Critical review on Data augmentation instead of explicit regularization by Alex Hernández-García and Peter König with code and paper - Deep Learning : Statistical perspective (Fall 2020)
Abstract
“Data augmentation instead of explicit regularization” by Alex Hernández-García and Peter König(2019) suggested new possibility and future of data augmentation. Normally, it was considered as one of the methods to supplement the lack of training datasets. However, what the paper had proposed was that this can actually replace explicit regularization such as weight decay and dropout. This critical review verifies the validity of the paper’s proposals regarding data augmentation methods, neural network designs and datasets. Finally, the result demonstrates that the data augmentation alone cannot always replace the explicit regularization.