300 Tasks Test Visual Reasoning, Retrieval Runs 12.4× Faster

Featured

• VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning score 10
  入选 HF Daily Papers; HF 热度: 205 upvotes (+4); 有代码实现; 关键词(2): scaling, reasoning
• Code World Model: Coding Agent as World Brain score 8
  入选 HF Daily Papers; HF 热度: 21 upvotes (+4); 关键词(3): fine-tuning, coding, reasoning

Also Worth Noting

• Groundhog Bit-Flip Attack: Seeding Infinite Generation Loops in Mixture-of-Experts LLMs through Bit Flips score 4
  关键词(4): lightweight, MoE, agentic, reasoning; 顶会接收: EMNLP
• GGSS: Geodesic-Gated Spherical Steering for Inference-Time Debiasing of Generative Vision-Language Models score 4
  关键词(1): vision-language; 顶会接收: EMNLP
• Distance Is Not Enough: Forget-Retain Alignment Gap Predicts LLM Relearning Robustness score 4
  关键词(2): pruning, fine-tuning; 顶会接收: EMNLP
• PonsRAG: A Pons-Inspired RAG Bridging Cognitive Islands for Coordinated Long Narrative Reasoning score 4
  关键词(3): retrieval-augmented, RAG, reasoning; 顶会接收: EMNLP
• EgoArgus: Benchmarking VLMs as Situational Assistants for Modality-Grounded User Supports score 4
  关键词(1): deployment; 顶会接收: EMNLP
• RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval score 10
  入选 HF Daily Papers; HF 热度: 3 upvotes (+1); 有代码实现; 关键词(2): lightweight, latency; 顶会接收: EMNLP
• Towards Purified Multi-Label Test-Time Adaptation of Vision-Language Models score 4
  关键词(1): vision-language; 顶会接收: ECCV
• Think-Probe-Respond: Improving Large Language Models as Judges of Research Idea Novelty score 4
  关键词(2): lightweight, reasoning; 顶会接收: EMNLP
• GUIDE: Generative Unsupervised Chinese Query Correction via Phonetic and Visual Shared-ID Encoding score 3
  顶会接收: EMNLP
• CRAMER: Control via Request-Aware Masking for Editing Recommenders score 3
  顶会接收: ICML
• Moving Beyond More Views: Redundancy-Aware Ego-Exo Fusion for Proficiency Estimation score 3
  顶会接收: ECCV
• Unlocking Multimodal Protein Language Models at Inference Time score 3
  顶会接收: EMNLP