On-Policy Feedback Makes Training Signals Useful

Featured

• Recursive Experiential-Working Memory Evolution for Long-Horizon Agent Harnesses score 9
  入选 HF Daily Papers; HF 热度: 22 upvotes (+4); 有代码实现
• On-Policy Self-Distillation in Diffusion Models score 10
  入选 HF Daily Papers; HF 热度: 54 upvotes (+4); 有代码实现; 关键词(2): distillation, post-training
• On-policy Distillation with Verifiable Reward score 9
  入选 HF Daily Papers; HF 热度: 15 upvotes (+3); 有代码实现; 关键词(4): distillation, GRPO, post-training, reasoning
• TorchMorph: CUDA-accelerated Morphological Transforms score 7
  入选 HF Daily Papers; HF 热度: 3 upvotes (+1); 有代码实现; 关键词(2): lightweight, throughput

Also Worth Noting

• CAFE: Self-Improving Search Agents Need Co-Evolving Feedback score 5
  入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 关键词(1): agentic
• Beyond Confidence: Test-Time Scaling for Multi-Turn Search Agents via Retrieval Grounding score 4
  关键词(2): scaling, reasoning; 顶会接收: EMNLP
• Compression Trinity: Exploring Sparsity, Quantization, and Low-Rank Approximations for LLM Compression score 4
  机构: University of Toronto; 关键词(6): compression, quantization, deployment, fine-tuning, post-training
• When Less Is More: An Empirical Study of Minimal Responses in Counseling Dialogues and the Behavior of LLMs score 4
  关键词(1): synthetic data; 顶会接收: EMNLP
• Robust Code RL via Faulty-Code-Driven Test case Synthesis and Dense Reward Shaping score 4
  关键词(3): fine-tuning, code generation, coding; 顶会接收: EMNLP
• MemUse: Moving Memory Evaluation from Direct QA to Natural Integration in Long-Term Human-AI Conversation score 9
  入选 HF Daily Papers; 有代码实现; 关键词(1): deployment; 顶会接收: EMNLP
• Who is the Agent to Blame? Localizing Faithfulness and Citation Mistakes in Agentic Deep Research score 4
  关键词(2): agentic, open-source; 顶会接收: EMNLP
• Low-Rank Ternary Adaptation for Fine-Tuning Transformers score 4
  关键词(3): quantization, fine-tuning, fine-tune; 顶会接收: ECCV
• IDeaL: Data-Free Multi-Teacher Distillation via Improved Dead Leaves score 4
  关键词(1): distillation; 顶会接收: ECCV
• StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing score 4
  关键词(2): GRPO, guardrails; 顶会接收: EMNLP
• Meta$^n$: Recursive Self-Improvement through Emergent Depth score 6
  入选 HF Daily Papers; HF 热度: 10 upvotes (+3)
• SENSESHIFT: Continuous Sentiment-Controlled Text Generation via Encoder-based Mask Infilling score 3
  顶会接收: EMNLP
• Equivariant Covariance Tensors: Guaranteed SPD Uncertainty for Tensor-Valued Geometric Learning score 3
  顶会接收: ICML
• Expectation, Backlash, Recovery, and Excitement: How Model Releases Shape Reddit Perceptions of Conversational AI Systems score 3
  顶会接收: EMNLP
• Arbitrary Polygon Oscillator: Generalizing Polygonal Synthesis to Arbitrary Shapes, Morphing, and Three-Dimensional Polyhedra score 3
  机构: MIT