Continual Learning of Diffusion Models with Generative Distillation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Masip, Sergi, Rodriguez, Pau, Tuytelaars, Tinne, van de Ven, Gido M.
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929739783995392
author Masip, Sergi
Rodriguez, Pau
Tuytelaars, Tinne
van de Ven, Gido M.
author_facet Masip, Sergi
Rodriguez, Pau
Tuytelaars, Tinne
van de Ven, Gido M.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2311_14028
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Continual Learning of Diffusion Models with Generative Distillation
Masip, Sergi
Rodriguez, Pau
Tuytelaars, Tinne
van de Ven, Gido M.
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
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.
title Continual Learning of Diffusion Models with Generative Distillation
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.14028