Improving Denoising Diffusion Probabilistic Models via Exploiting Shared Representations
This work addresses the problem of data efficiency in multi-task image generation for researchers and practitioners using DDPMs, representing an incremental improvement by adapting few-shot learning techniques to enhance existing models.
The paper tackles the challenge of multi-task image generation with limited data for denoising diffusion probabilistic models (DDPM) by proposing SR-DDPM, which leverages representation-based techniques from few-shot learning to learn from fewer samples across tasks, resulting in improved performance on standard image datasets with better FID and SSIM metrics compared to unconditional and conditional DDPM.
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.