Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules

A key challenge in leveraging data augmentation for neural network training choosing an effective augmentation policy from a large search space of operations. Properly chosen augmentation policies can lead to generalization improvements; however, state-of-the-art approaches as AutoAugment are computationally infeasible to run for the ordinary. In this paper, we introduce a new data augmentation algorithm, Population Augmentation (PBA), which generates nonstationary augmentation policy instead of a fixed augmentation policy. We show that PBA can match performance of AutoAugment on CIFAR-10, CIFAR-100, and SVHN, with three of magnitude less overall compute. On CIFAR-10 we achieve a mean test of 1.46%, which is a slight improvement upon the current-of-the-art. The code for PBA is open source and is available at://github.com/arcelien/pba.