Training Neural Networks 3: Augmentation and Regularization