Skip to content

AI论文速递 2026年05月13日（HuggingFace Daily Papers）

数据来源：https://huggingface.co/papers 采集时间：2026-05-13

📌 重点关注

  1. LychSim: A Controllable and Interactive Simulation Framework for Vision Research | arXiv — 【重点关注】 While self-supervised pretraining has reduced vision systems' reliance on syn... 💡 基于UE5的视觉仿真框架，原生集成MCP让LLM Agent闭环控制，仿真+Agent结合的新范式
  2. Rethinking Agentic Search with Pi-Serini: Is Lexical Retrieval Sufficient? | arXiv — 【重点关注】 Does a lexical retriever suffice as large language models (LLMs) become more ... 💡 BM25+GPT-5.5达83.1%准确率，证明Agent循环中简单检索就够了，成本降3-10倍
  3. Training-Free Dense Hand Contact Estimation with Multi-Modal Large Language Models | arXiv — 【重点关注】 Dense hand contact estimation requires both high-level semantic understanding... 💡 零训练零样本，MLLM推理能力直接超越有监督方法，Prompt Engineering的威力

📋 其他值得关注

  1. SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture | arXiv — Recent large vision-language models (VLMs) remain fundamentally constrained b...
  2. Images in Sentences: Scaling Interleaved Instructions for Unified Visual Generation | arXiv — While recent advancements in multimodal language models have enabled image ge...
  3. LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models | arXiv — Looped computation shows promise in improving the reasoning-oriented performa...
  4. Teaching Language Models to Think in Code | arXiv — Tool-integrated reasoning (TIR) has emerged as a dominant paradigm for mathem...
  5. DeepRefine: Agent-Compiled Knowledge Refinement via Reinforcement Learning | arXiv — Agent-compiled knowledge bases provide persistent external knowledge for larg...
  6. RoboMemArena: A Comprehensive and Challenging Robotic Memory Benchmark | arXiv — Memory is a critical component of robotic intelligence, as robots must rely o...
  7. TMAS: Scaling Test-Time Compute via Multi-Agent Synergy | arXiv — Test-time scaling has become an effective paradigm for improving the reasonin...