大模型推理能力 - arXivDaily 专题

2605.19944 2026-05-20 cs.LG cs.AI cs.CC cs.CL 82%

A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits

关于推理的测度论分析：结构泛化与近似限制

Yuyang Zhang, Yifu Zhang, Xuehai Zhou, Xiaoyin Chen

机构 * McGill University（麦吉尔大学）； Mila - Quebec AI Institute（魁北克AI研究所）； Université de Montréal（蒙特利尔大学）

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

AI总结本文通过最优传输理论分析推理过程，揭示了结构泛化和近似限制的理论机制，发现位置依赖注意力机制和Transformer电路深度对推理性能有显著影响。

Comments Preprint

2605.14036 2026-05-15 cs.AI cs.CC cs.CL cs.LG 82%

Enhanced and Efficient Reasoning in Large Learning Models

增强和高效的大型学习模型推理

Leslie G. Valiant

机构 * John A. Paulson School of Engineering and Applied Sciences（约翰·A·保罗森工程与应用科学学院）

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

AI总结本文提出一种高效且基于原理的推理方法，用于改进大型语言模型，通过预处理和机器学习流程提升推理能力，实现更稳健的逻辑推理。

2512.00775 2026-05-13 cs.RO cs.SY eess.SY 82%

SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning

SAGAS：语义感知的图辅助拼接用于离线时序逻辑规划

Ruijia Liu, Ancheng Hou, Xiang Yin

机构 * School of Automation and Intelligent Sensing（自动化与智能感知学院）

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

AI总结 SAGAS结合符号合成的组成性和从离线轨迹学习的数据驱动可达结构，实现零样本泛化，无需任务特定奖励或在线交互。

2505.00017 2026-04-10 cs.CL cs.AI cs.DB cs.LG 82%

ReCellTy: Domain-Specific Knowledge Graph Retrieval-Augmented LLMs Reasoning Workflow for Single-Cell Annotation

ReCellTy：领域特定的知识图谱检索增强型大语言模型推理工作流用于单细胞注释

Dezheng Han, Yibin Jia, Ruxiao Chen, Wenjie Han, Shuaishuai Guo, Jianbo Wang

机构 * Shandong University（山东大学）； Qilu Hospital of Shandong University（山东大学齐鲁医院）； Johns Hopkins University（约翰霍普金斯大学）； Cognicore Artificial Intelligence Co., Ltd.（科智人工智能有限公司）

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

AI总结本文提出ReCellTy，通过整合生物信息知识图谱和多任务推理工作流，提升单细胞注释的准确性与自动化水平，改进人类评估得分和语义相似度，缩小大中小语言模型性能差距。

2603.04124 2026-03-05 cs.AI cond-mat.mtrl-sci cs.CL cs.LG 82%

BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning

BeamPERL: 参数高效强化学习与可验证奖励用于结构梁力学推理

Tarjei Paule Hage, Markus J. Buehler

机构 * Department of Mechanical Engineering（机械工程系）； Massachusetts Institute of Technology（麻省理工学院）； Department of Civil and Environmental Engineering（土木与环境工程系）； Schwarzman College of Computing（施瓦茨曼计算学院）

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

AI总结 BeamPERL通过参数高效强化学习与可验证奖励，提升紧凑语言模型在结构梁力学推理中的能力，发现精确奖励无法保证可转移的物理推理，需结合结构化推理支架以实现稳健科学推理。

2601.11556 2026-03-02 cs.LG cs.AI cs.CL cs.SD eess.AS 82%

CSyMR: Benchmarking Compositional Music Information Retrieval in Symbolic Music Reasoning

CSyMR：在符号音乐推理中进行组合音乐信息检索的基准测试

Boyang Wang, Yash Vishe, Xin Xu, Zachary Novack, Xunyi Jiang, Julian McAuley, Junda Wu

机构 * University of California, San Diego（加州大学圣地亚哥分校）

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

AI总结 CSyMR-Bench通过工具增强的检索和推理框架，在符号音乐推理中实现组合音乐信息检索的准确率提升。

2511.19514 2026-02-26 cs.IR 82%

SCoTER: Structured Chain-of-Thought Transfer for Enhanced Recommendation

SCoTER: 结构化推理链转移以提升推荐

Jie Jiang, Yang Wu, Qian Li, Yuling Xiong, Hongbo Tang, Xun Liu, Haoze Wang, Jun Zhang, Huan Yu, Hailong Shi

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

AI总结 SCoTER通过结构化推理链转移提升推荐系统，解决推理模式发现和结构保持问题，实现推荐效果和成本的双重优化。

2602.17693 2026-02-23 cs.LG cs.AI cs.CL 82%

A Case Study of Selected PTQ Baselines for Reasoning LLMs on Ascend NPU

在Ascend NPU上对推理LLM的选定PTQ基线进行案例研究

Yuchen Luo, Fangyue Zhu, Ruining Zhou, Mingzhe Huang, Jian Zhu, Fanyu Fan, Wei Shao

机构 * School of Mathematics and Statistics, Wuhan University（武汉大学数学与统计学院）； Huawei（华为公司）

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

AI总结本文研究了在Ascend NPU上不同PTQ方法对推理LLM的影响，发现4位权重量化在大模型中可行，但激进方案存在稳定性问题，而8位量化更稳定，实际部署中动态量化开销限制了加速效果。

2601.09727 2026-01-16 cs.CL cs.AI cs.IR cs.LG 82%

SciNets: Graph-Constrained Multi-Hop Reasoning for Scientific Literature Synthesis

SciNets: 图约束多跳推理用于科学文献综合

Sauhard Dubey

机构 * Jaypee Institute of Information Technology（杰伊佩 Institute 信息科技学院）

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

AI总结 SciNets通过图约束多跳推理实现科学文献综合，通过多跳路径连接罕见共现的概念，提升机制解释的稳定性和多样性。

Comments 19 pages, 2 figures

2512.24165 2026-01-01 cs.CV 82%

DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models

DiffThinker: 向基于扩散模型的生成多模态推理迈进

Zefeng He, Xiaoye Qu, Yafu Li, Tong Zhu, Siyuan Huang, Yu Cheng

机构 * Shanghai AI Laboratory（上海人工智能实验室）； Nanjing University（南京大学）； Shanghai Jiao Tong University（上海交通大学）； The Chinese University of Hong Kong（香港中文大学）

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

AI总结 DiffThinker通过基于扩散模型的生成方法，在多模态推理任务中实现了更高效的视觉推理和更精确的空间处理，显著优于现有模型。

Comments Project page: https://diffthinker-project.github.io

2512.21250 2025-12-25 cs.CR cs.MA 82%

CoTDeceptor:Adversarial Code Obfuscation Against CoT-Enhanced LLM Code Agents

CoTDeceptor：对抗增强CoT的LLM代码代理的对抗性代码混淆

Haoyang Li, Mingjin Li, Jinxin Zuo, Siqi Li, Xiao Li, Hao Wu, Yueming Lu, Xiaochuan He

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

AI总结 CoTDeceptor通过构建多阶段混淆策略链，有效对抗增强CoT的LLM检测器，实现对14个漏洞类别的绕过。

2512.17062 2025-12-22 cs.RO 82%

Lang2Manip: A Tool for LLM-Based Symbolic-to-Geometric Planning for Manipulation

Lang2Manip: 一种基于LLM的符号到几何规划工具用于操作

Muhayy Ud Din, Jan Rosell, Waseem Akram, Irfan Hussain

机构 * Khalifa University Center for Autonomous Robotic Systems (KUCARS)（卡利法大学自主机器人系统中心）； Khalifa University（卡利法大学）； Institute of Industrial and Control Engineering (IOC)（工业与控制工程研究所）； Universitat Politècnica de Catalunya（加泰罗尼亚理工大学）

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

AI总结 Lang2Manip通过结合LLM和Kautham框架，实现通用的符号到几何操作规划，适用于多种机器人和任务。

Comments Submitted to ICARA

2512.07218 2025-12-09 cs.CL cs.AI cs.LG 82%

NeSTR: A Neuro-Symbolic Abductive Framework for Temporal Reasoning in Large Language Models

NeSTR：一种用于大语言模型时间推理的神经符号抽象框架

Feng Liang, Weixin Zeng, Runhao Zhao, Xiang Zhao

机构 * National Key Laboratory of Big Data and Decision（大数据与决策国家重点实验室）

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

AI总结 NeSTR通过结合符号表示与反思推理，提升大语言模型在复杂时间约束下的推理能力。

Comments Accepted by AAAI 2026

2511.02194 2025-11-05 cs.AI cs.CL cs.CY cs.LG 82%

Personalized Decision Modeling: Utility Optimization or Textualized-Symbolic Reasoning

Yibo Zhao, Yang Zhao, Hongru Du, Hao Frank Yang

机构 * Department of Civil and Systems Engineering（土木与系统工程系）； Johns Hopkins University（约翰霍普金斯大学）； Department of Systems and Information Engineering（系统与信息工程系）； University of Virginia（弗吉尼亚大学）； Johns Hopkins Data Science and AI Institute（约翰霍普金斯大学数据科学与人工智能研究所）

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

2407.07612 2025-10-27 cs.LG cs.AI cs.CL 82%

Teaching Transformers Causal Reasoning through Axiomatic Training

Aniket Vashishtha, Abhinav Kumar, Atharva Pandey, Abbavaram Gowtham Reddy, Kabir Ahuja, Vineeth N Balasubramanian, Amit Sharma

机构 * uiuc（伊利诺伊大学香槟分校）； mit（麻省理工学院）； msr（微软研究院）； cispa（CISPA信息安全研究中心）

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

2509.07017 2025-09-10 cs.AI cs.CL cs.LG 82%

From Eigenmodes to Proofs: Integrating Graph Spectral Operators with Symbolic Interpretable Reasoning

Andrew Kiruluta, Priscilla Burity

机构 * Andrew Kiruluta and Priscilla Burity（独立研究者）

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

2509.03537 2025-09-05 cs.CL cs.AI cs.LG 82%

AR$^2$: Adversarial Reinforcement Learning for Abstract Reasoning in Large Language Models

Cheng-Kai Yeh, Hsing-Wang Lee, Chung-Hung Kuo, Hen-Hsen Huang

机构 * Department of Mathematical Sciences National Chengchi University Taipei Taiwan（数学科学系国立政治大学台北台湾）； Department of Computer Science National Chengchi University Taipei Taiwan（计算机科学系国立政治大学台北台湾）； Institute of Information Science Academia Sinica Taipei Taiwan（信息科学研究所中国学术院台北台湾）； National Chengchi University（国立政治大学）

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

Comments 7 pages, accepted by CIKM 2025 as a short paper

2508.18988 2025-08-27 cs.CL cs.AI cs.LG 82%

Interpretable by AI Mother Tongue: Native Symbolic Reasoning in Neural Models

Hung Ming Liu

机构 * PARRAWA AI

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

Comments 25 pages, 9 figures. The AI Intuition Explorer dashboard is available at: https://cyrilliu1974.github.io/github.io/vi.html

2508.05005 2025-08-08 cs.SE 82%

Generative AI for Object-Oriented Programming: Writing the Right Code and Reasoning the Right Logic

Gang Xu, Airong Wang, Yushan Pan

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

2409.01667 2025-06-10 cs.CV 82%

VProChart: Answering Chart Question through Visual Perception Alignment Agent and Programmatic Solution Reasoning

Muye Huang, Lingling Zhang, Lai Han, Wenjun Wu, Xinyu Zhang, Jun Liu

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

2505.22407 2025-05-29 cs.CV 82%

Self-Reflective Reinforcement Learning for Diffusion-based Image Reasoning Generation

Jiadong Pan, Zhiyuan Ma, Kaiyan Zhang, Ning Ding, Bowen Zhou

机构 * Department of Electronic Engineering, Tsinghua University, Beijing, China（电子工程系，清华大学，北京，中国）； Shanghai AI Laboratory, Shanghai, China（上海人工智能实验室，上海，中国）

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

2505.15873 2025-05-27 cs.PL 82%

Abstractions-of-Thought: Intermediate Representations for LLM Reasoning in Hardware Design

Matthew DeLorenzo, Kevin Tieu, Prithwish Jana, Piyush Jha, Dileep Kalathil, Vijay Ganesh, Jeyavijayan Rajendran

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

2410.13018 2024-10-18 cs.AI cs.CL cs.LG 82%

Learning Representations for Reasoning: Generalizing Across Diverse Structures

Zhaocheng Zhu

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

Comments PhD thesis

2407.18062 2024-07-26 cs.SD eess.AS 82%

Audio Entailment: Assessing Deductive Reasoning for Audio Understanding

Soham Deshmukh, Shuo Han, Hazim Bukhari, Benjamin Elizalde, Hannes Gamper, Rita Singh, Bhiksha Raj

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

2406.06588 2024-06-12 cs.CL cs.AI cs.LG cs.NE 82%

Assessing the Emergent Symbolic Reasoning Abilities of Llama Large Language Models

Flavio Petruzzellis, Alberto Testolin, Alessandro Sperduti

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

Comments Accepted at 33rd International Conference on Artificial Neural Networks (ICANN24)

2402.14809 2024-06-04 cs.CL cs.AI cs.LG 82%

CriticBench: Benchmarking LLMs for Critique-Correct Reasoning

Zicheng Lin, Zhibin Gou, Tian Liang, Ruilin Luo, Haowei Liu, Yujiu Yang

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

Comments ACL 2024 Findings

2311.00445 2024-04-12 cs.CL cs.AI cs.LG 82%

A Systematic Comparison of Syllogistic Reasoning in Humans and Language Models

Tiwalayo Eisape, MH Tessler, Ishita Dasgupta, Fei Sha, Sjoerd van Steenkiste, Tal Linzen

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

Comments NAACL 2024

2404.01591 2024-04-03 cs.CV 82%

Language Model Guided Interpretable Video Action Reasoning

Ning Wang, Guangming Zhu, HS Li, Liang Zhang, Syed Afaq Ali Shah, Mohammed Bennamoun

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

Comments Accepted by CVPR 2024

2310.05253 2023-10-23 cs.CL cs.AI cs.LG 82%

Explainable Claim Verification via Knowledge-Grounded Reasoning with Large Language Models

Haoran Wang, Kai Shu

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

Comments Findings of EMNLP 2023

2309.13072 2023-09-26 cs.CL cs.AI cs.LG 82%

Weakly Supervised Reasoning by Neuro-Symbolic Approaches

Xianggen Liu, Zhengdong Lu, Lili Mou

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

Comments Compendium of Neurosymbolic Artificial Intelligence, 665--692, 2023, IOS Press

1. 逻辑推理 3033 篇

A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits

Enhanced and Efficient Reasoning in Large Learning Models

SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning

ReCellTy: Domain-Specific Knowledge Graph Retrieval-Augmented LLMs Reasoning Workflow for Single-Cell Annotation

BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning

CSyMR: Benchmarking Compositional Music Information Retrieval in Symbolic Music Reasoning

SCoTER: Structured Chain-of-Thought Transfer for Enhanced Recommendation

A Case Study of Selected PTQ Baselines for Reasoning LLMs on Ascend NPU

SciNets: Graph-Constrained Multi-Hop Reasoning for Scientific Literature Synthesis

DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models

CoTDeceptor:Adversarial Code Obfuscation Against CoT-Enhanced LLM Code Agents

Lang2Manip: A Tool for LLM-Based Symbolic-to-Geometric Planning for Manipulation

NeSTR: A Neuro-Symbolic Abductive Framework for Temporal Reasoning in Large Language Models

Personalized Decision Modeling: Utility Optimization or Textualized-Symbolic Reasoning

Teaching Transformers Causal Reasoning through Axiomatic Training

From Eigenmodes to Proofs: Integrating Graph Spectral Operators with Symbolic Interpretable Reasoning

AR$^2$: Adversarial Reinforcement Learning for Abstract Reasoning in Large Language Models

Interpretable by AI Mother Tongue: Native Symbolic Reasoning in Neural Models

Generative AI for Object-Oriented Programming: Writing the Right Code and Reasoning the Right Logic

VProChart: Answering Chart Question through Visual Perception Alignment Agent and Programmatic Solution Reasoning

Self-Reflective Reinforcement Learning for Diffusion-based Image Reasoning Generation

Abstractions-of-Thought: Intermediate Representations for LLM Reasoning in Hardware Design

Learning Representations for Reasoning: Generalizing Across Diverse Structures

Audio Entailment: Assessing Deductive Reasoning for Audio Understanding

Assessing the Emergent Symbolic Reasoning Abilities of Llama Large Language Models

CriticBench: Benchmarking LLMs for Critique-Correct Reasoning

A Systematic Comparison of Syllogistic Reasoning in Humans and Language Models

Language Model Guided Interpretable Video Action Reasoning

Explainable Claim Verification via Knowledge-Grounded Reasoning with Large Language Models

Weakly Supervised Reasoning by Neuro-Symbolic Approaches