chen, Disen Lan, Wen-Jie Shu, Qingyang Liu, Zihan Wang, Sirui Chen, Wenkai Cheng, Kanghao Chen, Hongfei Zhang, Zixin Zhang, Rongjin Guo, Yu Cheng, Ying-Cong Chen

专题命中 逻辑推理 ：reasoning(title,abstract)；planning(abstract)；logical reasoning(abstract)

AI总结 TiViBench提出用于评估视频生成模型的推理能力，结合VideoTPO提升推理性能，揭示商业与开源模型在推理潜力上的差异。

Comments Project: https://haroldchen19.github.io/TiViBench-Page/

DR. WELL: Dynamic Reasoning and Learning with Symbolic World Model for Embodied LLM-Based Multi-Agent Collaboration

Narjes Nourzad, Hanqing Yang, Shiyu Chen, Carlee Joe-Wong

机构 * University of Southern California（南加州大学） ； Carnegie Mellon University（卡内基梅隆大学） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；planning(abstract)；分类 cs.CL、cs.AI、cs.LG

ProofSketch: Efficient Verified Reasoning for Large Language Models

Disha Sheshanarayana, Tanishka Magar

机构 * Manipal University Jaipur（马普尔大学斋普尔） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted at NeurIPS 2025, ER Workshop

VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning

Qi Wang, Yanrui Yu, Ye Yuan, Rui Mao, Tianfei Zhou

机构 * Beijing Institute of Technology（北京理工大学） ； Shenzhen University（深圳大学） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；chain-of-thought(abstract)；CoT(abstract)

Comments Accepted by NeurIPS 2025. Code: https://github.com/QiWang98/VideoRFT

Method-Based Reasoning for Large Language Models: Extraction, Reuse, and Continuous Improvement

Hong Su

机构 * School of Computer Science, Chengdu University of Information Technology（计算机科学学院，成都信息科技学院） 纠错

专题命中 逻辑推理 ：reasoning(title,abstract)；logical reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG