cs.AI（2026-03-30）

📊 共 28 篇论文 | 🔗 1 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (21 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (4) 支柱一：机器人控制 (Robot Control) (1) 支柱五：交互与反应 (Interaction & Reaction) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (21 篇)

#题目一句话要点标签🔗⭐
1 MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models 提出MonitorBench，用于全面评估大语言模型中思维链的可监控性 large language model chain-of-thought
2 PReD: An LLM-based Foundation Multimodal Model for Electromagnetic Perception, Recognition, and Decision 提出PReD：首个电磁领域多模态大模型，实现感知、识别与决策闭环 large language model foundation model multimodal
3 CARV: A Diagnostic Benchmark for Compositional Analogical Reasoning in Multimodal LLMs CARV：多模态LLM中组合类比推理的诊断基准 large language model multimodal
4 The Scaffold Effect: How Prompt Framing Drives Apparent Multimodal Gains in Clinical VLM Evaluation 提出框架效应以解决临床VLM评估中的多模态表现问题 multimodal
5 MiroEval: Benchmarking Multimodal Deep Research Agents in Process and Outcome MiroEval：面向多模态深度研究Agent的过程与结果评测基准 multimodal
6 COvolve: Adversarial Co-Evolution of Large-Language-Model-Generated Policies and Environments via Two-Player Zero-Sum Game COvolve：通过零和博弈对抗协同进化LLM生成策略与环境，实现开放式学习。 large language model
7 Deep Research of Deep Research: From Transformer to Agent, From AI to AI for Science 构建Transformer到Agent的演进路线图，探索AI在科学研究中的应用 large language model multimodal
8 HeteroHub: An Applicable Data Management Framework for Heterogeneous Multi-Embodied Agent System HeteroHub：异构多具身智能体系统的数据管理框架 embodied AI multimodal
9 A Multi-Agent Rhizomatic Pipeline for Non-Linear Literature Analysis 提出基于多智能体Rhizomatic流程的非线性文献分析方法，突破传统线性综述局限。 large language model
10 Beyond the Answer: Decoding the Behavior of LLMs as Scientific Reasoners 利用GEPA优化提示词，揭示LLM在科学推理中的行为模式 large language model
11 SAGAI-MID: A Generative AI-Driven Middleware for Dynamic Runtime Interoperability SAGAI-MID：利用生成式AI中间件实现动态运行时互操作性 large language model
12 The Ultimate Tutorial for AI-driven Scale Development in Generative Psychometrics: Releasing AIGENIE from its Bottle 提出AIGENIE框架以自动化心理测量量表开发流程 large language model
13 Moving Beyond Review: Applying Language Models to Planning and Translation in Reflection Pensée：利用语言模型在反思写作的规划和翻译阶段提供支持，提升反思深度和质量。 large language model
14 Coherent Without Grounding, Grounded Without Success: Observability and Epistemic Failure 揭示大语言模型在可观测性差异下的能力与解释错位现象 large language model
15 Evaluating LLMs for Answering Student Questions in Introductory Programming Courses 评估LLM在编程入门课程中回答学生问题的能力，并提出评估框架。 large language model
16 Reasoning as Energy Minimization over Structured Latent Trajectories 提出基于能量最小化的结构化隐空间轨迹推理方法，解决单步解码和链式推理的不足。 chain-of-thought ✅
17 EpiPersona: Persona Projection and Episode Coupling for Pluralistic Preference Modeling EpiPersona：通过人物角色投影和情景耦合建模多元偏好 large language model
18 Designing AI for Real Users -- Accessibility Gaps in Retail AI Front-End 零售AI前端易用性设计缺陷：忽略残障用户体验，提出前端保障机制 multimodal
19 CoT2-Meta: Budgeted Metacognitive Control for Test-Time Reasoning CoT2-Meta：面向测试时推理的预算型元认知控制框架 chain-of-thought
20 ViviDoc: Generating Interactive Documents through Human-Agent Collaboration ViviDoc：提出一种人机协作框架，用于生成可交互文档，降低创作成本。 large language model
21 GEAKG: Generative Executable Algorithm Knowledge Graphs 提出GEAKG：一种生成式可执行算法知识图谱，实现跨领域算法知识的表示、学习与迁移。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (4 篇)

#题目一句话要点标签🔗⭐
22 Seeing with You: Perception-Reasoning Coevolution for Multimodal Reasoning 提出PRCO框架，通过感知-推理协同进化提升多模态推理能力 reinforcement learning large language model multimodal
23 SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology 提出SARL：通过奖励推理拓扑结构实现无标签强化学习，提升大语言模型的推理能力。 reinforcement learning PPO
24 FedFG: Privacy-Preserving and Robust Federated Learning via Flow-Matching Generation FedFG：基于流匹配生成器的隐私保护和鲁棒联邦学习框架 flow matching
25 Reward Hacking as Equilibrium under Finite Evaluation 揭示有限评估下奖励篡改是AI系统的结构性均衡，并提出可计算的篡改指标 RLHF DPO

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
26 Adversarial Attacks on Multimodal Large Language Models: A Comprehensive Survey 全面分析多模态大语言模型对抗攻击，揭示脆弱性根源并指导防御。 manipulation large language model multimodal

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

#题目一句话要点标签🔗⭐
27 Meta-Harness: End-to-End Optimization of Model Harnesses Meta-Harness：端到端优化LLM应用的代码框架，提升性能并降低token消耗。 IMoS large language model

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

#题目一句话要点标签🔗⭐
28 Differentiable Power-Flow Optimization 提出可微潮流计算(DPF)，解决电力系统仿真中传统方法扩展性差的问题。 differentiable simulation

⬅️ 返回 cs.AI 首页 · 🏠 返回主页