cs.AI（2026-04-16）

📊 共 51 篇论文 | 🔗 6 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (33 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (11 🔗3) 支柱一：机器人控制 (Robot Control) (3 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (2) 支柱三：空间感知与语义 (Perception & Semantics) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (33 篇)

#题目一句话要点标签🔗⭐
1 Predicting Power-System Dynamic Trajectories with Foundation Models 提出LASS-ODE-Power，利用大规模预训练预测电力系统动态轨迹。 foundation model
2 Intermediate Layers Encode Optimal Biological Representations in Single-Cell Foundation Models 单细胞Foundation模型中，中间层编码了最优的生物学表征，超越了传统末层特征提取方法。 foundation model
3 Dissecting Failure Dynamics in Large Language Model Reasoning 提出GUARD框架，通过不确定性信号探测并纠正大语言模型推理过程中的早期错误。 large language model
4 Rethinking Patient Education as Multi-turn Multi-modal Interaction 提出MedImageEdu基准，用于评估多模态交互式患者教育智能体 multimodal visual grounding
5 MirrorBench: Evaluating Self-centric Intelligence in MLLMs by Introducing a Mirror MirrorBench：通过引入镜像评估多模态大语言模型中的自我中心智能 large language model multimodal ✅
6 DR$^{3}$-Eval: Towards Realistic and Reproducible Deep Research Evaluation 提出DR³-Eval：一个用于评估深度研究Agent的现实且可复现的基准 multimodal instruction following
7 Context Over Content: Exposing Evaluation Faking in Automated Judges 揭示LLM评估中的情境偏见：下游影响信息会扭曲评估结果 chain-of-thought
8 AIPC: Agent-Based Automation for AI Model Deployment with Qualcomm AI Runtime AIPC：基于Agent的AI模型自动化部署框架，加速高通AI Runtime部署。 multimodal
9 VeriGraphi: A Multi-Agent Framework of Hierarchical RTL Generation for Large Hardware Designs 提出VeriGraphi框架，解决LLM生成大型分层硬件设计Verilog代码的难题 large language model
10 Scepsy: Serving Agentic Workflows Using Aggregate LLM Pipelines Scepsy：利用聚合LLM流水线服务Agentic工作流，提升吞吐量并降低延迟 large language model
11 Autonomous Evolution of EDA Tools: Multi-Agent Self-Evolved ABC 提出自我进化的逻辑综合框架以提升EDA工具性能 large language model
12 From Reactive to Proactive: Assessing the Proactivity of Voice Agents via ProVoice-Bench 提出ProVoice-Bench，用于评估主动式语音代理，填补现有基准测试的空白。 multimodal
13 Dr.~RTL: Autonomous Agentic RTL Optimization through Tool-Grounded Self-Improvement Dr.~RTL：通过工具驱动的自主Agent持续优化RTL设计 large language model
14 Discovering Novel LLM Experts via Task-Capability Coevolution 提出AC/DC框架，通过任务-能力协同进化发现具备新技能且更高效的LLM。 large language model
15 Governing Reflective Human-AI Collaboration: A Framework for Epistemic Scaffolding and Traceable Reasoning 提出一种人机协作框架，通过知识支架和可追溯推理提升AI治理能力 large language model
16 MemoSight: Unifying Context Compression and Multi Token Prediction for Reasoning Acceleration MemoSight：融合上下文压缩与多Token预测，加速LLM推理 chain-of-thought
17 The Missing Knowledge Layer in AI: A Framework for Stable Human-AI Reasoning 提出稳定人机推理框架，解决大语言模型推理漂移问题 large language model
18 The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows 揭示LLM谬误：AI辅助认知工作流中的能力误判现象 large language model
19 Bounded Autonomy for Enterprise AI: Typed Action Contracts and Consumer-Side Execution 提出面向企业AI的Bounded Autonomy架构，保障LLM安全执行并提升效率。 large language model
20 The Agentification of Scientific Research: A Physicist's Perspective AI Agent赋能科研：从工具到合作者，重塑科学研究范式 large language model
21 HWE-Bench: Benchmarking LLM Agents on Real-World Hardware Bug Repair Tasks HWE-Bench：首个面向真实硬件缺陷修复任务的大规模LLM Agent基准测试 large language model
22 Learning to Draw ASCII Improves Spatial Reasoning in Language Models 通过学习绘制ASCII图提升语言模型空间推理能力 large language model
23 El Agente Forjador: Task-Driven Agent Generation for Quantum Simulation El Agente Forjador：面向量子模拟的任务驱动型智能体生成框架 large language model
24 GDPR Auto-Formalization with AI Agents and Human Verification 提出基于AI Agent和人工验证的GDPR自动形式化框架，提升法律文本处理质量 large language model
25 Quantifying Cross-Query Contradictions in Multi-Query LLM Reasoning 提出一种基于求解器增强的多查询LLM推理方法，解决跨查询逻辑矛盾问题 large language model
26 Analyzing Chain of Thought (CoT) Approaches in Control Flow Code Deobfuscation Tasks 提出基于思维链（CoT）提示的大语言模型代码去混淆方法，提升控制流恢复和语义保持。 large language model chain-of-thought
27 LLMbench: A Comparative Close Reading Workbench for Large Language Models LLMbench：用于大语言模型比较性细读的浏览器工作台 large language model
28 Ragged Paged Attention: A High-Performance and Flexible LLM Inference Kernel for TPU 提出Ragged Paged Attention，为TPU上的LLM推理提供高性能和灵活的内核。 large language model
29 LACE: Lattice Attention for Cross-thread Exploration 提出LACE框架以解决大语言模型推理孤立问题 large language model
30 The Semi-Executable Stack: Agentic Software Engineering and the Expanding Scope of SE 提出半可执行栈模型，应对AI驱动下软件工程范畴扩展至半可执行工件的挑战。 foundation model
31 The Crutch or the Ceiling? How Different Generations of LLMs Shape EFL Student Writings 研究不同代LLM对EFL学生写作的影响：支柱还是天花板？ large language model
32 HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents? 提出HarmfulSkillBench，评估LLM智能体在恶意技能环境下的安全性 large language model ✅
33 Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation 提出基于LLM的Verilog代码生成方法，通过数据高效微调和自动化测试平台提升性能。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (11 篇)

#题目一句话要点标签🔗⭐
34 CoTEvol: Self-Evolving Chain-of-Thoughts for Data Synthesis in Mathematical Reasoning 提出CoTEvol，通过自进化思维链进行数学推理数据合成 distillation large language model chain-of-thought
35 Disentangle-then-Refine: LLM-Guided Decoupling and Structure-Aware Refinement for Graph Contrastive Learning 提出SDM-SCR框架以解决图对比学习中的信号与噪声纠缠问题 contrastive learning large language model instruction following
36 Towards Faster Language Model Inference Using Mixture-of-Experts Flow Matching 提出MoE-FM框架，加速非自回归语言模型推理，显著提升效率。 flow matching Mamba multimodal
37 Targeted Exploration via Unified Entropy Control for Reinforcement Learning 提出UEC-RL，通过统一熵控制解决强化学习中探索不足和策略坍塌问题 reinforcement learning large language model ✅
38 MARS$^2$: Scaling Multi-Agent Tree Search via Reinforcement Learning for Code Generation 提出MARS$^2$，通过多智能体强化学习树搜索扩展代码生成能力。 reinforcement learning reward shaping ✅
39 IG-Search: Step-Level Information Gain Rewards for Search-Augmented Reasoning 提出IG-Search，利用信息增益奖励提升搜索增强推理能力 reinforcement learning large language model
40 Temporal Cross-Modal Knowledge-Distillation-Based Transfer-Learning for Gas Turbine Vibration Fault Detection 提出时序跨模态知识蒸馏迁移学习框架，用于燃气轮机振动故障检测。 distillation
41 Acceptance Dynamics Across Cognitive Domains in Speculative Decoding 研究推测解码中认知领域对接受率的影响，为领域自适应优化提供依据。 RLHF large language model
42 Asking What Matters: Reward-Driven Clarification for Software Engineering Tasks 提出CLARITI，通过奖励驱动的澄清问题生成，提升软件工程任务效率。 reinforcement learning reward design
43 Targeted Exploration via Unified Entropy Control for Reinforcement Learning 提出UEC-RL，通过统一熵控制解决强化学习中探索不足和策略坍塌问题 reinforcement learning large language model ✅
44 Subliminal Transfer of Unsafe Behaviors in AI Agent Distillation 揭示AI智能体蒸馏中，即使数据经过严格过滤，仍存在不安全行为的隐性传递现象 distillation

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
45 Hijacking Large Audio-Language Models via Context-Agnostic and Imperceptible Auditory Prompt Injection 提出AudioHijack框架，实现对大型语音语言模型在不可察觉的音频提示注入攻击 manipulation
46 CSLE: A Reinforcement Learning Platform for Autonomous Security Management CSLE：用于自主安全管理的强化学习平台，弥合仿真与实际系统差距 recovery control reinforcement learning
47 SecureRouter: Encrypted Routing for Efficient Secure Inference SecureRouter：面向高效安全推理的加密路由框架 MPC ✅

🔬 支柱六：视频提取与匹配 (Video Extraction) (2 篇)

#题目一句话要点标签🔗⭐
48 Learning to Think Like a Cartoon Captionist: Incongruity-Resolution Supervision for Multimodal Humor Understanding 提出IRS框架，通过不协调-解决监督提升多模态幽默理解能力 HuMoR multimodal zero-shot transfer
49 GIST: Multimodal Knowledge Extraction and Spatial Grounding via Intelligent Semantic Topology GIST：通过智能语义拓扑实现多模态知识提取与空间定位 egocentric embodied AI multimodal

🔬 支柱三：空间感知与语义 (Perception & Semantics) (1 篇)

#题目一句话要点标签🔗⭐
50 ADAPT: Benchmarking Commonsense Planning under Unspecified Affordance Constraints ADAPT：针对未明确可供性约束的常识规划基准测试 affordance

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

#题目一句话要点标签🔗⭐
51 Sequence Search: Automated Sequence Design using Neural Architecture Search 提出基于神经架构搜索的Sequence Search，实现磁共振序列的自动设计。 PULSE

⬅️ 返回 cs.AI 首页 · 🏠 返回主页