Continual Learning of Diffusion Models with Generative Distillation

TL;DR

This paper introduces generative distillation to enable continual learning in diffusion models, effectively mitigating catastrophic forgetting and improving incremental learning performance with manageable computational overhead.

Contribution

The paper proposes a novel generative distillation method that distills the reverse process of diffusion models, enhancing continual learning capabilities.

Findings

Generative distillation significantly improves continual learning in diffusion models.

The approach maintains denoising quality with only modest additional computation.

It effectively mitigates catastrophic forgetting in generative replay.

Abstract

Diffusion models are powerful generative models that achieve state-of-the-art performance in image synthesis. However, training them demands substantial amounts of data and computational resources. Continual learning would allow for incrementally learning new tasks and accumulating knowledge, thus enabling the reuse of trained models for further learning. One potentially suitable continual learning approach is generative replay, where a copy of a generative model trained on previous tasks produces synthetic data that are interleaved with data from the current task. However, standard generative replay applied to diffusion models results in a catastrophic loss in denoising capabilities. In this paper, we propose generative distillation, an approach that distils the entire reverse process of a diffusion model. We demonstrate that our approach substantially improves the continual learning performance of generative replay with only a modest increase in the computational costs.