Less is More: Discovering Concise Network Explanations

TL;DR

This paper presents DCNE, a novel method for generating concise, diverse, and human-interpretable visual explanations for neural networks, improving interpretability while reducing explanation complexity.

Contribution

DCNE introduces a new approach that selects the most important explanations, building on CRP, to produce fewer, more comprehensible network explanations with minimal quality loss.

Findings

DCNE produces 1/30 of CRP's explanations.

DCNE maintains explanation quality with a slight reduction.

DCNE offers a better trade-off between conciseness and completeness.

Abstract

We introduce Discovering Conceptual Network Explanations (DCNE), a new approach for generating human-comprehensible visual explanations to enhance the interpretability of deep neural image classifiers. Our method automatically finds visual explanations that are critical for discriminating between classes. This is achieved by simultaneously optimizing three criteria: the explanations should be few, diverse, and human-interpretable. Our approach builds on the recently introduced Concept Relevance Propagation (CRP) explainability method. While CRP is effective at describing individual neuronal activations, it generates too many concepts, which impacts human comprehension. Instead, DCNE selects the few most important explanations. We introduce a new evaluation dataset centered on the challenging task of classifying birds, enabling us to compare the alignment of DCNE's explanations to those of human expert-defined ones. Compared to existing eXplainable Artificial Intelligence (XAI) methods, DCNE has a desirable trade-off between conciseness and completeness when summarizing network explanations. It produces 1/30 of CRP's explanations while only resulting in a slight reduction in explanation quality. DCNE represents a step forward in making neural network decisions accessible and interpretable to humans, providing a valuable tool for both researchers and practitioners in XAI and model alignment.