Equivariant score-based generative models provably learn distributions with symmetries efficiently

TL;DR

This paper provides the first theoretical analysis of equivariant score-based generative models, demonstrating their efficiency in learning symmetric distributions and comparing equivariant bias with data augmentation.

Contribution

It offers new theoretical guarantees for equivariant SGMs, showing they can learn symmetrized distributions without data augmentation and quantifying the impact of non-equivariant models.

Findings

Equivariant SGMs have improved Wasserstein-1 generalization bounds.

Equivariant vector fields can learn symmetrized distributions without data augmentation.

Non-equivariant vector fields lead to worse generalization bounds.

Abstract

Symmetry is ubiquitous in many real-world phenomena and tasks, such as physics, images, and molecular simulations. Empirical studies have demonstrated that incorporating symmetries into generative models can provide better generalization and sampling efficiency when the underlying data distribution has group symmetry. In this work, we provide the first theoretical analysis and guarantees of score-based generative models (SGMs) for learning distributions that are invariant with respect to some group symmetry and offer the first quantitative comparison between data augmentation and adding equivariant inductive bias. First, building on recent works on the Wasserstein-1 ($\mathbf{d}_1$) guarantees of SGMs and empirical estimations of probability divergences under group symmetry, we provide an improved $\mathbf{d}_1$ generalization bound when the data distribution is group-invariant. Second, we describe the inductive bias of equivariant SGMs using Hamilton-Jacobi-Bellman theory, and rigorously demonstrate that one can learn the score of a symmetrized distribution using equivariant vector fields without data augmentations through the analysis of the optimality and equivalence of score-matching objectives. This also provides practical guidance that one does not have to augment the dataset as long as the vector field or the neural network parametrization is equivariant. Moreover, we quantify the impact of not incorporating equivariant structure into the score parametrization, by showing that non-equivariant vector fields can yield worse generalization bounds. This can be viewed as a type of model-form error that describes the missing structure of non-equivariant vector fields. Numerical simulations corroborate our analysis and highlight that data augmentations cannot replace the role of equivariant vector fields.