cs.AI（2025-05-20）

📊 共 45 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) (31 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (10 🔗1) 支柱一：机器人控制 (Robot Control) (3) 支柱八：物理动画 (Physics-based Animation) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (31 篇)

#题目一句话要点标签🔗⭐
1 Large Language Model-Driven Distributed Integrated Multimodal Sensing and Semantic Communications 提出LLM驱动的分布式集成多模态感知与语义通信框架，提升复杂环境下的感知精度。 large language model multimodal
2 Multimodal RAG-driven Anomaly Detection and Classification in Laser Powder Bed Fusion using Large Language Models 提出基于多模态RAG的增材制造异常检测与分类框架，无需训练数据。 large language model multimodal
3 Towards a Foundation Model for Communication Systems 提出面向通信系统的Transformer基础模型，实现多特征估计。 foundation model multimodal
4 Debating for Better Reasoning: An Unsupervised Multimodal Approach 提出一种无监督多模态辩论框架，提升视觉问答模型推理能力。 large language model multimodal
5 Large Language Model Powered Decision Support for a Metal Additive Manufacturing Knowledge Graph 提出基于大语言模型的金属增材制造知识图谱决策支持系统，实现自然语言交互。 large language model
6 SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment 提出SAFEPATH以解决大型推理模型的安全性问题 chain-of-thought
7 Can Large Language Models Really Recognize Your Name? 揭示大语言模型在识别个人姓名方面存在的系统性缺陷，并提出评估基准AMBENCH。 large language model
8 Guarded Query Routing for Large Language Models 提出GQR-Bench基准测试，并研究了LLM在安全查询路由中的有效性和效率。 large language model ✅
9 Toward Embodied AGI: A Review of Embodied AI and the Road Ahead 提出具身通用人工智能（AGI）分级框架，并展望高阶机器人大脑设计 embodied AI
10 Multimodal Mixture of Low-Rank Experts for Sentiment Analysis and Emotion Recognition 提出多模态低秩专家混合模型MMoLRE，用于情感分析和情绪识别的多任务学习。 multimodal
11 The Multimodal Information Based Speech Processing (MISP) 2025 Challenge: Audio-Visual Diarization and Recognition MISP 2025挑战赛：提出基于多模态信息的会议场景音视频说话人分离与识别方案 multimodal
12 Agent Context Protocols Enhance Collective Inference 提出Agent Context Protocols (ACPs)以增强多智能体系统的集体推理能力 generalist agent multimodal
13 MLZero: A Multi-Agent System for End-to-end Machine Learning Automation MLZero：基于LLM的多智能体系统，实现端到端多模态机器学习自动化 large language model multimodal
14 Reasoning Models Better Express Their Confidence 思维链推理模型能更准确地表达其置信度 large language model chain-of-thought
15 ProMind-LLM: Proactive Mental Health Care via Causal Reasoning with Sensor Data ProMind-LLM：利用因果推理与传感器数据实现主动式心理健康护理 large language model chain-of-thought
16 DrugPilot: LLM-based Parameterized Reasoning Agent for Drug Discovery DrugPilot：基于LLM的参数化推理Agent，用于药物发现 large language model multimodal
17 Towards Embodied Cognition in Robots via Spatially Grounded Synthetic Worlds 提出基于空间感知的合成数据集，用于训练机器人视觉语言模型以实现具身认知。 embodied AI
18 Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training RICE：无需额外训练，仅需两位专家即可引导MoE推理模型进行更有效的思考 instruction following
19 JARVIS: A Multi-Agent Code Assistant for High-Quality EDA Script Generation JARVIS：用于高质量EDA脚本生成的多智能体代码助手 large language model
20 SATBench: Benchmarking LLMs' Logical Reasoning via Automated Puzzle Generation from SAT Formulas SATBench：通过SAT公式自动生成逻辑谜题，评估LLM的逻辑推理能力 large language model ✅
21 Balanced and Elastic End-to-end Training of Dynamic LLMs 提出DynMo，实现动态LLM训练的负载均衡与弹性伸缩，提升训练效率。 large language model
22 ContextAgent: Context-Aware Proactive LLM Agents with Open-World Sensory Perceptions ContextAgent：提出一种利用开放世界感知的上下文感知主动LLM Agent。 large language model ✅
23 Cost-Awareness in Tree-Search LLM Planning: A Systematic Study 系统研究树搜索LLM规划器的成本意识，揭示其在资源约束下的规划能力不足 large language model
24 From nuclear safety to LLM security: Applying non-probabilistic risk management strategies to build safe and secure LLM-powered systems 借鉴核安全等领域经验，提出非概率风险管理策略以提升LLM系统安全性 large language model
25 Towards Reliable Proof Generation with LLMs: A Neuro-Symbolic Approach 提出神经符号方法，提升LLM在几何证明生成中的可靠性 large language model
26 Choosing a Model, Shaping a Future: Comparing LLM Perspectives on Sustainability and its Relationship with AI 评估LLM在可持续性问题上的偏差，揭示模型选择对组织策略的影响 large language model
27 Knowledge Graph Based Repository-Level Code Generation 提出基于知识图谱的代码仓库级代码生成方法，提升上下文准确性。 large language model
28 SCAN: Semantic Document Layout Analysis for Textual and Visual Retrieval-Augmented Generation 提出SCAN：一种语义文档布局分析方法，提升文本和视觉RAG性能 large language model
29 Divide by Question, Conquer by Agent: SPLIT-RAG with Question-Driven Graph Partitioning 提出SPLIT-RAG，通过问题驱动的图划分提升大规模知识图谱上的RAG效率与准确性。 large language model
30 RAG/LLM Augmented Switching Driven Polymorphic Metaheuristic Framework 提出RAG/LLM增强的切换驱动多态元启发式框架，提升复杂优化问题求解效率。 multimodal
31 LLM-based Evaluation Policy Extraction for Ecological Modeling 提出基于LLM的生态建模评估策略提取框架，提升模型评估的解释性和自动化程度 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (10 篇)

#题目一句话要点标签🔗⭐
32 Reinforcement Learning from User Feedback 提出RLUF框架，利用用户隐式反馈直接对齐LLM，提升用户满意度。 reinforcement learning RLHF large language model
33 Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoning 对比强化学习与蒸馏，揭示LLM推理能力提升的差异化机制 reinforcement learning distillation
34 Causal Cartographer: From Mapping to Reasoning Over Counterfactual Worlds Causal Cartographer：构建因果知识图谱，提升LLM在反事实推理中的鲁棒性 world model large language model foundation model
35 DSMentor: Enhancing Data Science Agents with Curriculum Learning and Online Knowledge Accumulation DSMentor：利用课程学习和在线知识积累增强数据科学Agent能力 curriculum learning large language model
36 SHARP: Synthesizing High-quality Aligned Reasoning Problems for Large Reasoning Models Reinforcement Learning SHARP：合成高质量对齐推理问题，用于强化学习训练大型推理模型 reinforcement learning chain-of-thought
37 RL of Thoughts: Navigating LLM Reasoning with Inference-time Reinforcement Learning RLoT：利用推理时强化学习引导LLM进行复杂推理 reinforcement learning large language model
38 Visual Instruction Bottleneck Tuning 提出Visual Instruction Bottleneck Tuning (Vittle)，提升多模态大语言模型在分布偏移下的泛化性和鲁棒性。 representation learning large language model multimodal
39 Embedded Mean Field Reinforcement Learning for Perimeter-defense Game 提出嵌入式均值场强化学习框架EMFAC，解决复杂三维环境下的无人机防御问题。 reinforcement learning representation learning
40 PRL: Prompts from Reinforcement Learning 提出基于强化学习的提示生成方法PRL，提升大语言模型在多任务上的性能。 reinforcement learning ✅
41 TelePlanNet: An AI-Driven Framework for Efficient Telecom Network Planning TelePlanNet：AI驱动的电信网络高效规划框架，提升基站选址一致性。 reinforcement learning large language model

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
42 Self-Evolving Curriculum for LLM Reasoning 提出自适应课程学习（SEC）方法，提升LLM在推理任务中强化学习微调的性能。 dual-arm reinforcement learning curriculum learning
43 EVA: Red-Teaming GUI Agents via Evolving Indirect Prompt Injection EVA：通过演化间接提示注入进行GUI代理的红队测试 manipulation multimodal
44 Transductively Informed Inductive Program Synthesis 提出TIIPS框架，通过协同机制融合归纳与转导推理，提升程序合成的准确性和泛化性。 manipulation

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

#题目一句话要点标签🔗⭐
45 AudioJailbreak: Jailbreak Attacks against End-to-End Large Audio-Language Models 提出AudioJailbreak，一种针对端到端大型音频语言模型的异步、通用、隐蔽且具有空中鲁棒性的对抗性攻击。 PULSE ✅

⬅️ 返回 cs.AI 首页 · 🏠 返回主页