I. Introduction
The algorithmic advancements of deep neural networks (DNN) have enabled them to play irreplaceable functions in countless applications. Given enough data to train on, DNNs have the ability to learn high-level features and show performance comparable to or exceeding human’s [1]. The huge leap of accuracy in computer vision first drew the attention [2], but the hype also spread to other domains including natural language processing [3], autonomous driving [4] and game playing [5].