Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules

TL;DR

Population Based Augmentation (PBA) offers an efficient method for learning augmentation schedules that match AutoAugment's performance with significantly less computational cost, improving neural network generalization.

Contribution

The paper introduces PBA, a novel augmentation algorithm that generates dynamic augmentation schedules, reducing computational requirements compared to existing methods like AutoAugment.

Findings

PBA achieves state-of-the-art results on CIFAR-10 with 1.46% error.

PBA matches AutoAugment's performance on CIFAR-10, CIFAR-100, and SVHN.

PBA requires three orders of magnitude less compute than AutoAugment.

Abstract

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