Hands-On Deep Learning for IoT

Capsule networks

In CNNs, each layer understands an image at a much more granular level through a slow receptive field or max pooling operations. If the images have rotation, tilt, or very different shapes or orientation, CNNs fail to extract such spatial information and show very poor performance at image processing tasks. Even the pooling operations in CNNs cannot be much help against such positional invariance. This issue in CNNs has led us to the recent advancement of CapsNet through the paper entitled Dynamic Routing Between Capsules (see more at https://arxiv.org/abs/1710.09829) by Geoffrey Hinton et al:

"A capsule is a group of neurons whose activity vector represents the instantiation parameters of a specific type of entity, such ...