Using hard examples for data augmentation, a team of researchers created two novel algorithms for self-supervised pretraining of image representation learning models. Learn how hard examples lead to stronger learning signals and state-of-the-art performance: https://aka.ms/AAbc4he | Microsoft Research