PAGANDA: An Adaptive Task-Independent Automatic Data Augmentation

Anonymous

Abstract: Data augmentation is a useful technique to enlarge the size of the training set and prevent overfitting for different machine learning tasks when training data is scarce. However, current data augmentation techniques rely heavily on human design and domain knowledge, and existing automated approaches are yet to fully exploit the latent features in the training dataset. In this paper we propose \textit{Parallel Adaptive GAN Data Augmentation}(PAGANDA), where the training set adaptively enriches itself with sample images automatically constructed from Generative Adversarial Networks (GANs) trained in parallel. We demonstrate by experiments that our data augmentation strategy, with little model-specific considerations, can be easily adapted to cross-domain deep learning/machine learning tasks such as image classification and image inpainting, while significantly improving model performance in both tasks. Our source code and experimental details are available at \url{https://github.com/miaojiang1987/k-folder-data-augmentation-gan/}.

TL;DR: We present an automated adaptive data augmentation that works for multiple different tasks.