In the original BatchNorm paper, the authors Sergey Ioffe and Christian Szegedy of Google introduced a method to address a phenomenon called internal covariate shift.  This occurs because the distribution of each layer’s inputs changes during training, as the parameters of the previous layers change. This slows down the training by requiring lower learning rates…

The post What Are The Alternatives To Batch Normalization In Deep Learning? appeared first on Analytics India Magazine.