PDE替掉attention误差低2倍，局部RL省3/4算力

重点关注

• LongCat-Flash-Prover: Advancing Native Formal Reasoning via Agentic Tool-Integrated Reinforcement Learning score 10
  入选 HF Daily Papers；HF 热度: 63 upvotes (+4)；有代码实现；关键词(4): MoE, agentic, reasoning, open-source
• F4Splat: Feed-Forward Predictive Densification for Feed-Forward 3D Gaussian Splatting score 10
  入选 HF Daily Papers；HF 热度: 31 upvotes (+4)；有代码实现；关键词(1): real-time
• RoboAlign: Learning Test-Time Reasoning for Language-Action Alignment in Vision-Language-Action Models score 8
  入选 HF Daily Papers；HF 热度: 21 upvotes (+4)；关键词(4): reasoning, vision-language, robotics, embodied
• PivotRL: High Accuracy Agentic Post-Training at Low Compute Cost score 7
  入选 HF Daily Papers；HF 热度: 12 upvotes (+3)；关键词(5): production, fine-tuning, post-training, agentic, coding
• Generalized Discrete Diffusion from Snapshots score 7
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)；有代码实现
• FluidWorld: Reaction-Diffusion Dynamics as a Predictive Substrate for World Models score 5
  入选 HF Daily Papers；有代码实现

也值得关注

• Reading Between the Lines: How Electronic Nonverbal Cues shape Emotion Decoding score 4
  关键词(2): reasoning, embodied；顶会接收: AAAI
• CoVFT: Context-aware Visual Fine-tuning for Multimodal Large Language Models score 4
  关键词(2): fine-tuning, reasoning；顶会接收: CVPR
• Mixture of Chapters: Scaling Learnt Memory in Transformers score 4
  关键词(4): scaling, fine-tuning, pre-training, pretraining；顶会接收: ICLR
• Frequency Switching Mechanism for Parameter-E!cient Multi-Task Learning score 4
  关键词(2): lightweight, fine-tuning；顶会接收: CVPR
• ORACLE: Optimizing Reasoning Abilities of Large Language Models via Constraint-Led Synthetic Data Elicitation score 4
  关键词(2): reasoning, synthetic data；顶会接收: AAAI
• Plant Taxonomy Meets Plant Counting: A Fine-Grained, Taxonomic Dataset for Counting Hundreds of Plant Species score 4
  关键词(1): reasoning；顶会接收: CVPR
• Uncertainty-Aware Knowledge Distillation for Multimodal Large Language Models score 4
  关键词(2): distillation, vision-language；顶会接收: CVPR
• CTFS : Collaborative Teacher Framework for Forward-Looking Sonar Image Semantic Segmentation with Extremely Limited Labels score 3
  顶会接收: CVPR
• Taming Sampling Perturbations with Variance Expansion Loss for Latent Diffusion Models score 3
  顶会接收: CVPR
• LiFR-Seg: Anytime High-Frame-Rate Segmentation via Event-Guided Propagation score 3
  顶会接收: ICLR
• Reframing Long-Tailed Learning via Loss Landscape Geometry score 3
  顶会接收: CVPR
• EmoTaG: Emotion-Aware Talking Head Synthesis on Gaussian Splatting with Few-Shot Personalization score 3
  顶会接收: CVPR
• FluidGaussian: Propagating Simulation-Based Uncertainty Toward Functionally-Intelligent 3D Reconstruction score 3
  顶会接收: CVPR
• Constrained Online Convex Optimization with Memory and Predictions score 3
  顶会接收: AAAI