Learning Data Augmentation with Online Bilevel Optimization for Image Classification

Data augmentation is a key practice in machine learning for improving\ngeneralization performance. However, finding the best data augmentation\nhyperparameters requires domain knowledge or a computationally demanding\nsearch. We address this issue by proposing an efficient approach to\nautomatically train a network that learns an effective distribution of\ntransformations to improve its generalization. Using bilevel optimization, we\ndirectly optimize the data augmentation parameters using a validation set. This\nframework can be used as a general solution to learn the optimal data\naugmentation jointly with an end task model like a classifier. Results show\nthat our joint training method produces an image classification accuracy that\nis comparable to or better than carefully hand-crafted data augmentation. Yet,\nit does not need an expensive external validation loop on the data augmentation\nhyperparameters.\n