We provide a mathematically rigorous framework to study the use of Deep Mutual Learning in the context of conditional diffusion models for long tailed generation.

We propose an effective long-tailed semi-supervised learning method with unknown unlabeled data distribution.