cs.CL（2025-08-14）

📊 共 29 篇论文 | 🔗 5 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (25 🔗5) 支柱二：RL算法与架构 (RL & Architecture) (3) 支柱一：机器人控制 (Robot Control) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (25 篇)

#题目一句话要点标签🔗⭐
1 When Language Overrules: Revealing Text Dominance in Multimodal Large Language Models 系统性揭示多模态大语言模型中文本主导现象，并提出token压缩方法有效缓解该问题。 large language model multimodal
2 MAC: A Live Benchmark for Multimodal Large Language Models in Scientific Understanding 提出MAC：一个用于评估多模态大语言模型科学理解能力的动态基准 large language model multimodal ✅
3 DiFaR: Enhancing Multimodal Misinformation Detection with Diverse, Factual, and Relevant Rationales DiFaR：通过多样、真实、相关的理由增强多模态错误信息检测 multimodal chain-of-thought
4 MM-BrowseComp: A Comprehensive Benchmark for Multimodal Browsing Agents 提出MM-BrowseComp多模态浏览代理基准，评估模型在复杂网页环境下的多模态推理能力。 multimodal
5 SproutBench: A Benchmark for Safe and Ethical Large Language Models for Youth SproutBench：面向青少年安全和伦理的大语言模型评测基准 large language model
6 DLLMQuant: Quantizing Diffusion-based Large Language Models DLLMQuant：为基于扩散的大语言模型提出量化方案，提升压缩效率。 large language model
7 Yet another algorithmic bias: A Discursive Analysis of Large Language Models Reinforcing Dominant Discourses on Gender and Race 提出一种定性分析框架，揭示大型语言模型在性别和种族议题上强化主流话语的偏见。 large language model
8 Computational Economics in Large Language Models: Exploring Model Behavior and Incentive Design under Resource Constraints 提出基于计算经济学框架的LLM训练方法，提升资源约束下的模型效率与可解释性 large language model
9 Large Language Models for Summarizing Czech Historical Documents and Beyond 利用大型语言模型在捷克历史文档摘要任务中取得新突破 large language model
10 Rule2Text: A Framework for Generating and Evaluating Natural Language Explanations of Knowledge Graph Rules Rule2Text框架利用大语言模型为知识图谱规则生成自然语言解释，提升可理解性。 large language model chain-of-thought ✅
11 Psyche-R1: Towards Reliable Psychological LLMs through Unified Empathy, Expertise, and Reasoning 提出Psyche-R1，首个融合共情、专业知识和推理能力的中文心理学大语言模型。 large language model chain-of-thought
12 Thinking Inside the Mask: In-Place Prompting in Diffusion LLMs 提出ICE框架，利用扩散LLM的In-Place Prompting提升推理性能并加速计算。 large language model chain-of-thought
13 A Survey on Diffusion Language Models 综述扩散语言模型：探索并行生成、双向上下文建模及可控生成的新范式。 multimodal ✅
14 Reinforced Language Models for Sequential Decision Making 提出MS-GRPO算法，用于提升小规模语言模型在序列决策任务中的性能。 large language model
15 Chain-of-Query: Unleashing the Power of LLMs in SQL-Aided Table Understanding via Multi-Agent Collaboration 提出Chain-of-Query，通过多智能体协作提升LLM在SQL辅助表格理解中的能力 large language model ✅
16 Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics 研究LLM跨语言知识迁移失败问题，揭示统一表征的重要性并提出调控方法 large language model
17 SurfaceLogicKV: Surface and Logic Attention Behaviors are All You Need for Robust KV Cache Compression SurfaceLogicKV：利用表面和逻辑注意力实现鲁棒的KV缓存压缩 large language model
18 Layer-Wise Perturbations via Sparse Autoencoders for Adversarial Text Generation 提出基于稀疏自编码器的分层扰动方法，用于对抗性文本生成，提升大语言模型安全性。 large language model
19 ReportBench: Evaluating Deep Research Agents via Academic Survey Tasks ReportBench：通过学术综述任务评估深度研究智能体的性能 large language model ✅
20 Towards Reliable Multi-Agent Systems for Marketing Applications via Reflection, Memory, and Planning 提出RAMP框架，通过反思、记忆和规划提升营销应用中多智能体系统的可靠性 large language model
21 Approaching the Source of Symbol Grounding with Confluent Reductions of Abstract Meaning Representation Directed Graphs 利用融合归约抽象意义表示有向图，探索符号 grounding 的源头 large language model
22 BIPOLAR: Polarization-based granular framework for LLM bias evaluation BIPOLAR：提出一种基于极化的细粒度框架，用于评估LLM中的偏见。 large language model
23 eDIF: A European Deep Inference Fabric for Remote Interpretability of LLM 构建欧洲深度推理平台eDIF，实现LLM远程可解释性研究 large language model
24 Jailbreaking Commercial Black-Box LLMs with Explicitly Harmful Prompts 提出DH-CoT攻击，有效破解商业黑盒LLM的恶意内容防御 chain-of-thought
25 Cross-Prompt Encoder for Low-Performing Languages 提出跨提示编码器XPE，提升低资源语言在参数高效微调中的性能 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (3 篇)

#题目一句话要点标签🔗⭐
26 Vocabulary Expansion of Large Language Models via Kullback-Leibler-Based Self-Distillation 提出基于KL散度的自蒸馏方法，用于扩展冻结LLM的词汇表以适应特定领域术语。 distillation large language model
27 Making Qwen3 Think in Korean with Reinforcement Learning 提出基于强化学习的两阶段微调方法，使Qwen3具备韩语思维能力 reinforcement learning large language model chain-of-thought
28 ALAS: Autonomous Learning Agent for Self-Updating Language Models ALAS：用于自更新语言模型的自主学习Agent，解决LLM知识截止问题。 DPO direct preference optimization distillation

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
29 SSRL: Self-Search Reinforcement Learning 提出SSRL：利用自搜索强化学习提升大语言模型在Agentic搜索任务中的效率。 sim-to-real reinforcement learning large language model

⬅️ 返回 cs.CL 首页 · 🏠 返回主页