cs.AI（2026-05-12）

📊 共 63 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) (41 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (15) 支柱一：机器人控制 (Robot Control) (4) 支柱四：生成式动作 (Generative Motion) (2) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (41 篇)

#题目一句话要点标签🔗⭐
1 SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models SafeSteer：一种多模态大语言模型解码层面的防御机制 large language model multimodal
2 MM-OptBench: A Solver-Grounded Benchmark for Multimodal Optimization Modeling 提出MM-OptBench，用于评估多模态优化建模中语言模型生成优化模型和代码的能力。 large language model multimodal
3 On the Limitations of Large Language Models for Conceptual Database Modeling 分析大型语言模型在概念数据库建模中的局限性 large language model chain-of-thought
4 Measuring What Matters Beyond Text: Evaluating Multimodal Summaries by Quality, Alignment, and Diversity 提出MM-Eval：一个综合评估多模态摘要质量、对齐性和多样性的统一框架 large language model multimodal
5 OOM-Free Alpamayo via CPU-GPU Memory Swapping for Vision-Language-Action Models 提出OOM-Free Alpamayo框架，通过CPU-GPU内存交换实现VLA模型在低显存GPU上的高效推理。 vision-language-action VLA
6 Very Efficient Listwise Multimodal Reranking for Long Documents ZipRerank：高效列表式多模态重排序，加速长文档检索与生成。 multimodal ✅
7 $δ$-mem: Efficient Online Memory for Large Language Models 提出$δ$-mem，通过高效在线记忆增强大语言模型处理长程依赖任务的能力。 large language model
8 Towards Automated Air Traffic Safety Assessment Around Non-Towered Airports Using Large Language Models 提出基于大语言模型的非塔台机场空域安全自动评估框架 large language model
9 Large Language Models as Amortized Pareto-Front Generators for Constrained Bi-Objective Convex Optimization DIPS：利用大语言模型生成约束双目标凸优化的Pareto前沿 large language model
10 OmniRefine: Alignment-Aware Cooperative Compression for Efficient Omnimodal Large Language Models 提出OmniRefine，用于高效压缩Omni-LLM中的音视频token，提升推理效率。 large language model
11 Automated Reformulation of Robust Optimization via Memory-Augmented Large Language Models 提出AutoREM，利用记忆增强大语言模型自动重构鲁棒优化问题。 large language model
12 When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel 揭示思维链推理过程中的不一致性：推理轨迹并非完美监督通道 chain-of-thought
13 A CAP-like Trilemma for Large Language Models: Correctness, Non-bias, and Utility under Semantic Underdetermination 针对大语言模型，提出语义欠定下的正确性、非偏见性和效用性三难困境 large language model
14 Allegory of the Cave: Measurement-Grounded Vision-Language Learning 提出PRISM-VL，通过测量域视觉语言学习提升多模态推理能力 multimodal
15 SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces SkillSafetyBench：评估技能驱动攻击下Agent的安全性 large language model
16 Formalize, Don't Optimize: The Heuristic Trap in LLM-Generated Combinatorial Solvers 形式化而非优化：LLM生成组合求解器中的启发式陷阱 large language model
17 LISA: Cognitive Arbitration for Signal-Free Autonomous Intersection Management LISA：基于LLM的认知仲裁无信号灯自动交叉口管理 large language model
18 Iterative Audit Convergence in LLM-Managed Multi-Agent Systems: A Case Study in Prompt Engineering Quality Assurance 提出基于LLM驱动的多Agent迭代审计方法，用于提升Prompt工程质量保障。 large language model
19 How Useful Is Cross-Domain Generalization for Training LLM Monitors? 研究跨领域泛化能力对训练LLM监控器的有效性 instruction following
20 Reconnecting Fragmented Citation Networks with Semantic Augmentation 提出基于语义增强的混合框架，用于补全碎片化的引文网络。 large language model
21 No Action Without a NOD: A Heterogeneous Multi-Agent Architecture for Reliable Service Agents 提出NOD异构多Agent架构，提升服务Agent在长程任务中的可靠性 large language model
22 Uncertainty Quantification for LLM-based Code Generation RisCoSet：利用多重假设检验为LLM代码生成构建风险可控的预测集合 large language model
23 BoolXLLM: LLM-Assisted Explainability for Boolean Models BoolXLLM：提出一种LLM辅助的布尔模型可解释性框架 large language model
24 Intermediate Artifacts as First-Class Citizens: A Data Model for Durable Intermediate Artifacts in Agentic Systems 提出一种数据模型，用于在Agent系统中持久化中间产物，提升可维护性。 chain-of-thought
25 SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory SAGE：一种自进化Agentic图记忆引擎，用于结构化关联记忆 foundation model
26 LLMs and the ZPD 基于维果茨基ZPD理论，探索LLM的“原始思维”模式与交互的重要性 large language model
27 LegalCheck: Retrieval- and Context-Augmented Generation for Drafting Municipal Legal Advice Letters LegalCheck：结合检索与上下文增强生成，辅助起草市政法律咨询函 large language model
28 CR^2: Cost-Aware Risk-Controlled Routing for Wireless Device-Edge LLM Inference 提出CR^2框架，解决无线边缘设备上LLM推理的成本感知风险控制路由问题 large language model
29 BadSKP: Backdoor Attacks on Knowledge Graph-Enhanced LLMs with Soft Prompts 提出BadSKP，针对知识图谱增强LLM的软提示后门攻击方法 large language model
30 Counterfactual Trace Auditing of LLM Agent Skills 提出CTA框架，通过反事实轨迹审计评估LLM Agent技能的影响，揭示现有评估方法的局限性。 large language model
31 Domain Restriction via Multi SAE Layer Transitions 利用多层稀疏自编码器转换进行领域限制，解决大语言模型领域外交互问题。 large language model
32 Rethinking Supervision Granularity: Segment-Level Learning for LLM-Based Theorem Proving 提出段级监督以解决大语言模型定理证明中的训练数据构建问题 large language model ✅
33 From Clever Hans to Scientific Discovery: Interpreting EEG Foundational Transformers with LRP 利用LRP解释EEG Transformer，揭示脑电信号中隐藏的行为模式与生物学假设 foundation model
34 Why Users Go There: World Knowledge-Augmented Generative Next POI Recommendation 提出AWARE，利用LLM增强POI推荐，解决模型知识固化问题。 large language model
35 OptArgus: A Multi-Agent System to Detect Hallucinations in LLM-based Optimization Modeling 提出OptArgus以解决LLM优化建模中的幻觉检测问题 large language model
36 Toward Stable Value Alignment: Introducing Independent Modules for Consistent Value Guidance 提出SVGT，通过独立价值模块实现大语言模型稳定价值观对齐 large language model ✅
37 Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion 提出MORA，通过扩展奖励维度打破大语言模型安全性-有用性瓶颈 large language model
38 Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermark 提出一种基于二项式编码的多比特LLM水印方案，提升消息准确性和鲁棒性。 TAMP
39 GAR: Carbon-Aware Routing for LLM Inference via Constrained Optimization 提出GAR：通过约束优化实现LLM推理的碳感知路由 large language model
40 Controllable User Simulation 提出因果一致的可控用户模拟器，解决对话Agent离线评估中的偏差问题。 large language model
41 AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive 提出AutoLLMResearch以解决高成本LLM实验配置问题 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (15 篇)

#题目一句话要点标签🔗⭐
42 Reinforcing VLAs in Task-Agnostic World Models RAW-Dream：在任务无关世界模型中强化VLA，实现零样本任务泛化。 reinforcement learning world model world models
43 Pretraining Strategies and Scaling for ECG Foundation Models: A Systematic Study 心电图(ECG)基础模型：预训练策略、规模化与模型架构的系统性研究 JEPA state space model representation learning
44 TOPPO: Rethinking PPO for Multi-Task Reinforcement Learning with Critic Balancing TOPPO：通过平衡Critic优化多任务强化学习中的PPO算法 reinforcement learning PPO SAC
45 EHR-RAGp: Retrieval-Augmented Prototype-Guided Foundation Model for Electronic Health Records EHR-RAGp：一种检索增强的原型引导电子病历基础模型 predictive model foundation model
46 Why Conclusions Diverge from the Same Observations: Formalizing World-Model Non-Identifiability via an Inference 形式化世界模型非唯一性，解释相同观察得出不同结论的现象 world model world models representation learning
47 High-lift Wing Separation Control via Bayesian Optimization and Deep Reinforcement Learning 通过贝叶斯优化和深度强化学习实现高升力翼分离控制 reinforcement learning deep reinforcement learning DRL
48 Towards Visually Grounded Multimodal Summarization via Cross-Modal Transformer and Gated Attention SPeCTrA-Sum：提出深度对齐跨模态Transformer和门控注意力机制，用于视觉引导的多模态摘要生成。 distillation multimodal
49 Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics 针对企业系统，提出基于运行时发现的智能体，提升动态环境适应性 world model world models
50 Hindsight Hint Distillation: Scaffolded Reasoning for SWE Agents from CoT-free Answers 提出HHD，利用无CoT问答对为软件工程Agent构建脚手架式推理，提升长程任务性能。 distillation chain-of-thought
51 Semantic Reward Collapse and the Preservation of Epistemic Integrity in Adaptive AI Systems 提出语义奖励坍塌概念，并设计宪法奖励分层框架，旨在提升自适应AI系统的认知完整性。 reinforcement learning RLHF large language model
52 Minimax Rates and Spectral Distillation for Tree Ensembles 提出基于谱分析的树集成模型压缩方法，实现资源受限场景下的高性能预测。 distillation
53 QAP-Router: Tackling Qubit Routing as Dynamic Quadratic Assignment with Reinforcement Learning QAP-Router：利用强化学习和动态二次分配解决量子比特路由问题 reinforcement learning
54 Assessing and Mitigating Miscalibration in LLM-Based Social Science Measurement 提出软标签蒸馏校准方法，提升LLM在社会科学测量中的置信度可靠性 distillation large language model
55 Modulation Consistency-based Contrastive Learning for Self-Supervised Automatic Modulation Classification 提出基于调制一致性的对比学习框架Mod-CL，用于自监督自动调制分类。 contrastive learning
56 Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning 提出自适应教师暴露的自蒸馏方法ATESD，提升LLM推理能力 distillation

🔬 支柱一：机器人控制 (Robot Control) (4 篇)

#题目一句话要点标签🔗⭐
57 Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models 提出MCF-Proto以解决VLA模型动作头同质化问题 manipulation vision-language-action VLA
58 When Simulation Lies: A Sim-to-Real Benchmark and Domain-Randomized RL Recipe for Tool-Use Agents 提出RobustBench-TC基准与ToolRL-DR方法，提升工具使用Agent在噪声环境下的鲁棒性 sim-to-real domain randomization reinforcement learning
59 Beyond Inefficiency: Systemic Costs of Incivility in Multi-Agent Monte Carlo Simulations 利用LLM多智能体模拟揭示不文明行为对效率的系统性影响 manipulation large language model
60 Trade-offs in Decentralized Agentic AI Discovery Across the Compute Continuum 研究分布式Agentic AI在计算连续体中基于DHT的发现机制的性能权衡 shared control

🔬 支柱四：生成式动作 (Generative Motion) (2 篇)

#题目一句话要点标签🔗⭐
61 Cochise: A Reference Harness for Autonomous Penetration Testing Cochise：用于自主渗透测试的参考框架，促进模型、架构和轨迹对比研究。 penetration
62 Crash Assessment via Mesh-Based Graph Neural Networks and Physics-Aware Attention 提出基于网格图神经网络和物理感知注意力机制的混合代理模型，用于快速车辆碰撞评估。 contact-aware

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

#题目一句话要点标签🔗⭐
63 Self-organized MT Direction Maps Emerge from Spatiotemporal Contrastive Optimization 提出时空对比优化方法，模拟灵长类MT区方向选择性图的自组织涌现 spatiotemporal

⬅️ 返回 cs.AI 首页 · 🏠 返回主页