cs.AI（2026-05-08）

📊 共 29 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (19 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (8 🔗1) 支柱一：机器人控制 (Robot Control) (1 🔗1) 支柱四：生成式动作 (Generative Motion) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (19 篇)

#题目一句话要点标签🔗⭐
1 Efficient Data Selection for Multimodal Models via Incremental Optimization Utility 提出One-Step-Train (OST)框架，通过增量优化效用评估实现多模态模型高效数据筛选 multimodal
2 Vaporizer: Breaking Watermarking Schemes for Large Language Model Outputs 提出Vaporizer攻击框架，通过语义保持的文本修改有效破解大语言模型水印方案 large language model
3 Abductive Reasoning with Probabilistic Commonsense 提出概率性溯因常识推理框架PACS，通过集成LLM与形式化求解器解决常识认知差异问题。 large language model chain-of-thought
4 Do Joint Audio-Video Generation Models Understand Physics? 提出AV-Phys Bench基准与AV-Phys Agent评估框架，揭示音视频生成模型在物理常识理解上的局限性。 multimodal
5 MathlibPR: Pull Request Merge-Readiness Benchmark for Formal Mathematical Libraries 提出MathlibPR基准测试，旨在利用大语言模型辅助形式化数学库的Pull Request评审 large language model
6 TraceFix: Repairing Agent Coordination Protocols with TLA+ Counterexamples 提出TraceFix框架：利用TLA+模型检测与反例反馈实现多智能体协作协议的自动修复与验证 large language model
7 CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios 提出CyBiasBench基准测试，揭示大模型智能体在网络攻击场景中的选择性偏见与行为模式 large language model ✅
8 RuleSafe-VL: Evaluating Rule-Conditioned Decision Reasoning in Vision-Language Content Moderation 提出RuleSafe-VL基准，通过规则条件化决策推理评估多模态内容审核能力 multimodal
9 LLM hallucinations in the wild: Large-scale evidence from non-existent citations 大规模实证研究揭示：大语言模型幻觉正导致科学文献中虚假引用激增 large language model
10 The AI-Native Large-Scale Agile Software Development Manifesto 提出AI原生大规模敏捷软件开发宣言，通过AI代理重塑组织级开发范式 large language model
11 Hierarchical Task Network Planning with LLM-Generated Heuristics 利用LLM生成启发式函数，提升分层任务网络规划效率 large language model
12 GASim: A Graph-Accelerated Hybrid Framework for Social Simulation 提出GASim图加速混合框架，通过图神经网络优化大规模社会模拟的计算效率 large language model ✅
13 LARAG: Link-Aware Retrieval Strategy for RAG Systems in Hyperlinked Technical Documentation 提出LARAG检索策略，通过利用超链接拓扑结构提升技术文档RAG系统的检索准确性与效率。 large language model
14 Prompt Engineering Strategies for LLM-based Qualitative Coding of Psychological Safety in Software Engineering Communities: A Controlled Empirical Study 评估LLM在软件工程心理安全定性编码中的提示工程策略，揭示模型稳定性与偏差规律 large language model
15 GraphReAct: Reasoning and Acting for Multi-step Graph Inference 提出GraphReAct框架，通过推理与行动交替机制实现图结构数据的多步推理 large language model
16 SOM: Structured Opponent Modeling for LLM-based Agents via Structural Causal Model 提出结构化对手建模（SOM）框架，利用结构因果模型提升LLM智能体在多智能体环境下的预测能力。 large language model
17 HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization 提出HMACE异构多智能体协作进化框架，以解决组合优化问题中的启发式算法自动设计难题。 large language model
18 ARMOR: An Agentic Framework for Reaction Feasibility Prediction via Adaptive Utility-aware Multi-tool Reasoning 提出ARMOR代理框架，通过自适应效用感知多工具推理解决化学反应可行性预测难题 large language model
19 2.5-D Decomposition for LLM-Based Spatial Construction 提出2.5-D分解方法，通过解耦空间维度提升大模型在自主构建任务中的空间推理精度 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (8 篇)

#题目一句话要点标签🔗⭐
20 Structured Role-Aware Policy Optimization for Multimodal Reasoning 提出结构化角色感知策略优化SRPO，提升多模态推理中证据可靠性。 reinforcement learning multimodal
21 DCGL: Dual-Channel Graph Learning with Large Language Models for Knowledge-Aware Recommendation 提出双通道图学习框架DCGL，通过解耦语义与行为模式提升知识感知推荐性能 contrastive learning large language model ✅
22 Three-in-One World Model: Energy-Based Consistency, Prediction, and Counterfactual Inference for Marketing Intervention 提出“三合一”世界模型，通过深度玻尔兹曼机实现营销干预中的一致性评估、预测与反事实推理。 world model world models
23 LiteGUI: Distilling Compact GUI Agents with Reinforcement Learning 提出LiteGUI框架：通过强化学习与多解蒸馏提升轻量级GUI智能体性能 reinforcement learning imitation learning distillation
24 Tools as Continuous Flow for Evolving Agentic Reasoning 提出FlowAgent框架：利用条件流匹配实现连续轨迹生成，解决智能体长程推理中的误差累积问题。 flow matching large language model
25 Dependence on Early and Late Reverberation of Single-Channel Speaker Distance Estimation 揭示单通道说话人距离估计中早期反射与后期混响的依赖机制 MAE PULSE
26 Learning to Communicate Locally for Large-Scale Multi-Agent Pathfinding 提出LC-MAPF框架，通过局部通信机制提升大规模多智能体路径规划的协作效率 reinforcement learning imitation learning
27 Offline Policy Optimization with Posterior Sampling 提出基于后验采样的离线策略优化方法（PSPO），平衡模型泛化与鲁棒性 reinforcement learning offline reinforcement learning

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
28 Reason to Play: Behavioral and Brain Alignment Between Frontier LRMs and Human Game Learners 评估前沿大推理模型（LRMs）在复杂游戏学习中的行为与神经对齐机制 manipulation reinforcement learning deep reinforcement learning ✅

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
29 BeeVe: Unsupervised Acoustic State Discovery in Honey Bee Buzzing 提出BeeVe无监督学习框架，通过VQ-VAE实现蜜蜂蜂鸣声的声学状态自动发现 VQ-VAE

⬅️ 返回 cs.AI 首页 · 🏠 返回主页