PhD student, Massachusetts Institute of Technology
3 papers at NeurIPS 2025
This paper presents FlashBias to speed up computation of attention with bias, which brings 1.5x speedup for AlphaFold and 2x speedup for SwinV2.