Machine learningML & AI Methods

Score-based generative models are provably robust: an uncertainty quantification perspective

The authors demonstrate that score-based generative models are resilient to errors, using the Wasserstein uncertainty propagation theorem to explain how learning errors affect the model's quality.