Diffusion Models (DMs) are powerful generative models that add Gaussian noise to the data and learn to remove it. We wanted to determine which noise distribution (Gaussian or non-Gaussian) led to better generated data in DMs. Since DMs do not work by design with non-Gaussian noise, we built a framework that allows reversing a diffusion process with non-Gaussian location-scale noise. We use that framework to show that the Gaussian distribution performs the best over a wide range of other distributions (Laplace, Uniform, t, Generalized-Gaussian).
@article{arxiv.2304.05907,
title = {Diffusion models with location-scale noise},
author = {Alexia Jolicoeur-Martineau and Kilian Fatras and Ke Li and Tal Kachman},
journal= {arXiv preprint arXiv:2304.05907},
year = {2023}
}