A Licensing Information

CIFAR-10 [ 33 ] is licensed under the MIT license. We sample 20% data from Split CIFAR-10 training set for validation, and we use grid-search on this validation set to help us select the mentioned best hyperparameters. We use the same set of hyperparameters for both datasets. Activation of layers are stored in memory for the backward pass. We omit the FLOPs needed for batch normalization and cross entropy.