AI论文速递 2026年04月27日（HuggingFace Daily Papers）¶

数据来源：https://huggingface.co/papers 采集时间：2026-04-27

📌 重点关注¶

OpenMobile: Building Open Mobile Agents with Task and Trajectory Synthesis | arXiv — 【重点关注】 Mobile agents powered by vision-language models have demonstrated impressive ... 💡 基于VLM的移动智能体，轨迹合成能力值得关注
WebGen-R1: Incentivizing Large Language Models to Generate Functional and Aesthetic Websites with Reinforcement Learning | arXiv — 【重点关注】 While Large Language Models (LLMs) excel at function-level code generation, p... 💡 强化学习指导LLM生成功能与美学兼备的网站，创意编程新方向
LLaTiSA: Towards Difficulty-Stratified Time Series Reasoning from Visual Perception to Semantics | arXiv — 【重点关注】 Comprehensive understanding of time series remains a significant challenge fo... 💡 从视觉感知到语义理解的时间序列推理分层框架，多模态分析利器

PersonalAI: A Systematic Comparison of Knowledge Graph Storage and Retrieval Approaches for Personalized LLM agents | arXiv — Personalizing language models by effectively incorporating user interaction h...
Co-Evolving LLM Decision and Skill Bank Agents for Long-Horizon Tasks | arXiv — Long horizon interactive environments are a testbed for evaluating agents ski...
3D-VCD: Hallucination Mitigation in 3D-LLM Embodied Agents through Visual Contrastive Decoding | arXiv — Large multimodal models are increasingly used as the reasoning core of embodi...
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges | arXiv — Reinforcement Learning from Human Feedback (RLHF) and related alignment parad...
Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs | arXiv — Prior work shows that fine-tuning aligned models on benign data degrades safe...
Visual Reasoning through Tool-supervised Reinforcement Learning | arXiv — In this paper, we investigate the problem of how to effectively master tool-u...
AgriIR: A Scalable Framework for Domain-Specific Knowledge Retrieval | arXiv — This paper introduces AgriIR, a configurable retrieval augmented generation (...