OPSA Lifts AIME24 by 35.41 Points

Featured

• Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement score 10
  入选 HF Daily Papers; HF 热度: 126 upvotes (+4); 有代码实现; 关键词(1): distillation
• PaperGym: Rubric-Centered Evolution for Research-Plan Generation score 10
  入选 HF Daily Papers; HF 热度: 35 upvotes (+4); 有代码实现; 关键词(2): fine-tuning, GRPO

Also Worth Noting

• Efficient and High-Quality Depth Estimation via Pixel-Space Diffusion with Linear Attention score 4
  关键词(2): scaling, latency; 顶会接收: ECCV
• CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents score 9
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 关键词(2): fine-tuning, agentic; 顶会接收: EMNLP
• Reactivating Test-Time Scaling for Plane Geometry Problem Solving score 4
  关键词(2): scaling, reasoning; 顶会接收: EMNLP
• Doc-REFRAG: Rethinking Multimodal Document Retrieval-Augmented Generation score 4
  关键词(4): lightweight, latency, retrieval-augmented, RAG; 顶会接收: EMNLP
• NoisEasier: Test-Time Noise Optimization for Text-to-Video Generation score 4
  关键词(3): scaling, fine-tuning, text-to-video; 顶会接收: ECCV
• Quantifying and Mitigating Korean Jamo-Level Typographical Vulnerabilities in Large Language Models score 4
  关键词(1): scaling; 顶会接收: EMNLP
• Beyond Token-Level Guidance: Inference-Time Alignment of Specialized LLMs via Cross-Family Representation Steering score 4
  关键词(1): finetuning; 顶会接收: EMNLP
• CapFrame: Text-Instructed Viewpoint Grounding in 3D Gaussian Scenes via Geometric Pseudo Labels score 4
  关键词(1): real-time; 顶会接收: ECCV
• Auditing MCQA Benchmarks through Probability Landscapes score 4
  关键词(1): lightweight; 顶会接收: EMNLP
• Real-Time Scene-Adaptive Tone Mapping for High-Dynamic Range Object Detection score 4
  关键词(3): scaling, real-time, finetuning; 顶会接收: NeurIPS
• EvoSkill Injection: Red-Teaming Autonomous Skill Generation and Evolution in Self-Evolving Agents score 4
  关键词(1): reasoning; 顶会接收: EMNLP
• Enhancing Low-Resource Language Reasoning via High-Resource Language Feature Transfer score 4
  关键词(3): fine-tuning, pretraining, reasoning; 顶会接收: EMNLP
• VisER: Visual Evidence and Reliance for Object Hallucination Detection in LVLMs score 4
  关键词(1): vision-language; 顶会接收: EMNLP
• PAC: Progress-Augmented Advantage Curriculum for Multi-Task Reinforcement Learning of LLMs score 4
  关键词(3): GRPO, post-training, reasoning; 顶会接收: EMNLP
• Modality Disentangled Learning for Incomplete Multimodal Emotion Recognition: A Primitive Memory Distillation Perspective score 4
  关键词(1): distillation; 顶会接收: EMNLP