cs.CL（2025-09-05）

📊 共 29 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) (17 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (8) 支柱五：交互与反应 (Interaction & Reaction) (1) 支柱六：视频提取与匹配 (Video Extraction) (1) 支柱一：机器人控制 (Robot Control) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (17 篇)

#题目一句话要点标签🔗⭐
1 Do Large Language Models Need Intent? Revisiting Response Generation Strategies for Service Assistant 对比研究意图识别在服务型AI响应生成中的必要性，挑战传统假设 large language model
2 HoPE: Hyperbolic Rotary Positional Encoding for Stable Long-Range Dependency Modeling in Large Language Models 提出HoPE：一种用于稳定长程依赖建模的双曲旋转位置编码 large language model
3 CTCC: A Robust and Stealthy Fingerprinting Framework for Large Language Models via Cross-Turn Contextual Correlation Backdoor 提出CTCC：一种鲁棒且隐蔽的跨轮次上下文相关后门指纹框架，用于保护大型语言模型。 large language model ✅
4 Creativity Benchmark: A benchmark for marketing creativity for large language models 提出Creativity Benchmark以评估大语言模型的市场创意能力 large language model
5 A Study of Large Language Models for Patient Information Extraction: Model Architecture, Fine-Tuning Strategy, and Multi-task Instruction Tuning 研究大型语言模型在患者信息抽取中的应用，探索模型架构、微调策略和多任务指令调优。 large language model
6 Memorization $\neq$ Understanding: Do Large Language Models Have the Ability of Scenario Cognition? 提出双视角评估框架，揭示大语言模型在情景认知方面依赖记忆而非理解 large language model
7 Evaluating Cognitive-Behavioral Fixation via Multimodal User Viewing Patterns on Social Media 提出一种多模态用户行为分析框架，用于评估社交媒体中的认知行为固着现象。 multimodal ✅
8 KERAG: Knowledge-Enhanced Retrieval-Augmented Generation for Advanced Question Answering KERAG：知识增强的检索增强生成框架，提升复杂问答覆盖率与准确性 large language model chain-of-thought
9 Knowledge Collapse in LLMs: When Fluency Survives but Facts Fail under Recursive Synthetic Training 揭示LLM递归合成训练中的知识崩塌现象，提出领域特定训练缓解策略 large language model instruction following
10 WildScore: Benchmarking MLLMs in-the-Wild Symbolic Music Reasoning WildScore：提出一个在真实场景下评估多模态大语言模型音乐推理能力的基准。 large language model multimodal
11 Code Review Without Borders: Evaluating Synthetic vs. Real Data for Review Recommendation 利用LLM生成合成数据，解决新兴语言代码审查推荐系统训练数据不足问题 large language model
12 Research on Multi-hop Inference Optimization of LLM Based on MQUAKE Framework 基于MQUAKE框架的多跳推理优化LLM方法，提升复杂问题解答能力 large language model
13 The Token Tax: Systematic Bias in Multilingual Tokenization 揭示多语言分词偏差：Token Tax对低资源语言的影响与应对 large language model
14 A Lightweight Framework for Trigger-Guided LoRA-Based Self-Adaptation in LLMs 提出SAGE框架，通过触发器引导LoRA自适应调整LLM以提升推理时性能 large language model
15 From Staff Messages to Actionable Insights: A Multi-Stage LLM Classification Framework for Healthcare Analytics 提出多阶段LLM分类框架，从医护人员消息中提取可执行的医疗分析洞见。 large language model
16 Triadic Fusion of Cognitive, Functional, and Causal Dimensions for Explainable LLMs: The TAXAL Framework TAXAL框架：融合认知、功能和因果维度，提升Agentic LLM的可解释性 large language model
17 L1RA: Dynamic Rank Assignment in LoRA Fine-Tuning L1RA：LoRA微调中基于L1正则化的动态秩分配方法 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (8 篇)

#题目一句话要点标签🔗⭐
18 Less is More Tokens: Efficient Math Reasoning via Difficulty-Aware Chain-of-Thought Distillation 提出难度感知的CoT蒸馏方法，提升数学推理效率并减少冗余token生成。 DPO direct preference optimization distillation
19 Post-training Large Language Models for Diverse High-Quality Responses 提出DQO方法，提升大型语言模型后训练阶段生成回复的多样性和质量 reinforcement learning large language model instruction following
20 PLaMo 2 Technical Report PLaMo 2：面向日语的混合架构大型语言模型，通过持续预训练支持32K上下文。 DPO direct preference optimization large language model
21 ACE-RL: Adaptive Constraint-Enhanced Reward for Long-form Generation Reinforcement Learning 提出ACE-RL框架，通过自适应约束增强奖励解决长文本生成中细粒度控制问题。 reinforcement learning large language model
22 AFD-SLU: Adaptive Feature Distillation for Spoken Language Understanding 提出自适应特征蒸馏框架以解决语音理解中的数据稀缺问题 distillation large language model
23 Phonological Representation Learning for Isolated Signs Improves Out-of-Vocabulary Generalization 提出基于音位表征学习的孤立手语识别模型，提升未见手语的泛化能力 representation learning
24 Crosscoding Through Time: Tracking Emergence & Consolidation Of Linguistic Representations Throughout LLM Pretraining 提出基于稀疏互编码器的RelIE方法，追踪LLM预训练过程中语言表征的演化。 representation learning large language model
25 Elucidating the Design Space of Decay in Linear Attention 深入研究线性注意力衰减机制，揭示其设计空间的关键维度 linear attention

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

#题目一句话要点标签🔗⭐
26 ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions 提出ToM-SSI基准，用于评估具身社交互动中智能体的心理理论能力。 dyadic interaction foundation model multimodal

🔬 支柱六：视频提取与匹配 (Video Extraction) (1 篇)

#题目一句话要点标签🔗⭐
27 Decoders Laugh as Loud as Encoders 解码器在幽默理解上可与编码器媲美：GPT-4o在幽默理解上达到RoBERTa水平 HuMoR large language model

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
28 Personality as a Probe for LLM Evaluation: Method Trade-offs and Downstream Effects 研究LLM中人格控制的方法权衡与下游影响，提出多层次稳定性评估框架。 manipulation large language model

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
29 Non-Termination Proving: 100 Million LoC and Beyond Pulse Infinite：利用证明技术检测大型程序中的非终止问题 PULSE

⬅️ 返回 cs.CL 首页 · 🏠 返回主页