Meta Approach to Data Augmentation Optimization

TL;DR

This paper introduces a novel method for jointly optimizing image recognition models and data augmentation policies directly via gradient descent, enhancing performance without proxy tasks or extensive hyperparameter tuning.

Contribution

It presents a scalable, efficient approach that approximates policy gradients using implicit gradients and Neumann series, avoiding prior search space reductions.

Findings

Improves ImageNet classification accuracy

Enhances fine-grained recognition performance

Eliminates need for dataset-specific hyperparameter tuning

Abstract

Data augmentation policies drastically improve the performance of image recognition tasks, especially when the policies are optimized for the target data and tasks. In this paper, we propose to optimize image recognition models and data augmentation policies simultaneously to improve the performance using gradient descent. Unlike prior methods, our approach avoids using proxy tasks or reducing search space, and can directly improve the validation performance. Our method achieves efficient and scalable training by approximating the gradient of policies by implicit gradient with Neumann series approximation. We demonstrate that our approach can improve the performance of various image classification tasks, including ImageNet classification and fine-grained recognition, without using dataset-specific hyperparameter tuning.