Safe Augmentation: Learning Task-Specific Transformations from Data The repository demonstrates our approach for learning safe augmentations that do not shift the data distribution.