Understanding Augmentation-based Self-Supervised Representation Learning via RKHS Approximation and Regression