Preventing Dimensional Collapse in Self-Supervised Learning via Orthogonality Regularization | Read Paper on Bytez