Associate Professor, Tsinghua University
3 papers at NeurIPS 2025
We propose a synergistic two-stage framework that learns to jointly reconstruct physical fields and to optimize sensor placements with theoretical guarantees, both aiming for accurate physics sensing.
We pioneer training world models through reinforcement learning with verifiable rewards (RLVR), demonstrating substantial performance gains on both language- and video-based world models.
This paper presents FlashBias to speed up computation of attention with bias, which brings 1.5x speedup for AlphaFold and 2x speedup for SwinV2.
