cs.AI（2026-05-18）

📊 共 37 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (24 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (10) 支柱一：机器人控制 (Robot Control) (1) 支柱六：视频提取与匹配 (Video Extraction) (1) 支柱四：生成式动作 (Generative Motion) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (24 篇)

#题目一句话要点标签🔗⭐
1 Qumus: Realization of An Embodied AI Quantum Material Experimentalist Qumus：实现具身AI量子材料实验家，首次AI创建石墨烯和纳米器件。 embodied AI large language model multimodal
2 SVFSearch: A Multimodal Knowledge-Intensive Benchmark for Short-Video Frame Search in the Gaming Vertical Domain 提出SVFSearch：一个面向游戏短视频帧搜索的多模态知识密集型基准 large language model multimodal visual grounding
3 Visualizing the Invisible: Generative Visual Grounding Empowers Universal EEG Understanding in MLLMs 提出生成式视觉 grounding (GVG) 框架，提升 MLLM 对脑电信号的理解能力 foundation model visual grounding
4 Safety Geometry Collapse in Multimodal LLMs and Adaptive Drift Correction 针对多模态LLM安全几何坍塌问题，提出自适应漂移校正方法ReGap large language model multimodal
5 Estimating Item Difficulty with Large Language Models as Experts 利用大型语言模型作为专家评估项目难度，无需响应数据。 large language model
6 TeleCom-Bench: How Far Are Large Language Models from Industrial Telecommunication Applications? TeleCom-Bench：评估大语言模型在工业电信应用中的能力差距，并提供领域对齐指导。 large language model ✅
7 TierCheck: Tiered Checkpointing for Fault Tolerance in Large Language Model Training TierCheck：面向大语言模型训练的异构容错分层检查点系统 large language model
8 DuIVRS-2: An LLM-based Interactive Voice Response System for Large-scale POI Attribute Acquisition DuIVRS-2：基于LLM的大规模POI属性获取交互式语音应答系统 large language model chain-of-thought
9 Evaluating Cognitive Age Alignment in Interactive AI Agents 提出ChildAgentEval，评估交互式AI智能体认知年龄对齐程度 large language model multimodal
10 Prompt2Fingerprint: Plug-and-Play LLM Fingerprinting via Text-to-Weight Generation 提出Prompt2Fingerprint，通过文本到权重的生成实现即插即用的LLM指纹识别。 large language model
11 Guard: Scalable Straggler Detection and Node Health Management for Large-Scale Training Guard：用于大规模训练的可扩展Straggler检测和节点健康管理系统 foundation model
12 Democratizing Large-Scale Re-Optimization with LLM-Guided Model Patches 提出基于LLM引导模型补丁的大规模重优化框架，赋能非专家用户。 large language model
13 SCICONVBENCH: Benchmarking LLMs on Multi-Turn Clarification for Task Formulation in Computational Science SCICONVBENCH：用于评估LLM在计算科学中多轮澄清任务构建能力的基准测试。 large language model ✅
14 Prompts Don't Protect: Architectural Enforcement via MCP Proxy for LLM Tool Access Control 提出MCP代理架构，通过强制访问控制保障LLM工具使用的安全性 large language model
15 QSTRBench: a New Benchmark to Evaluate the Ability of Language Models to Reason with Qualitative Spatial and Temporal Calculi 提出QSTRBench以评估语言模型的空间与时间推理能力 large language model
16 The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human-AI Collaboration 研究揭示LLM在人机协作中存在语境性谄媚问题，并探讨AI素养干预的有效性 large language model
17 Evidence-Grounded Frontier Mapping and Agentic Hypothesis Generation in Nanomedicine pArticleMap：一种基于证据的纳米医学前沿探索与假设生成系统 large language model
18 Generative AI and the Productivity Divide: Human-AI Complementarities in Education 研究表明，生成式AI在教育中生产力提升存在差异，AI交互能力是关键。 large language model
19 A-ProS: Towards Reliable Autonomous Programming Through Multi-Model Feedback A-ProS：通过多模型反馈实现可靠的自主编程 large language model
20 Babel: Jailbreaking Safety Attention via Obfuscation Distribution Optimized Sampling Babel：通过优化混淆分布采样破解安全注意力机制 large language model
21 Reconciling Contradictory Views on the Effectiveness of SFT in LLMs: An Interaction Perspective 基于交互视角，揭示SFT在LLM中效果不一致的原因并提供训练指导 large language model
22 BLAgent: Agentic RAG for File-Level Bug Localization BLAgent：面向文件级缺陷定位的Agentic RAG框架 large language model
23 Agentic Chunking and Bayesian De-chunking of AI Generated Fuzzy Cognitive Maps: A Model of the Thucydides Trap 提出基于LLM Agent的FCM自动构建与贝叶斯解耦方法，用于分析大国冲突。 large language model
24 Interactive Evaluation Requires a Design Science 设计科学视角下的交互式评估框架，应对LLM在复杂环境中的评估挑战。 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🔗⭐
25 DARE-EEG: A Foundation Model for Mining Dual-Aligned Representation of EEG DARE-EEG：通过双重对齐表征学习脑电图的通用基础模型 representation learning contrastive learning foundation model
26 Actionable World Representation 提出WorldString，统一建模可交互对象状态，构建可执行的世界表征。 policy learning world model world models
27 AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment 提出AMR-SD以解决大语言模型的信用分配瓶颈问题 reinforcement learning distillation large language model
28 SD-Search: On-Policy Hindsight Self-Distillation for Search-Augmented Reasoning SD-Search：基于On-Policy Hindsight Self-Distillation的搜索增强推理 reinforcement learning distillation
29 LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning 提出LLM引导的多智能体通信(LMAC)以提升协作式MARL性能 reinforcement learning
30 Latent Action Reparameterization for Efficient Agent Inference 提出Latent Action Reparameterization，提升LLM Agent推理效率 representation learning large language model
31 When Outcome Looks Right But Discipline Fails: Trace-Based Evaluation Under Hidden Competitor State 提出纪律稳定性评估以解决隐藏竞争状态下的经济安全问题 PPO behavior cloning
32 AdaptiveLoad: Towards Efficient Video Diffusion Transformer Training 提出AdaptiveLoad以解决视频扩散Transformer训练中的负载不均问题 world model world models
33 Scalable Environments Drive Generalizable Agents 提出环境尺度扩展方法，提升智能体在多样化环境中的泛化能力 world model world models
34 Learning to Solve Compositional Geometry Routing Problems 提出DiCon，解决组合几何路由问题中的复杂表征与决策挑战。 representation learning contrastive learning

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
35 Not What You Asked For: Typographic Attacks in Household Robot Manipulation 揭示家庭机器人操作中印刷文字攻击的风险：语义劫持导致物理操作失败 manipulation semantic map open-vocabulary

🔬 支柱六：视频提取与匹配 (Video Extraction) (1 篇)

#题目一句话要点标签🔗⭐
36 Beyond the Cartesian Illusion: Testing Two-Stage Multi-Modal Theory of Mind under Perceptual Bottlenecks 提出基于锚点的具身空间分解CoT，提升MLLM在感知瓶颈下的二阶ToM能力 egocentric embodied AI large language model

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
37 KISS - Knowledge Infrastructure for Scientific Simulation: A Scaffolding for Agentic Earth Science 提出KISS知识基础设施，赋能Agent自主执行地球科学模拟。 physically plausible

⬅️ 返回 cs.AI 首页 · 🏠 返回主页