Visualizing Deep Neural Network Decisions: Prediction Difference Analysis

TL;DR

This paper introduces the prediction difference analysis method for visualizing deep neural network decisions, highlighting image regions that support or oppose specific classifications to improve interpretability.

Contribution

The paper proposes a novel visualization technique that overcomes limitations of previous methods, enhancing understanding of neural network decision processes.

Findings

Effective visualization on natural images (ImageNet)

Insightful analysis of medical images (MRI scans)

Improved interpretability of deep models

Abstract

This article presents the prediction difference analysis method for visualizing the response of a deep neural network to a specific input. When classifying images, the method highlights areas in a given input image that provide evidence for or against a certain class. It overcomes several shortcoming of previous methods and provides great additional insight into the decision making process of classifiers. Making neural network decisions interpretable through visualization is important both to improve models and to accelerate the adoption of black-box classifiers in application areas such as medicine. We illustrate the method in experiments on natural images (ImageNet data), as well as medical images (MRI brain scans).