Extracting Interpretable Concept-Based Decision Trees from CNNs

TL;DR

This paper introduces a method to interpret CNNs by extracting concept-based decision trees from hidden layer activations, enabling better understanding of model reasoning with human-understandable concepts.

Contribution

The paper presents a novel approach to derive interpretable decision trees from CNNs that reveal concept importance and interactions, enhancing model transparency.

Findings

Decision trees accurately represent CNN classifications at low depths.

The method enables human-in-the-loop understanding of CNN concepts.

Extracted trees highlight concept importance and interactions.

Abstract

In an attempt to gather a deeper understanding of how convolutional neural networks (CNNs) reason about human-understandable concepts, we present a method to infer labeled concept data from hidden layer activations and interpret the concepts through a shallow decision tree. The decision tree can provide information about which concepts a model deems important, as well as provide an understanding of how the concepts interact with each other. Experiments demonstrate that the extracted decision tree is capable of accurately representing the original CNN's classifications at low tree depths, thus encouraging human-in-the-loop understanding of discriminative concepts.