Is normalization indispensable for training deep neural network?

NeurIPS 2020