cs.CL（2025-07-21）

📊 共 29 篇论文 | 🔗 3 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) (7 🔗1)

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

#题目一句话要点标签🔗⭐
1 Discrete Tokenization for Multimodal LLMs: A Comprehensive Survey 综述：多模态LLM的离散Token化方法，聚焦向量量化及其在LLM中的应用。 large language model multimodal ✅
2 Efficient Compositional Multi-tasking for On-device Large Language Models 提出面向端侧LLM的高效组合式多任务学习方法，解决资源受限场景下的复杂任务执行问题。 large language model
3 MD-LLM-1: A Large Language Model for Molecular Dynamics 提出MD-LLM-1，利用大语言模型学习蛋白质动力学并预测构象状态 large language model
4 Is Large Language Model Performance on Reasoning Tasks Impacted by Different Ways Questions Are Asked? 研究不同提问方式对大语言模型推理任务性能的影响 large language model
5 Leveraging Context for Multimodal Fallacy Classification in Political Debates 利用上下文信息，解决政治辩论中多模态谬误分类问题 multimodal
6 Metaphor and Large Language Models: When Surface Features Matter More than Deep Understanding 揭示大语言模型在隐喻理解中过度依赖表面特征的现象 large language model
7 A Novel Self-Evolution Framework for Large Language Models 提出双阶段自进化框架DPSE，提升大语言模型领域认知和用户偏好对齐能力。 large language model
8 The Prompt Makes the Person(a): A Systematic Evaluation of Sociodemographic Persona Prompting for Large Language Models 系统评估社会人口学角色提示对大语言模型的影响，揭示刻板印象与模型选择的关键因素 large language model
9 Understanding Large Language Models' Ability on Interdisciplinary Research 提出IDRBench：评估大语言模型在跨学科研究中产生创新性研究思路能力的基准。 large language model
10 Reasoning Models are Test Exploiters: Rethinking Multiple-Choice 揭示推理模型在多项选择题中的作弊行为，重新评估其推理能力 large language model chain-of-thought
11 Learning without training: The implicit dynamics of in-context learning 揭示Transformer在上下文学习中的隐式动态机制，无需额外训练即可实现泛化 large language model
12 CoLD: Counterfactually-Guided Length Debiasing for Process Reward Models 提出CoLD框架，通过反事实引导消除过程奖励模型中的长度偏差。 large language model
13 ASPERA: A Simulated Environment to Evaluate Planning for Complex Action Execution ASPERA：用于评估复杂动作执行规划的模拟环境 large language model
14 Probing Information Distribution in Transformer Architectures through Entropy Analysis 利用熵分析探究Transformer架构中的信息分布 large language model
15 Semantic Convergence: Investigating Shared Representations Across Scaled LLMs 研究表明，不同规模Gemma-2模型在内部概念表示上具有趋同性。 large language model
16 How and Where to Translate? The Impact of Translation Strategies in Cross-lingual LLM Prompting 研究跨语言LLM提示中翻译策略对RAG分类任务的影响，优化提示策略提升跨语言知识共享。 large language model
17 Deep Researcher with Test-Time Diffusion 提出Test-Time Diffusion Deep Researcher (TTD-DR)，解决LLM在生成复杂研究报告时性能瓶颈问题。 large language model
18 3LM: Bridging Arabic, STEM, and Code through Benchmarking 3LM：构建阿拉伯语、STEM和代码的LLM评测基准 large language model
19 Reservoir Computing as a Language Model 提出水库计算作为语言模型以解决能耗与速度瓶颈 large language model
20 P3: Prompts Promote Prompting P3：通过迭代优化系统和用户提示，提升大语言模型性能 large language model
21 STITCH: Simultaneous Thinking and Talking with Chunked Reasoning for Spoken Language Models 提出STITCH，一种用于语音语言模型的同步思考与说话的分块推理方法 chain-of-thought ✅
22 On the Inevitability of Left-Leaning Political Bias in Aligned Language Models 对齐语言模型中左倾政治偏见的必然性分析 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (7 篇)

#题目一句话要点标签🔗⭐
23 Chinchunmei at SemEval-2025 Task 11: Boosting the Large Language Model's Capability of Emotion Perception using Contrastive Learning 利用对比学习增强大语言模型的情感感知能力，应用于SemEval-2025情感检测任务。 DPO contrastive learning large language model
24 ChiMed 2.0: Advancing Chinese Medical Dataset in Facilitating Large Language Modeling ChiMed 2.0：构建大规模中文医疗数据集，促进大型语言模型发展 reinforcement learning RLHF large language model
25 Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models 提出基于过程验证的混合测试时缩放方法，提升大语言模型推理能力 reinforcement learning large language model
26 The Impact of Language Mixing on Bilingual LLM Reasoning 研究表明双语LLM推理中的语言混合是一种策略性行为，并提出引导解码方法提升推理精度。 reinforcement learning large language model chain-of-thought
27 Learning to Extract Rational Evidence via Reinforcement Learning for Retrieval-Augmented Generation EviOmni：提出一种基于强化学习的检索增强生成证据抽取方法，提升LLM生成质量。 reinforcement learning large language model
28 Collaborative Distillation Strategies for Parameter-Efficient Language Model Deployment 提出多教师协同蒸馏策略，用于参数高效的语言模型部署 distillation large language model
29 Stabilizing Knowledge, Promoting Reasoning: Dual-Token Constraints for RLVR 提出Archer，通过双Token约束强化学习提升LLM推理能力并稳定知识。 reinforcement learning large language model ✅

⬅️ 返回 cs.CL 首页 · 🏠 返回主页