cs.AI（2026-02-06）

📊 共 44 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) (27 🔗4) 支柱二：RL算法与架构 (RL & Architecture) (11) 支柱一：机器人控制 (Robot Control) (3) 支柱八：物理动画 (Physics-based Animation) (2 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (27 篇)

#题目一句话要点标签🔗⭐
1 Trifuse: Enhancing Attention-Based GUI Grounding via Multimodal Fusion Trifuse：通过多模态融合增强基于注意力的GUI元素定位 large language model multimodal
2 POP: Online Structural Pruning Enables Efficient Inference of Large Foundation Models POP：在线结构剪枝实现大模型高效推理，兼顾精度与速度 large language model foundation model
3 Federated Prompt-Tuning with Heterogeneous and Incomplete Multimodal Client Data 提出异构不完全多模态联邦Prompt Tuning框架，解决跨客户端数据缺失和语义对齐问题。 multimodal
4 Is there "Secret Sauce'' in Large Language Model Development? 大规模语言模型性能主要由算力驱动，但开发者效率差异显著影响非前沿模型 large language model
5 Sequences as Nodes for Contrastive Multimodal Graph Recommendation 提出MuSICRec，通过多模态对比图推荐缓解冷启动和数据稀疏问题。 multimodal
6 Multimodal Enhancement of Sequential Recommendation 提出MuSTRec，融合多模态信息与序列推荐，提升推荐性能。 multimodal
7 PreFlect: From Retrospective to Prospective Reflection in Large Language Model Agents PreFlect：大型语言模型Agent中从回顾性反思到前瞻性反思的转变 large language model ✅
8 ShallowJail: Steering Jailbreaks against Large Language Models 提出ShallowJail攻击，利用浅层对齐破解大语言模型的安全防护 large language model ✅
9 The Quantum Sieve Tracer: A Hybrid Framework for Layer-Wise Activation Tracing in Large Language Models 提出量子筛追踪器，用于分析大语言模型中的逐层激活追踪，揭示模型架构差异。 large language model
10 GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models GhostCite：大规模分析大语言模型时代下引文有效性问题 large language model
11 Multimodal Generative Retrieval Model with Staged Pretraining for Food Delivery on Meituan 针对美团外卖场景，提出基于分阶段预训练的多模态生成式检索模型 multimodal
12 LogicSkills: A Structured Benchmark for Formal Reasoning in Large Language Models LogicSkills：一个用于评估大语言模型形式推理能力的结构化基准 large language model
13 Same Answer, Different Representations: Hidden instability in VLMs 揭示视觉语言模型内部表征不稳定性：相同答案，不同表征 multimodal
14 How Well Can LLM Agents Simulate End-User Security and Privacy Attitudes and Behaviors? SP-ABCBench评估LLM智能体模拟用户安全隐私态度的能力，发现仍有提升空间 large language model
15 TamperBench: Systematically Stress-Testing LLM Safety Under Fine-Tuning and Tampering TamperBench：系统性压力测试LLM在微调和篡改下的安全性 large language model ✅
16 TraceCoder: A Trace-Driven Multi-Agent Framework for Automated Debugging of LLM-Generated Code TraceCoder：基于运行时追踪的多智能体框架，用于自动调试LLM生成的代码 large language model
17 ScaleEnv: Scaling Environment Synthesis from Scratch for Generalist Interactive Tool-Use Agent Training ScaleEnv：从零扩展环境合成，用于通用交互式工具使用Agent训练 generalist agent
18 Bridging 6G IoT and AI: LLM-Based Efficient Approach for Physical Layer's Optimization Tasks 提出基于LLM的PE-RTFV框架，用于6G IoT物理层优化 large language model
19 Wild Guesses and Mild Guesses in Active Concept Learning 研究主动概念学习中查询策略对神经符号贝叶斯学习器的影响，揭示了确认偏差的潜在合理性。 large language model
20 Evidence for Daily and Weekly Periodic Variability in GPT-4o Performance 揭示GPT-4o性能的每日和每周周期性波动，挑战时间不变性假设 large language model
21 AgentStepper: Interactive Debugging of Software Development Agents AgentStepper：用于软件开发Agent交互式调试的工具 large language model
22 Lemon Agent Technical Report Lemon Agent：基于AgentCortex框架的多智能体协同系统，提升复杂任务处理效率。 multimodal
23 HyPER: Bridging Exploration and Exploitation for Scalable LLM Reasoning with Hypothesis Path Expansion and Reduction HyPER：通过假设路径扩展与缩减，桥接探索与利用，实现可扩展的LLM推理 chain-of-thought
24 Evaluating Retrieval-Augmented Generation Variants for Natural Language-Based SQL and API Call Generation 评估检索增强生成变体在自然语言到SQL和API调用生成中的应用 large language model
25 BEAGLE: Behavior-Enforced Agent for Grounded Learner Emulation BEAGLE：行为增强的智能体，用于模拟扎根学习者的学习过程 large language model
26 Rethinking Scientific Modeling: Toward Physically Consistent and Simulation-Executable Programmatic Generation 提出物理一致的程序化生成框架，用于自动创建可执行的结构建模代码。 large language model ✅
27 Intrinsic Stability Limits of Autoregressive Reasoning: Structural Consequences for Long-Horizon Execution 揭示自回归推理的内在稳定性极限，提出长程执行的结构性治理方案 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🔗⭐
28 Next-generation cyberattack detection with large language models: anomaly analysis across heterogeneous logs 利用大语言模型进行下一代网络攻击检测，实现跨异构日志的异常分析 distillation large language model
29 Unlocking Noisy Real-World Corpora for Foundation Model Pre-Training via Quality-Aware Tokenization 提出QA-Token，通过质量感知分词提升噪声数据上预训练模型效果 reinforcement learning foundation model
30 LatentChem: From Textual CoT to Latent Thinking in Chemical Reasoning 提出LatentChem，通过隐空间推理提升化学大模型效率与性能。 latent dynamics large language model chain-of-thought
31 Towards Understanding What State Space Models Learn About Code 首个SSM代码理解分析：揭示其在代码建模中的优势与局限性 SSM state space model
32 SeeUPO: Sequence-Level Agentic-RL with Convergence Guarantees 提出SeeUPO，一种具备收敛保证的序列级Agentic-RL算法，解决多轮交互场景下的训练不稳定性问题。 reinforcement learning PPO large language model
33 Semantically Labelled Automata for Multi-Task Reinforcement Learning with LTL Instructions 提出基于语义标记自动机的多任务强化学习方法，解决LTL指令下的泛化问题。 reinforcement learning
34 Prism: Spectral Parameter Sharing for Multi-Agent Reinforcement Learning Prism：基于谱参数共享的多智能体强化学习框架，提升资源效率 reinforcement learning
35 Progress Constraints for Reinforcement Learning in Behavior Trees 提出基于进度约束的强化学习行为树方法，提升任务性能与样本效率。 reinforcement learning
36 AgentCPM-Explore: Realizing Long-Horizon Deep Exploration for Edge-Scale Agents AgentCPM-Explore：面向边缘规模Agent的长程深度探索 reinforcement learning large language model
37 AbFlow : End-to-end Paratope-Centric Antibody Design by Interaction Enhanced Flow Matching AbFlow：通过交互增强的Flow Matching实现以互补位为中心的端到端抗体设计 flow matching
38 Sample-Efficient Policy Space Response Oracles with Joint Experience Best Response 提出Joint Experience Best Response，提升PSRO在多智能体强化学习中的样本效率。 reinforcement learning offline RL

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
39 Empirical Analysis of Adversarial Robustness and Explainability Drift in Cybersecurity Classifiers 针对网络安全分类器，研究对抗鲁棒性与可解释性漂移问题 manipulation
40 Incentive-Aware AI Safety via Strategic Resource Allocation: A Stackelberg Security Games Perspective 提出基于Stackelberg安全博弈的激励感知AI安全框架，解决AI系统开发与部署中的对抗性风险。 manipulation
41 Malicious Agent Skills in the Wild: A Large-Scale Security Empirical Study 构建恶意Agent技能数据集，揭示LLM Agent生态系统中的安全漏洞 manipulation

🔬 支柱八：物理动画 (Physics-based Animation) (2 篇)

#题目一句话要点标签🔗⭐
42 SAS-Net: Scene-Appearance Separation Network for Robust Spatiotemporal Registration in Bidirectional Photoacoustic Microscopy 提出SAS-Net，解决双向光声显微镜中时空配准的域偏移和几何失真问题。 spatiotemporal ✅
43 GraFSTNet: Graph-based Frequency SpatioTemporal Network for Cellular Traffic Prediction GraFSTNet：基于图的频率时空网络用于蜂窝流量预测 spatiotemporal

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

#题目一句话要点标签🔗⭐
44 Reasoning-Augmented Representations for Multimodal Retrieval 提出推理增强表示框架，提升通用多模态检索中隐式推理能力 feature matching multimodal ✅

⬅️ 返回 cs.AI 首页 · 🏠 返回主页