Hierarchically branched diffusion models leverage dataset structure for class-conditional generation

TL;DR

This paper introduces hierarchically branched diffusion models that utilize dataset structure for improved class-conditional generation, enabling continual learning, analogy-based transmutation, and enhanced interpretability across diverse scientific datasets.

Contribution

The paper proposes a novel branched diffusion framework that learns separate reverse processes for each hierarchy branch, leveraging dataset structure for better class-conditional generation.

Findings

Effective on multiple scientific datasets across modalities

Enables continual learning of new classes

Provides interpretability into the generation process

Abstract

Class-labeled datasets, particularly those common in scientific domains, are rife with internal structure, yet current class-conditional diffusion models ignore these relationships and implicitly diffuse on all classes in a flat fashion. To leverage this structure, we propose hierarchically branched diffusion models as a novel framework for class-conditional generation. Branched diffusion models rely on the same diffusion process as traditional models, but learn reverse diffusion separately for each branch of a hierarchy. We highlight several advantages of branched diffusion models over the current state-of-the-art methods for class-conditional diffusion, including extension to novel classes in a continual-learning setting, a more sophisticated form of analogy-based conditional generation (i.e. transmutation), and a novel interpretability into the generation process. We extensively evaluate branched diffusion models on several benchmark and large real-world scientific datasets spanning many data modalities.