Data Augmentation with Generative Neural Networks
Supervised machine learning models have been successfully used in computer vision, speech recognition and natural language. However, data related issues are often responsible for the poor performance of these models. Specifically, overfitting due to limited data has been a common issue for different machine learning tasks. In training tasks, especially some Deep Learning training with high dimensional data, we can only retrieve a limited amount of data. To address this issue, we investigate data augmentation(DA) technique that enlarges the size of the training, although existing data augmentation solutions cannot fully exploit the latent features in the training dataset. In this paper, we propose an adaptive DA strategy based on generative models. The training set adaptively enriches itself and sample images automatically constructed from deep generative models, which are trained in a cluster-based environment to exploit latent factors. We demonstrate that the proposed adaptive data augmentation for Deep Learning significantly improves the model performance, where two applications are validated with image classification and image inpainting.