Across 600 Hard Cases, RSB Searches 50% Fewer Branches Than State-of-the-Art Heuristics

Featured

• Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models score 10
  入选 HF Daily Papers; HF 热度: 20 upvotes (+4); 有代码实现; 关键词(2): lightweight, pruning

Also Worth Noting

• ControlScope: Workflow Revision and Reliability in LLM Agents score 4
  入选 HF Daily Papers; 关键词(1): reasoning
• CASS: Contribution-Aware Structured Sparsity for Model Merging score 4
  关键词(2): pruning, fine-tuning; 顶会接收: NeurIPS
• Commutator Memory: Sparse, Path-Local Reading and Steering in Language Models score 4
  关键词(2): DPO, GRPO; 顶会接收: NeurIPS
• OSPD: On-Policy Self-Distillation for Persona-Consistent Dialogue score 4
  关键词(2): distillation, fine-tuning; 顶会接收: EMNLP
• AgentHop: A Diagnostic Benchmark for Agentic Multi-Hop Scientific Question Answering score 4
  关键词(2): agentic, leaderboard; 顶会接收: NeurIPS
• Verifying Neural Networks with Reinforcement Learning score 4
  关键词(1): pruning; 顶会接收: NeurIPS
• PersonaManifold: Revealing and Exploiting Curved Geometry in LLM Persona Representations score 4
  关键词(2): scaling, open-source; 顶会接收: NeurIPS
• WorldAttention: An Efficient Attention Architecture for Interactive Video World Models score 12
  机构: Alibaba; 入选 HF Daily Papers; HF 热度: 14 upvotes (+3); 有代码实现; 关键词(2): latency, embodied
• Revitalizing Medical Time Series with Vision-Informed Retrieval: A Vision-Language Perspective score 4
  关键词(1): vision-language; 顶会接收: NeurIPS
• Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling score 4
  关键词(2): serving, edge; 顶会接收: NeurIPS
• From Migration to Calibration: Preserving Agent Capabilities across Models, Jurisdictions, and Scale score 4
  机构: Tsinghua; 关键词(4): scaling, deployment, DPO, GRPO
• Modeling Whole-Slide Images as Dynamic Tumor Microenvironment Fields score 3
  顶会接收: NeurIPS
• Preference-Guided Adaptation for Open-Vocabulary Semantic Segmentation via Prompt Disagreement score 3
  顶会接收: NeurIPS
• Papers Without Code: Availability of GitHub Repositories Linked in *CL Publications score 3
  机构: MIT
• In-game Toxic Detection: Bi-directional Representations with Attention Residuals score 3
  顶会接收: AAAI