Bing Wang
6 papers · Latest:
Prefix Teach, Suffix Fade: Local Teachability Collapse in Strong-to-Weak On-Policy Distillation
A new on-policy distillation method, "Prefix Teach, Suffix Fade," improves strong-to-weak model training by focusing supervision on locally teachable trajectory segments.
AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs
AGoQ introduces novel 4-bit activation and 8-bit gradient quantization to significantly reduce memory and speed up distributed LLM training.
OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation
OneVL introduces a unified VLA and World Model framework, achieving state-of-the-art latent Chain-of-Thought reasoning at real-time speed.
PhysInOne: Visual Physics Learning and Reasoning in One Suite
PhysInOne is a new large-scale dataset with 2 million videos and detailed annotations for training AI in physics-grounded visual reasoning.
Beyond Dense Connectivity: Explicit Sparsity for Scalable Recommendation
SSR introduces explicit sparsity to recommender systems, outperforming dense models by filtering low-utility connections for better scalability and performance.
UniDriveVLA: Unifying Understanding, Perception, and Action Planning for Autonomous Driving
UniDriveVLA unifies autonomous driving tasks by decoupling perception and reasoning with expert Mixture-of-Transformers, achieving SOTA performance.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.