Optimizing Data Augmentation Policy Through Random Unidimensional Search

TL;DR

This paper introduces Random Unidimensional Augmentation, a method that significantly reduces the computational cost of optimizing data augmentation policies in deep learning, achieving comparable results with fewer training runs.

Contribution

The authors propose a novel, efficient search method for data augmentation policies that requires only 6 training runs, drastically reducing computational overhead compared to existing approaches.

Findings

Achieves similar performance to state-of-the-art methods with fewer trainings

Reduces computational cost of augmentation policy search by over 80%

Source code is publicly available for reproducibility

Abstract

It is no secret amongst deep learning researchers that finding the optimal data augmentation strategy during training can mean the difference between state-of-the-art performance and a run-of-the-mill result. To that end, the community has seen many efforts to automate the process of finding the perfect augmentation procedure for any task at hand. Unfortunately, even recent cutting-edge methods bring massive computational overhead, requiring as many as 100 full model trainings to settle on an ideal configuration. We show how to achieve equivalent performance using just 6 trainings with Random Unidimensional Augmentation. Source code is available at https://github.com/fastestimator/RUA/tree/v1.0