Diffusion models with location-scale noise

TL;DR

This paper introduces a framework for diffusion models with non-Gaussian location-scale noise and demonstrates that Gaussian noise yields superior data generation quality compared to other distributions.

Contribution

It develops a novel framework enabling diffusion processes with non-Gaussian noise and empirically shows Gaussian noise's superiority in generative performance.

Findings

Gaussian noise outperforms other distributions in diffusion models

Framework allows reversing diffusion with non-Gaussian noise

Non-Gaussian noise generally less effective for data generation

Abstract

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).