cs.AI（2026-03-19）

📊 共 31 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) (20 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (8 🔗3) 支柱一：机器人控制 (Robot Control) (2) 支柱五：交互与反应 (Interaction & Reaction) (1)

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

#题目一句话要点标签🔗⭐
1 Cognitive Mismatch in Multimodal Large Language Models for Discrete Symbol Understanding 揭示多模态大语言模型在离散符号理解中的认知失配问题，并提出评测基准。 large language model multimodal
2 dTRPO: Trajectory Reduction in Policy Optimization of Diffusion Large Language Models dTRPO：通过轨迹缩减优化扩散大语言模型的策略 large language model instruction following
3 How Uncertainty Estimation Scales with Sampling in Reasoning Models 研究推理模型中基于采样的不确定性估计方法，并提出混合估计器。 chain-of-thought
4 Serendipity by Design: Evaluating the Impact of Cross-domain Mappings on Human and LLM Creativity 跨领域映射提升人类与LLM创造力：设计中的意外发现 large language model
5 Evaluating 5W3H Structured Prompting for Intent Alignment in Human-AI Interaction 提出基于5W3H结构化提示的PPS框架，提升人机交互中意图对齐效果 large language model
6 I Can't Believe It's Corrupt: Evaluating Corruption in Multi-Agent Governance Systems 评估多智能体治理系统中腐败现象，强调制度设计的重要性 large language model
7 Quantitative Introspection in Language Models: Tracking Internal States Across Conversation 提出数值自报告以追踪语言模型的内部状态 large language model
8 Geography According to ChatGPT -- How Generative AI Represents and Reasons about Geography 评估ChatGPT的地理知识表示与推理能力，揭示生成式AI的地理认知局限性 foundation model
9 Measuring and Exploiting Confirmation Bias in LLM-Assisted Security Code Review 研究LLM辅助代码审查中的确认偏差，揭示软件供应链攻击风险 large language model
10 Analysis Of Linguistic Stereotypes in Single and Multi-Agent Generative AI Architectures 分析单智能体和多智能体生成式AI架构中的语言刻板印象 chain-of-thought
11 Accurate and Efficient Multi-Channel Time Series Forecasting via Sparse Attention Mechanism 提出Li-Net，通过稀疏注意力机制实现准确高效的多通道时间序列预测。 multimodal
12 An Onto-Relational-Sophic Framework for Governing Synthetic Minds 提出Onto-Relational-Sophic框架，为通用人工智能的治理提供哲学基础。 foundation model
13 ZEBRAARENA: A Diagnostic Simulation Environment for Studying Reasoning-Action Coupling in Tool-Augmented LLMs ZebraArena：用于研究工具增强LLM中推理-行动耦合的诊断模拟环境 large language model
14 The Spillover Effects of Peer AI Rinsing on Corporate Green Innovation 提出针对企业AI洗涤行为的政策工具以促进绿色创新 large language model
15 PlanTwin: Privacy-Preserving Planning Abstractions for Cloud-Assisted LLM Agents 提出PlanTwin以解决云端规划中的隐私泄露问题 large language model
16 ItinBench: Benchmarking Planning Across Multiple Cognitive Dimensions with Large Language Models ItinBench：利用大语言模型在多认知维度上进行规划的基准测试 large language model ✅
17 Learning to Disprove: Formal Counterexample Generation with Large Language Models 提出基于大语言模型的形式化反例生成方法，提升数学推理能力 large language model
18 The Autonomy Tax: Defense Training Breaks LLM Agents 揭示防御训练导致LLM Agent能力退化的“自主性税”，并分析其根本原因。 large language model
19 POET: Power-Oriented Evolutionary Tuning for LLM-Based RTL PPA Optimization POET：面向功耗优化的LLM驱动RTL代码演化调优框架 large language model
20 PlanTwin: Privacy-Preserving Planning Abstractions for Cloud-Assisted LLM Agents PlanTwin：为云辅助LLM代理提供隐私保护的规划抽象 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🔗⭐
21 Balanced Thinking: Improving Chain of Thought Training in Vision Language Models 提出SCALe，通过动态损失加权改进视觉语言模型中的思维链训练 reinforcement learning multimodal chain-of-thought
22 AlignMamba-2: Enhancing Multimodal Fusion and Sentiment Analysis with Modality-Aware Mamba AlignMamba-2：利用模态感知Mamba增强多模态融合与情感分析 Mamba multimodal
23 Bridging Network Fragmentation: A Semantic-Augmented DRL Framework for UAV-aided VANETs 提出SA-DRL框架，利用语义增强的DRL解决UAV辅助VANET中的网络碎片问题。 reinforcement learning deep reinforcement learning DRL
24 RewardFlow: Topology-Aware Reward Propagation on State Graphs for Agentic RL with Large Language Models RewardFlow：利用状态图拓扑结构的奖励传播，提升LLM智能体强化学习效果 reinforcement learning large language model ✅
25 Functional Subspace Watermarking for Large Language Models 提出功能子空间水印(FSW)方法，增强大语言模型水印对参数扰动的鲁棒性。 distillation large language model
26 LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modeling 提出LuMamba以解决EEG建模中的电极拓扑不变性与计算效率问题 Mamba foundation model ✅
27 Box Maze: A Process-Control Architecture for Reliable LLM Reasoning 提出Box Maze框架，通过过程控制架构提升LLM推理的可靠性 reinforcement learning RLHF large language model
28 Memento-Skills: Let Agents Design Agents Memento-Skills：提出一种通用、可持续学习的LLM Agent系统，实现Agent的自主设计。 reinforcement learning generalist agent ✅

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
29 Thinking with Constructions: A Benchmark and Policy Optimization for Visual-Text Interleaved Geometric Reasoning 提出A2PO，通过强化学习提升MLLM在几何推理中利用辅助线的能力 manipulation reinforcement learning reward shaping
30 Cross-Domain Demo-to-Code via Neurosymbolic Counterfactual Reasoning 提出NeSyCR，通过神经符号反事实推理实现跨域Demo-to-Code manipulation

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
31 Secure Linear Alignment of Large Language Models 提出一种安全线性对齐框架，用于跨独立训练的大语言模型进行隐私保护的交叉推理。 OMOMO large language model

⬅️ 返回 cs.AI 首页 · 🏠 返回主页