Improving Denoising Diffusion Probabilistic Models via Exploiting Shared Representations

TL;DR

This paper introduces SR-DDPM, a novel approach that enhances denoising diffusion probabilistic models for multi-task image generation with limited data by leveraging shared representations and meta-learning techniques.

Contribution

The paper presents SR-DDPM, a new method that uses shared representations and task-specific layers to improve multi-task image generation with fewer samples.

Findings

Outperforms standard DDPM in FID and SSIM metrics

Effective in multi-task settings with limited data

Scales well across diverse image datasets

Abstract

In this work, we address the challenge of multi-task image generation with limited data for denoising diffusion probabilistic models (DDPM), a class of generative models that produce high-quality images by reversing a noisy diffusion process. We propose a novel method, SR-DDPM, that leverages representation-based techniques from few-shot learning to effectively learn from fewer samples across different tasks. Our method consists of a core meta architecture with shared parameters, i.e., task-specific layers with exclusive parameters. By exploiting the similarity between diverse data distributions, our method can scale to multiple tasks without compromising the image quality. We evaluate our method on standard image datasets and show that it outperforms both unconditional and conditional DDPM in terms of FID and SSIM metrics.