ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Schedule

概要

arXiv:2601.18681v2 Announce Type: replace-cross Abstract: We consider time discretization for score-based diffusion models to generate samples from a learned reverse-time dynamic on a finite grid. Uniform and hand-crafted grids can be suboptimal given a budget on the number of time steps. We introd…