Capsule Networks

Nigel Schuster

Capsule Networks are a novel approach for processing data with a neural network. They were first introduced a couple years ago, but gained some popularity after Hinton et al. published [Dynamic Routing between Capsules] (https://arxiv.org/abs/1710.09829). We researched the performance of CapsNets during our Deep Learning class. Our main findings were, that CapsNets exhibit some desirable properties, such as resistance to Adversarial Attacks and better adaptiveness, but are orders of magnitude slower. You can find the full report here.

Published on 2018-02-20.