医疗AI、科学Agent、机器人VLA：垂直领域正在成为AI的主战场

重点关注

• MedXIAOHE: A Comprehensive Recipe for Building Medical MLLMs score 9
  入选 HF Daily Papers；HF 热度: 52 upvotes (+4)；关键词(8): scaling, preference, pretraining, agentic, reasoning
• On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs score 6
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(6): fine-tuning, alignment, reasoning, multimodal, benchmark
• GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics score 7
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；关键词(2): agent, reasoning
• Towards Universal Video MLLMs with Attribute-Structured and Quality-Verified Instructions score 7
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；关键词(4): fine-tuning, audio, open-source, data curation
• SciAgentGym: Benchmarking Multi-Step Scientific Tool-use in LLM Agents score 6
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；关键词(5): fine-tuning, agents, agentic, reasoning, evaluation
• FLAC: Maximum Entropy RL via Kinetic Energy Regularized Bridge Matching score 5
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(1): diffusion
• Self-EvolveRec: Self-Evolving Recommender Systems with LLM-based Directional Feedback score 5
  入选 HF Daily Papers；关键词(3): search, recommendation, evaluation
• Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Execution score 5
  入选 HF Daily Papers；关键词(9): fast, deployment, inference, latency, throughput
• CoPE-VideoLM: Codec Primitives For Efficient Video Language Models score 6
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(6): efficient, lightweight, fine-tuning, pre-training, transformer
• RelBench v2: A Large-Scale Benchmark and Repository for Relational Data score 5
  机构: Stanford；关键词(5): pretraining, planning, recommendation, benchmark, evaluation

也值得关注

• Bootstrapping MLLM for Weakly-Supervised Class-Agnostic Object Counting score 4
  关键词(1): fine-tuning；顶会接收: ICLR
• Gradient-Enhanced Partitioned Gaussian Processes for Real-Time Quadrotor Dynamics Modeling score 2
  关键词(4): efficient, inference, real-time, cost
• Layer-Specific Fine-Tuning for Improved Negation Handling in Medical Vision-Language Models score 2
  关键词(5): fine-tuning, alignment, vision-language, benchmark, safety
• A Theoretical Analysis of Mamba's Training Dynamics: Filtering Relevant Features for Generalization in State Space Models score 2
  关键词(5): transformer, attention, state space, mamba, synthetic data
• Favia: Forensic Agent for Vulnerability-fix Identification and Analysis score 5
  入选 HF Daily Papers；关键词(5): efficient, alignment, agent, reasoning, search
• Building Large-Scale Drone Defenses from Small-Team Strategies score 2
  关键词(3): efficient, agent, evaluation
• Visual RAG Toolkit: Scaling Multi-Vector Visual Retrieval with Training-Free Pooling and Multi-Stage Search score 2
  关键词(11): scaling, efficiency, fast, lightweight, distillation
• Bench-MFG: A Benchmark Suite for Learning in Stationary Mean Field Games score 2
  关键词(3): agent, benchmark, evaluation
• Multi-Agent Model-Based Reinforcement Learning with Joint State-Action Learned Embeddings score 2
  关键词(4): efficient, agent, agents, planning
• Constraint-Rectified Training for Efficient Chain-of-Thought score 2
  关键词(7): efficient, efficiency, pruning, inference, post-training
• DiffuRank: Effective Document Reranking with Diffusion Language Models score 2
  关键词(4): efficiency, latency, diffusion, reasoning
• Flow-Factory: A Unified Framework for Reinforcement Learning in Flow-Matching Models score 2
  关键词(3): production, GRPO, diffusion
• Reasoning to Rank: An End-to-End Solution for Exploiting Large Language Models for Recommendation score 2
  关键词(2): reasoning, recommendation
• AMPS: Adaptive Modality Preference Steering via Functional Entropy score 2
  关键词(4): scaling, inference, preference, multimodal
• Self-Supervised JEPA-based World Models for LiDAR Occupancy Completion and Forecasting score 2
  关键词(2): agent, planning