Improving Diffusion Models's Data-Corruption Resistance using Scheduled Pseudo-Huber Loss

TL;DR

This paper introduces a time-dependent pseudo-Huber loss function for diffusion models, enhancing robustness to outliers and dataset corruption in image and audio generation tasks without needing data filtering.

Contribution

It proposes a novel pseudo-Huber loss with a time-dependent parameter, improving diffusion models' resistance to data corruption while maintaining high-quality outputs.

Findings

Better performance on corrupted datasets in image and audio domains

Enhanced robustness to outliers without data filtering

Potential for resisting dataset corruption in diffusion training

Abstract

Diffusion models are known to be vulnerable to outliers in training data. In this paper we study an alternative diffusion loss function, which can preserve the high quality of generated data like the original squared $L_{2}$ loss while at the same time being robust to outliers. We propose to use pseudo-Huber loss function with a time-dependent parameter to allow for the trade-off between robustness on the most vulnerable early reverse-diffusion steps and fine details restoration on the final steps. We show that pseudo-Huber loss with the time-dependent parameter exhibits better performance on corrupted datasets in both image and audio domains. In addition, the loss function we propose can potentially help diffusion models to resist dataset corruption while not requiring data filtering or purification compared to conventional training algorithms.