Data augmentation in the learning process can be useful especially in domains like medical imaging where geometric augmentation is not very valid for the human body. It has been a key factor in creating robust neural networks for visual self-supervised learning.