AI论文速递 2026年05月22日（HuggingFace Daily Papers）¶

数据来源：https://huggingface.co/papers 采集时间：2026-05-22

📌 重点关注¶

IndusAgent: Reinforcing Open-Vocabulary Industrial Anomaly Detection with Agentic Tools | arXiv — 【重点关注】 Multimodal large language models (MLLMs) have shown remarkable capability in ... 💡 工业异常检测引入Agent工具调用，对端侧AI质检场景有直接落地参考价值。
Maestro: Reinforcement Learning to Orchestrate Hierarchical Model-Skill Ensembles | arXiv — 【重点关注】 The proliferation of large language models (LLMs) and modular skills has endo... 💡 用RL编排多模型多技能的层级架构，为复杂Agent系统的编排策略提供了新思路。
MINTEval: Evaluating Memory under Multi-Target Interference in Long-Horizon Agent Systems | arXiv — 【重点关注】 Real-world agents operate over long and evolving horizons, where information ... 💡 多目标干扰下的长程记忆评估，揭示了Agent记忆管理的关键瓶颈，对Agent架构设计启发大。

Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMs | arXiv — LLM agents have recently emerged as a powerful paradigm for solving complex t...
Bernini: Latent Semantic Planning for Video Diffusion | arXiv — Multimodal large language models (MLLMs) and diffusion models have each reach...
ACC: Compiling Agent Trajectories for Long-Context Training | arXiv — Recent development of agents has renewed demand for long-context reasoning ca...
Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection | arXiv — Safety post-training can improve the harmfulness and policy compliance of Lar...
It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs | arXiv — Contextual Integrity (CI) defines privacy not merely as keeping information h...
A Survey of Large Audio Language Models: Generalization, Trustworthiness, and Outlook | arXiv — The foundational capabilities established by Large Language Models (LLMs) hav...
Perception or Prejudice: Can MLLMs Go Beyond First Impressions of Personality? | arXiv — Multimodal Large Language Models (MLLMs) are increasingly deployed in human-f...