Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025) - View it on GitHub