Structure Preserving Diffusion Models

Haoye Lu,Spencer Szabados,Yaoliang Yu

2024-03-01

Abstract:Diffusion models have become the leading distribution-learning method in recent years. Herein, we introduce structure-preserving diffusion processes, a family of diffusion processes for learning distributions that possess additional structure, such as group symmetries, by developing theoretical conditions under which the diffusion transition steps preserve said symmetry. While also enabling equivariant data sampling trajectories, we exemplify these results by developing a collection of different symmetry equivariant diffusion models capable of learning distributions that are inherently symmetric. Empirical studies, over both synthetic and real-world datasets, are used to validate the developed models adhere to the proposed theory and are capable of achieving improved performance over existing methods in terms of sample equality. We also show how the proposed models can be used to achieve theoretically guaranteed equivariant image noise reduction without prior knowledge of the image orientation.

Machine Learning,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The research of this paper is on the preservation of structural diffusion models, aiming to solve how to preserve additional structure of data, such as group symmetry, when learning distributions. By ensuring symmetry preservation in the diffusion step through theoretical conditions, the paper proposes a series of symmetric equivariant diffusion models that can learn the inherent symmetric distribution. Experimental results show that these models outperform existing methods in terms of sample quality and preservation of invariance, especially for symmetry tasks that require theoretical guarantees, such as image denoising and medical image processing, to ensure result consistency and reduce potential biases or errors.