Group Equivariant Capsule Networks

TL;DR

This paper introduces group equivariant capsule networks that guarantee equivariance and invariance properties, connecting capsule networks with group convolutional networks to enhance interpretability and control over neural network representations.

Contribution

It presents a generic routing algorithm on group elements and links capsule networks with group convolutional networks, enabling equivariance, invariance, and interpretable representations.

Findings

Provides a routing algorithm ensuring equivariance and invariance.

Connects capsule networks with group convolutional networks.

Enables sparse group convolution evaluation and interpretable capsules.

Abstract

We present group equivariant capsule networks, a framework to introduce guaranteed equivariance and invariance properties to the capsule network idea. Our work can be divided into two contributions. First, we present a generic routing by agreement algorithm defined on elements of a group and prove that equivariance of output pose vectors, as well as invariance of output activations, hold under certain conditions. Second, we connect the resulting equivariant capsule networks with work from the field of group convolutional networks. Through this connection, we provide intuitions of how both methods relate and are able to combine the strengths of both approaches in one deep neural network architecture. The resulting framework allows sparse evaluation of the group convolution operator, provides control over specific equivariance and invariance properties, and can use routing by agreement instead of pooling operations. In addition, it is able to provide interpretable and equivariant representation vectors as output capsules, which disentangle evidence of object existence from its pose.