cs.CL（2025-10-11）

📊 共 28 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (22 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (5) 支柱五：交互与反应 (Interaction & Reaction) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (22 篇)

#题目一句话要点标签🔗⭐
1 Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning 提出REFRAIN框架以解决链式思维推理中的过度思考问题 large language model chain-of-thought
2 On-device System of Compositional Multi-tasking in Large Language Models 针对LLM在端侧的组合多任务处理，提出一种高效的适配器融合方法。 large language model
3 Meronymic Ontology Extraction via Large Language Models 利用大型语言模型自动提取产品部件关系的本体，提升产品知识组织效率。 large language model
4 A Survey of Inductive Reasoning for Large Language Models 对大型语言模型中的归纳推理进行综述，并提出统一的评估方法。 large language model
5 Large Language Model Sourcing: A Survey 针对大语言模型溯源问题，提出包含模型、结构、数据和外部信息的多维度溯源方法综述。 large language model
6 BILLY: Steering Large Language Models via Merging Persona Vectors for Creative Generation BILLY：通过融合Persona向量引导大语言模型进行创造性生成 large language model
7 Revisiting the UID Hypothesis in LLM Reasoning Traces 揭示LLM推理轨迹中信息密度非均匀性，挑战UID假设 large language model chain-of-thought
8 Audit-of-Understanding: Posterior-Constrained Inference for Mathematical Reasoning in Language Models 提出Audit-of-Understanding以解决语言模型推理中的幻觉问题 large language model chain-of-thought
9 Debiasing LLMs by Masking Unfairness-Driving Attention Heads DiffHeads：通过屏蔽不公平驱动的注意力头来消除LLM的偏见 large language model chain-of-thought
10 Hybrid OCR-LLM Framework for Enterprise-Scale Document Information Extraction Under Copy-heavy Task 提出混合OCR-LLM框架，解决企业级海量重复文档的信息抽取难题 large language model multimodal
11 Scheming Ability in LLM-to-LLM Strategic Interactions 研究LLM在策略互动中的欺骗能力，揭示其潜在的欺骗倾向 large language model chain-of-thought
12 BenchPress: A Human-in-the-Loop Annotation System for Rapid Text-to-SQL Benchmark Curation BenchPress：一种人机协同的标注系统，用于快速构建Text-to-SQL基准 large language model ✅
13 ConsistencyAI: A Benchmark to Assess LLMs' Factual Consistency When Responding to Different Demographic Groups ConsistencyAI：评估LLM对不同人群回答的事实一致性基准 large language model
14 EvoEdit: Evolving Null-space Alignment for Robust and Efficient Knowledge Editing EvoEdit：通过演化零空间对齐实现鲁棒高效的知识编辑 large language model
15 Are LLMs Empathetic to All? Investigating the Influence of Multi-Demographic Personas on a Model's Empathy 提出多元化人群框架以研究LLMs的同理心表现 large language model
16 Language steering in latent space to mitigate unintended code-switching 提出基于隐空间语言引导的方法，用于缓解大型语言模型中意外的代码切换问题。 large language model
17 On the Entity-Level Alignment in Crosslingual Consistency 通过实体对齐增强跨语言一致性，提升多语言大语言模型的事实性知识回忆能力。 large language model
18 Backdoor Collapse: Eliminating Unknown Threats via Known Backdoor Aggregation in Language Models 提出Backdoor Collapse框架以解决语言模型中的后门攻击问题 large language model
19 ImCoref-CeS: An Improved Lightweight Pipeline for Coreference Resolution with LLM-based Checker-Splitter Refinement 提出ImCoref-CeS框架，结合增强的监督模型与LLM推理，提升共指消解性能。 large language model
20 Weed Out, Then Harvest: Dual Low-Rank Adaptation is an Effective Noisy Label Detector for Noise-Robust Learning Delora：双低秩自适应噪声标签检测器，提升噪声鲁棒学习效果 large language model
21 LinearRAG: Linear Graph Retrieval Augmented Generation on Large-scale Corpora 提出LinearRAG以解决大规模语料库中的信息检索问题 large language model ✅
22 Beyond Fertility: Analyzing STRR as a Metric for Multilingual Tokenization Evaluation 提出STRR指标，用于评估多语言分词器的跨语言公平性 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (5 篇)

#题目一句话要点标签🔗⭐
23 Unilaw-R1: A Large Language Model for Legal Reasoning with Reinforcement Learning and Iterative Inference 提出Unilaw-R1：一个基于强化学习和迭代推理的法律领域大语言模型 reinforcement learning large language model chain-of-thought
24 Path Drift in Large Reasoning Models:How First-Person Commitments Override Safety 揭示大型推理模型中的路径漂移现象，提出防御策略以提升安全性 RLHF large language model chain-of-thought
25 Unifying Tree Search Algorithm and Reward Design for LLM Reasoning: A Survey 统一树搜索算法与奖励设计的LLM推理综述：形式化与未来方向 reward design large language model
26 Text2Token: Unsupervised Text Representation Learning with Token Target Prediction 提出Text2Token框架，通过预测目标token实现无监督文本表示学习 representation learning contrastive learning
27 A-IPO: Adaptive Intent-driven Preference Optimization 提出A-IPO以解决偏好优化中用户意图缺失和对抗鲁棒性不足的问题 DPO direct preference optimization

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
28 Mathematics with large language models as provers and verifiers 利用大型语言模型作为证明器和验证器解决数学问题 IMoS large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页