CDeepEx: Contrastive Deep Explanations

Amir Feghahati, Christian R. Shelton, Michael J. Pazzani, Kevin Tang

Abstract:

We propose a method which can visually explain the classification decision of deep neural networks (DNNs). There are many proposed methods in machine learning and computer vision seeking to clarify the decision of machine learning black boxes, specifically DNNs. All of these methods try to gain insight into why the network "chose class A" as an answer. Humans, when searching for explanations, ask two types of questions. The first question is, "Why did you choose this answer?" The second question asks, "Why did you not choose answer B over A?" The previously proposed methods are either not able to provide the latter directly or efficiently.

We introduce a method capable of answering the second question both directly and efficiently. In this work, we limit the inputs to be images. In general, the proposed method generates explanations in the input space of any model capable of efficient evaluation and gradient evaluation. We provide results, showing the superiority of this approach for gaining insight into the inner representation of machine learning models.

Keywords: Deep learning, Explanation, Network interpretation, Contrastive explanation

TL;DR: A method to answer "why not class B?" for explaining deep networks

Data: [Fashion-MNIST](https://paperswithcode.com/dataset/fashion-mnist), [MNIST](https://paperswithcode.com/dataset/mnist)