Haifeng Wang

ProxyAttn: Guided Sparse Attention via Representative Heads

Enough is as good as a feast: A Comprehensive Analysis of How Reinforcement Learning Mitigates Task Conflicts in LLMs

Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts

Knowledge-Level Consistency Reinforcement Learning: Dual-Fact Alignment for Long-Form Factuality

SimReg: Achieving Higher Convergence and Generalization in the LLM Pretraining via Embedding Similarity Regularization

A2R: An Asymmetric Two-Stage Reasoning Framework for Parallel Reasoning

Unified Multimodal Model as Auto-Encoder

Semantic Energy: Detecting LLM Hallucination Beyond Entropy

Mixture of Hidden-Dimensions: Not All Hidden-States’ Dimensions are Needed in Transformer

FlashMask: Efficient and Rich Mask Extension of FlashAttention

Towards Large-Scale In-Context Reinforcement Learning by Meta-Training in Randomized Worlds

BEVWorld: A Multimodal World Model for Autonomous Driving via Unified BEV Latent Space