cs.AI（2026-04-28）

📊 共 31 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) (20 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (9) 支柱一：机器人控制 (Robot Control) (2)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (20 篇)

#题目一句话要点标签🔗⭐
1 Learning Generalizable Multimodal Representations for Software Vulnerability Detection 提出MultiVul多模态对比学习框架，提升软件漏洞检测的泛化性 large language model multimodal
2 Walking Through Uncertainty: An Empirical Study of Uncertainty Estimation for Audio-Aware Large Language Models 首个音频感知大语言模型不确定性估计的系统性实证研究 large language model
3 DualFact+: A Multimodal Fact Verification Framework for Procedural Video Understanding 提出DualFact+框架，用于程序视频理解中的多模态事实核查。 multimodal
4 From Insight to Action: A Novel Framework for Interpretability-Guided Data Selection in Large Language Models 提出IGDS框架，利用可解释性指导大语言模型的数据选择，提升模型性能。 large language model
5 From Soliloquy to Agora: Memory-Enhanced LLM Agents with Decentralized Debate for Optimization Modeling 提出Agora-Opt，利用去中心化辩论和记忆增强LLM Agent解决优化建模问题 large language model ✅
6 Making AI-Assisted Grant Evaluation Auditable without Exposing the Model 提出基于TEE的架构，在不暴露模型的前提下，实现AI辅助资助评估的可审计性。 large language model
7 Doing More With Less: Revisiting the Effectiveness of LLM Pruning for Test-Time Scaling 非结构化剪枝提升LLM在测试时计算扩展中的推理性能 large language model
8 Towards Agentic Investigation of Security Alerts 提出基于LLM的Agentic安全告警调查工作流，提升告警判定的准确性。 large language model
9 SAFEdit: Does Multi-Agent Decomposition Resolve the Reliability Challenges of Instructed Code Editing? SAFEdit：多智能体分解框架提升指令驱动代码编辑的可靠性 large language model
10 Think Before You Act -- A Neurocognitive Governance Model for Autonomous AI Agents 提出神经认知治理模型PAGRL，提升自主AI Agent在复杂环境下的安全性与合规性 large language model
11 HotComment: A Benchmark for Evaluating Popularity of Online Comments 提出HotComment基准，用于评估在线评论的受欢迎程度，并引入StyleCmt模型。 multimodal
12 The Nonverbal Syntax Framework: An Evidence-Based Tiered System for Inferring Learner States from Observable Behavioral Cues 提出非语言语法框架，通过可观察行为线索推断学习者状态 multimodal
13 Emotive Architectures: The Role of LLMs in Adjusting Work Environments 利用LLM构建情感感知工作环境，提升用户体验与福祉 large language model
14 SnapGuard: Lightweight Prompt Injection Detection for Screenshot-Based Web Agents SnapGuard：针对截图Web代理的轻量级Prompt注入检测方法 multimodal
15 Assistants, Not Architects: The Role of LLMs in Networked Systems Design 提出Kepler框架，解决LLM在网络系统架构设计中不可靠的问题 large language model
16 SciEval: A Benchmark for Automatic Evaluation of K-12 Science Instructional Materials SciEval：构建K-12科学教学材料自动评估基准，并验证领域微调的有效性。 large language model
17 AHASD: Asynchronous Heterogeneous Architecture for LLM Adaptive Drafting Speculative Decoding on Mobile Devices 提出AHASD以解决移动设备上LLM自适应草拟的效率问题 large language model
18 DATAREEL: Automated Data-Driven Video Story Generation with Animations DataReel：提出一个自动生成动画数据视频故事的基准和多智能体框架 large language model ✅
19 Where Did It Go Wrong? Capability-Oriented Failure Attribution for Vision-and-Language Navigation Agents 提出面向能力的测试方法，用于视觉-语言导航Agent的故障归因 VLN
20 Agentic Architect: An Agentic AI Framework for Architecture Design Exploration and Optimization Agentic Architect：基于Agentic AI的计算机体系结构设计探索与优化框架 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (9 篇)

#题目一句话要点标签🔗⭐
21 Three Models of RLHF Annotation: Extension, Evidence, and Authority 提出RLHF标注的三种模型，优化人类反馈强化学习流程 reinforcement learning RLHF large language model
22 How Can Reinforcement Learning Achieve Expert-level Placement? 提出基于专家布局学习的强化学习方法，提升芯片布局质量 reinforcement learning reward design
23 Semi-Markov Reinforcement Learning for City-Scale EV Ride-Hailing with Feasibility-Guaranteed Actions 提出基于半马尔可夫强化学习的城市级电动汽车网约车控制方法，保证动作可行性。 reinforcement learning SAC
24 Sample-efficient Neuro-symbolic Proximal Policy Optimization 提出神经符号近端策略优化，提升DRL在稀疏奖励和长规划任务中的样本效率 reinforcement learning deep reinforcement learning DRL
25 Improving Zero-Shot Offline RL via Behavioral Task Sampling 提出基于行为任务采样的离线零样本强化学习方法，提升泛化性能。 reinforcement learning offline RL
26 RADD: Retrieval-Augmented Discrete Diffusion for Multi-Modal Knowledge Graph Completion 提出RADD框架，解耦检索与重排序，提升多模态知识图谱补全性能。 distillation multimodal
27 JURY-RL: Votes Propose, Proofs Dispose for Label-Free RLVR JURY-RL：基于投票提议与形式化验证的无标签强化学习 reinforcement learning large language model
28 Multi-action Tangled Program Graphs for Multi-task Reinforcement Learning with Continuous Control 提出基于多动作缠结程序图的MATPG算法，用于连续控制多任务强化学习。 reinforcement learning
29 Evaluating Risks in Weak-to-Strong Alignment: A Bias-Variance Perspective 通过偏差-方差视角评估弱到强对齐中的风险，揭示强模型方差是欺骗性错误的早期预警信号。 reinforcement learning RLHF

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
30 Large language models eroding science understanding: an experimental study 大型语言模型易受伪科学影响，损害科学认知 manipulation large language model
31 PHISHREV: A Hybrid Machine Learning and Post-Hoc Non-monotonic Reasoning Framework for Context-Aware Phishing Website Classification 提出PHISHREV框架以解决网络钓鱼网站分类中的上下文推理问题 manipulation

⬅️ 返回 cs.AI 首页 · 🏠 返回主页