Diffused Redundancy in Pre-trained Representations | Read Paper on Bytez