蒸馏砍掉模型的犹豫，OOD暴跌40%

重点关注

• Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs? score 10
  入选 HF Daily Papers；HF 热度: 37 upvotes (+4)；有代码实现；关键词(3): distillation, post-training, reasoning
• CUA-Suite: Massive Human-annotated Video Demonstrations for Computer-Use Agents score 8
  入选 HF Daily Papers；HF 热度: 85 upvotes (+4)；关键词(2): scaling, reasoning
• UI-Voyager: A Self-Evolving GUI Agent Learning via Failed Experience score 10
  入选 HF Daily Papers；HF 热度: 37 upvotes (+4)；有代码实现；关键词(2): distillation, fine-tuning
• Understanding the Challenges in Iterative Generative Optimization with LLMs score 8
  入选 HF Daily Papers；HF 热度: 18 upvotes (+3)；有代码实现
• CarePilot: A Multi-Agent Framework for Long-Horizon Computer Task Automation in Healthcare score 8
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现；关键词(4): agentic, reasoning, vision-language, open-source
• GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents score 7
  入选 HF Daily Papers；HF 热度: 17 upvotes (+3)；关键词(4): agentic, reasoning, robotics, embodied
• SpectralSplats: Robust Differentiable Tracking via Spectral Moment Supervision score 7
  入选 HF Daily Papers；HF 热度: 11 upvotes (+3)；关键词(1): real-time
• OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning score 5
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；关键词(3): pretraining, reasoning, open-source

也值得关注

• Toward Physically Consistent Driving Video World Models under Challenging Trajectories score 4
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)
• SCoOP: Semantic Consistent Opinion Pooling for Uncertainty Quantification in Multiple Vision-Language Model Systems score 4
  关键词(2): reasoning, vision-language；顶会接收: ICLR
• Off-Policy Safe Reinforcement Learning with Constrained Optimistic Exploration score 4
  关键词(1): deployment；顶会接收: ICLR
• MMTIT-Bench: A Multilingual and Multi-Scenario Benchmark with Cognition-Perception-Reasoning Guided Text-Image Machine Translation score 4
  关键词(2): reasoning, vision-language；顶会接收: CVPR
• Decompose and Transfer: CoT-Prompting Enhanced Alignment for Open-Vocabulary Temporal Action Detection score 4
  关键词(1): reasoning；顶会接收: CVPR
• A^3: Towards Advertising Aesthetic Assessment score 4
  关键词(1): deployment；顶会接收: CVPR
• When Understanding Becomes a Risk: Authenticity and Safety Risks in the Emerging Image Generation Paradigm score 4
  关键词(1): edge；顶会接收: CVPR
• Tutor-Student Reinforcement Learning: A Dynamic Curriculum for Robust Deepfake Detection score 4
  关键词(1): PPO；顶会接收: CVPR
• LightSplat: Fast and Memory-Efficient Open-Vocabulary 3D Scene Understanding in Five Seconds score 4
  关键词(1): lightweight；顶会接收: CVPR
• RS-SSM: Refining Forgotten Specifics in State Space Model for Video Semantic Segmentation score 4
  关键词(2): compression, state space；顶会接收: CVPR
• ViHOI: Human-Object Interaction Synthesis with Visual Priors score 4
  关键词(2): vision-language, text-to-image；顶会接收: CVPR
• Unleashing Vision-Language Semantics for Deepfake Video Detection score 4
  关键词(1): vision-language；顶会接收: CVPR
• Video-Only ToM: Enhancing Theory of Mind in Multimodal Large Language Models score 4
  关键词(1): reasoning；顶会接收: CVPR