Interpretable Graph Capsule Networks for Object Recognition

TL;DR

This paper introduces Graph Capsule Networks (GraCapsNets), a new interpretable model that replaces routing with attention-based graph pooling, improving classification, interpretability, and robustness over traditional CapsNets.

Contribution

The paper presents a novel GraCapsNet architecture with built-in interpretability by replacing routing with attention-based graph pooling, enhancing performance and robustness.

Findings

GraCapsNets outperform CapsNets in classification accuracy.

GraCapsNets exhibit improved adversarial robustness.

GraCapsNets maintain disentangled representations and affine invariance.

Abstract

Capsule Networks, as alternatives to Convolutional Neural Networks, have been proposed to recognize objects from images. The current literature demonstrates many advantages of CapsNets over CNNs. However, how to create explanations for individual classifications of CapsNets has not been well explored. The widely used saliency methods are mainly proposed for explaining CNN-based classifications; they create saliency map explanations by combining activation values and the corresponding gradients, e.g., Grad-CAM. These saliency methods require a specific architecture of the underlying classifiers and cannot be trivially applied to CapsNets due to the iterative routing mechanism therein. To overcome the lack of interpretability, we can either propose new post-hoc interpretation methods for CapsNets or modifying the model to have build-in explanations. In this work, we explore the latter. Specifically, we propose interpretable Graph Capsule Networks (GraCapsNets), where we replace the routing part with a multi-head attention-based Graph Pooling approach. In the proposed model, individual classification explanations can be created effectively and efficiently. Our model also demonstrates some unexpected benefits, even though it replaces the fundamental part of CapsNets. Our GraCapsNets achieve better classification performance with fewer parameters and better adversarial robustness, when compared to CapsNets. Besides, GraCapsNets also keep other advantages of CapsNets, namely, disentangled representations and affine transformation robustness.