Interpretable Compositional Convolutional Neural Networks

TL;DR

This paper introduces a method to transform traditional CNNs into interpretable compositional CNNs that learn meaningful visual parts without additional supervision, enhancing explainability in AI models.

Contribution

It presents a novel approach to modify CNNs for interpretability by learning filters that encode meaningful visual patterns without requiring part annotations.

Findings

Effective learning of interpretable filters

Applicable to various CNN architectures

Demonstrated improved interpretability

Abstract

The reasonable definition of semantic interpretability presents the core challenge in explainable AI. This paper proposes a method to modify a traditional convolutional neural network (CNN) into an interpretable compositional CNN, in order to learn filters that encode meaningful visual patterns in intermediate convolutional layers. In a compositional CNN, each filter is supposed to consistently represent a specific compositional object part or image region with a clear meaning. The compositional CNN learns from image labels for classification without any annotations of parts or regions for supervision. Our method can be broadly applied to different types of CNNs. Experiments have demonstrated the effectiveness of our method.