Visualizing interpretations of deep neural networks

Abstract

Deep neural networks are notoriously black boxes that defy human interpretations. The lack of understanding of the decision process of neural networks erode public trust and prevent wide application of AI. In this project, we will develop a set of tools that visualize interpretations of deep neural networks, so that the general public can intuitively understand how these networks make decisions. For example, for a given prediction made by the network, we can visualize how data points in the training set affect the prediction.