Learning to Remove: Towards Isotropic Pre-trained BERT Embedding | Read Paper on Bytez