On Learning and Learned Data Representation by Capsule Networks

TL;DR

This paper explores how capsule networks' routing influences model fitting, data representation, and generalization, revealing their advantages over traditional CNNs in capturing data structures and adaptability.

Contribution

It provides an in-depth analysis of capsule networks' routing effects, their ability to discover global data structures, and their improved adaptability to new tasks compared to CNNs.

Findings

Routing affects model certainty and fitting.

Capsule representations reveal meaningful data manifolds.

Capsules are less coupled and more adaptable than CNN neurons.

Abstract

In this work, we investigate the following: 1) how the routing affects the CapsNet model fitting; 2) how the representation using capsules helps discover global structures in data distribution, and; 3) how the learned data representation adapts and generalizes to new tasks. Our investigation yielded the results some of which have been mentioned in the original paper of CapsNet, they are: 1) the routing operation determines the certainty with which a layer of capsules pass information to the layer above and the appropriate level of certainty is related to the model fitness; 2) in a designed experiment using data with a known 2D structure, capsule representations enable a more meaningful 2D manifold embedding than neurons do in a standard convolutional neural network (CNN), and; 3) compared with neurons of the standard CNN, capsules of successive layers are less coupled and more adaptive to new data distribution.