PhD student, University of Alberta
1 paper at NeurIPS 2025
We propose a very sparse attention mechanism for diffusion models on T2V and T2I tasks through re-use of softmax statistics.