Improving Diffusion Models's Data-Corruption Resistance using Scheduled Pseudo-Huber Loss
This addresses data corruption issues for diffusion model users, offering an incremental improvement over existing methods.
The paper tackles the problem of diffusion models being vulnerable to outliers in training data by proposing a scheduled pseudo-Huber loss function, which improves robustness to corrupted datasets in image and audio domains while preserving generation quality.
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.