cs.AI（2026-02-23）

📊 共 17 篇论文 | 🔗 1 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (10 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (6) 支柱一：机器人控制 (Robot Control) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (10 篇)

#题目一句话要点标签🔗⭐
1 Hiding in Plain Text: Detecting Concealed Jailbreaks via Activation Disentanglement 提出ReDAct与FrameShield，通过激活解耦检测大语言模型中的隐蔽越狱攻击。 large language model
2 A Multimodal Framework for Aligning Human Linguistic Descriptions with Visual Perceptual Data 提出多模态框架，对齐人类语言描述与视觉感知数据，解决跨模态参照理解问题。 multimodal
3 DReX: An Explainable Deep Learning-based Multimodal Recommendation Framework DReX：一种基于深度学习的可解释多模态推荐框架 multimodal
4 Can Large Language Models Replace Human Coders? Introducing ContentBench ContentBench：评估低成本大语言模型在内容分析编码任务中的能力。 large language model
5 Classroom Final Exam: An Instructor-Tested Reasoning Benchmark 提出 Classroom Final Exam (CFE)，用于评估大语言模型在 STEM 领域的推理能力。 large language model multimodal ✅
6 CausalFlip: A Benchmark for LLM Causal Judgment Beyond Semantic Matching CausalFlip：提出因果判断基准，评估LLM在语义匹配之外的因果推理能力 large language model chain-of-thought
7 AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization AdaEvolve：自适应LLM驱动的零阶优化框架，提升进化搜索效率 large language model
8 The LLMbda Calculus: AI Agents, Conversations, and Information Flow 提出LLMbda演算以解决AI代理安全性问题 large language model
9 LLM-enabled Applications Require System-Level Threat Monitoring LLM应用面临系统级威胁，提出安全监控方案以保障可靠运行 large language model
10 Red-Teaming Claude Opus and ChatGPT-based Security Advisors for Trusted Execution Environments 针对TEE安全，提出TEE-RedBench评估方法，用于评估LLM安全顾问的可靠性 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (6 篇)

#题目一句话要点标签🔗⭐
11 IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking IR$^3$：通过对比逆强化学习实现奖励篡改的可解释检测与缓解 reinforcement learning inverse reinforcement learning RLHF
12 Human-Guided Agentic AI for Multimodal Clinical Prediction: Lessons from the AgentDS Healthcare Benchmark 提出人机协作的Agentic AI，用于多模态临床预测，并在AgentDS医疗基准上验证有效性。 MAE multimodal
13 Tri-Subspaces Disentanglement for Multimodal Sentiment Analysis 提出Tri-Subspace Disentanglement框架，解决多模态情感分析中模态间信息融合不充分的问题。 MAE multimodal
14 Ada-RS: Adaptive Rejection Sampling for Selective Thinking Ada-RS：自适应拒绝采样提升工具型LLM选择性推理效率 DPO large language model chain-of-thought
15 Beyond Mimicry: Toward Lifelong Adaptability in Imitation Learning 面向终身适应性的模仿学习：超越单纯模仿，实现组合泛化 imitation learning
16 Meta-Learning and Meta-Reinforcement Learning - Tracing the Path towards DeepMind's Adaptive Agent 综述元学习与元强化学习，追溯DeepMind自适应Agent的技术演进路径 reinforcement learning

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
17 Agentic AI as a Cybersecurity Attack Surface: Threats, Exploits, and Defenses in Runtime Supply Chains 针对Agentic AI运行时供应链的攻击面研究：威胁、利用与防御 manipulation large language model

⬅️ 返回 cs.AI 首页 · 🏠 返回主页