How to Create a Custom Explainable Deep Learning Model with Keras

In this article, I'll create and train the Deep Neural Network architecture proposed by Wieland Brendel and Matthias Bethge in this research paper. As they state in their paper, the model performance I achieved in this test is similar to the performance of equivalent non-explainable Deep Learning models, and the proposed model, named BagNets, is really able to provide insights into how it achieves its predictions. This tutorial assumes you know the basics of Tensorflow and Keras Layers API. What is an Explainable Deep Learning (DL) Model? There's no single definition, but here and in my daily work, I chose the definition of Holzinger and his contributors, stated in their very enlighted paper.