大模型推理能力 - arXivDaily 专题

2310.15258 2023-10-25 cs.CL 83%

Breaking the Language Barrier: Improving Cross-Lingual Reasoning with Structured Self-Attention

Negar Foroutan, Mohammadreza Banaei, Karl Aberer, Antoine Bosselut

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

Comments EMNLP 2023 - Findings

2212.08286 2023-10-02 cs.CL 83%

ALERT: Adapting Language Models to Reasoning Tasks

Ping Yu, Tianlu Wang, Olga Golovneva, Badr AlKhamissi, Siddharth Verma, Zhijing Jin, Gargi Ghosh, Mona Diab, Asli Celikyilmaz

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

2305.16572 2023-05-29 cs.CL 83%

Counterfactual reasoning: Testing language models' understanding of hypothetical scenarios

Jiaxuan Li, Lang Yu, Allyson Ettinger

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

Comments Accepted to ACL 2023. arXiv admin note: substantial text overlap with arXiv:2212.03278

2302.00618 2023-02-02 cs.CL 83%

Synthetic Prompting: Generating Chain-of-Thought Demonstrations for Large Language Models

Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, Weizhu Chen

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

Comments Preprint

2211.15566 2023-01-16 cs.AI 83%

Neuro-Symbolic Spatio-Temporal Reasoning

Jae Hee Lee, Michael Sioutis, Kyra Ahrens, Marjan Alirezaie, Matthias Kerzel, Stefan Wermter

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

Comments Contribution to the book "A Compendium of Neuro-Symbolic Artificial Intelligence", which is to appear in the first half of 2023

2212.03278 2022-12-08 cs.CL 83%

Counterfactual reasoning: Do language models need world knowledge for causal understanding?

Jiaxuan Li, Lang Yu, Allyson Ettinger

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

Comments 5 pages, accepted to nCSI workshop at NeurIPS 2022

2208.11556 2022-10-20 cs.AI cs.MA 83%

Knowledge-based and Data-driven Reasoning and Learning for Ad Hoc Teamwork

Hasra Dodampegama, Mohan Sridharan

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

Comments Presented at the AI-HRI Symposium at AAAI Fall Symposium Series(FSS), 2022 (arXiv:cs/2209.14292)

2208.13841 2022-08-31 cs.AI 83%

Visual-Imagery-Based Analogical Construction in Geometric Matrix Reasoning Task

Yuan Yang, Keith McGreggor, Maithilee Kunda

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

2009.00418 2021-01-19 cs.AI cs.LO 83%

Machine Reasoning Explainability

Kristijonas Cyras, Ramamurthy Badrinath, Swarup Kumar Mohalik, Anusha Mujumdar, Alexandros Nikou, Alessandro Previti, Vaishnavi Sundararajan, Aneta Vulgarakis Feljan

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

2011.04864 2020-11-11 cs.CL 83%

Natural Language Inference in Context -- Investigating Contextual Reasoning over Long Texts

Hanmeng Liu, Leyang Cui, Jian Liu, Yue Zhang

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

1905.10621 2019-05-28 cs.AI 83%

Dynamic Epistemic Logic with ASP Updates: Application to Conditional Planning

Pedro Cabalar, Jorge Fandinno, Luis Fariñas del Cerro

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

2608.08514 2026-08-13 cs.AI cs.CL cs.LG 交叉投稿 83%

Reproducing and Stress-Testing Two Approaches to LLM Reasoning Reliability: Test-Time Probability Aggregation and Logic-Representation Editing

复现与压力测试两种提升大语言模型（LLM）推理可靠性的方法：测试时概率聚合与逻辑表示编辑

Minhan Cho, Jimin Kweon

机构 * Sungkyunkwan University（成均馆大学）

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

AI总结本研究复现了RPC与LCF两种提升LLM推理可靠性的方法，在多任务域与多模型上压力测试，发现RPC优势未达显著且样本量增大后效果逆转，LCF效果弱且无统计显著性。

Comments 16 pages, 3 figures, 9 tables. Code, data, and experiment logs: https://github.com/rabqatab/llm-reasoning-reliability-reproduction

2608.03450 2026-08-05 cs.MM cs.AI cs.CL cs.CV 新提交 82%

Balancing Efficiency and Efficacy: Training-Free Attention-Guided Switching Between Explicit and Latent Thoughts for MLLMs

平衡效率与效能：面向多模态大语言模型（MLLM）的无训练注意力引导式显式与隐式思维切换

Haoqian Kang, Liupeng Li, Kuofeng Gao, Jinpeng Wang, Zhenyu Lu, Bin Chen, Ke Chen, Yaowei Wang

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

AI总结针对MLLM推理中感知与逻辑混淆的问题，提出无训练的注意力引导式切换框架，通过视觉-文本注意力比率自适应切换显式与隐式推理，实现性能与效率的双重提升。

Comments Accepted by ACM MM 2026. 10 pages, 6 figures, 5 tables

2511.09149 2026-07-14 cs.LG cs.AI cs.MA 版本更新 82%

Enabling Agents to Communicate Entirely in Latent Space

使智能体能够在潜在空间中完全交流

Zhuoyun Du, Runze Wang, Huiyu Bai, Zouying Cao, Xiaoyong Zhu, Yu Cheng, Bo Zheng, Wei Chen, Haochao Ying

机构 * State Key Lab of CAD&CG（CAD与CG国家重点实验室）； Future Living Lab of Alibaba（阿里巴巴未来生活实验室）； Zhejiang Key Laboratory of Medical Imaging Artificial Intelligence（浙江医学影像人工智能重点实验室）； Shanghai Jiao Tong University（上海交通大学）

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

AI总结本文提出Interlat方法，通过利用LLM的连续隐藏状态实现智能体间的潜在空间通信，实验表明其在跨异构模型中表现优异，提升了推理速度并保持性能。

Comments Accepted to ACL 2026

2505.12415 2026-05-14 cs.CL cs.AI 82%

Table-R1: Region-based Reinforcement Learning for Table Understanding

基于区域的表格理解强化学习：Table-R1

Zhenhe Wu, Jian Yang, Zhongjiang He, Changzai Pan, Jie Zhang, Jiaheng Liu, Xianjie Wu, Yu Zhao, Shuangyong Song, Yongxiang Li, Zhoujun Li, Xueling Li

机构 * Beihang University（北京航空航天大学）； Xingchen AGI Lab（星辰AGI实验室）； China Telecom Artificial Intelligence Technology (Beijing) Co., Ltd（中国电信人工智能技术（北京）有限公司）； Institute of Artificial Intelligence (TeleAI), China Telecom（中国电信人工智能研究院（TeleAI））

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

AI总结本文提出Table-R1，通过区域证据增强和TARPO优化提升LLM的表格理解能力，在多个数据集上提升14.36个百分点，同时降低响应token消耗。

2604.12426 2026-04-15 cs.LG cs.CL 82%

Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task

Transformer 是否会根据深度进行自适应使用？来自关系推理任务的证据

Alicia Curth, Rachel Lawrence, Sushrut Karmalkar, Niranjani Prasad

机构 * Microsoft Research Cambridge（微软研究院剑桥分部）

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

AI总结本文通过控制多跳关系推理任务研究Transformer是否自适应使用深度，发现预训练模型在简单任务中使用较少层，而微调模型在更复杂的任务中更一致地自适应调整深度。

Comments Accepted at the ICLR 2026 Workshop on Logical Reasoning of Large Language Models

2507.20999 2025-09-17 cs.LG cs.CL 82%

LoRA-PAR: A Flexible Dual-System LoRA Partitioning Approach to Efficient LLM Fine-Tuning

Yining Huang, Bin Li, Keke Tang, Meilian Chen

机构 * School of Politics and Public Administration, South China Normal University（华南师范大学政治与公共管理学院）； Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences（中国科学院深圳先进技术研究所）； University of Chinese Academy of Sciences（中国科学院大学）； Shenyang institute of computing technology, Chinese academy of sciences（中国科学院沈阳计算技术研究所）