In classification tasks, classifiers trained with finite examples might generalize poorly to new data with unknown variance. For this issue, data augmentation is a successful solution where numerous artificial examples are added to training sets. In …