Align Your Steps: Optimizing Sampling Schedules in Diffusion Models

arXiv:2404.14507v1 Announce Type: cross
Abstract: Diffusion models (DMs) have established themselves as the state-of-the-art generative modeling approach in the visual domain and beyond. A crucial drawback of DMs is their slow sampling speed, relying on many sequential function evaluations through large neural networks. Sampling from DMs can be seen as solving a differential equation through a discretized set of noise levels known as the sampling schedule. While past works primarily focused on deriving efficient solvers, little attention has been given …

abstract art arxiv beyond cs.cv cs.lg diffusion diffusion models domain function generative generative modeling modeling networks neural networks sampling speed state through type visual