After a long hiatus, I am back with a new episode. In today’s episode, I talk about Capsule Networks. What excites me the most about this network is that it takes the concepts of linear algebra and applies it to visual recognition. Vectors, a concept which any high school student is well aware of, is used to capture spatial relationships of objects with respect to one another. It attempts to redress a fundamental flaw of Convolutional Neural Network in the process. This paper is a brain child of Geoffrey Hinton - the pioneer of Deep Learning. I always enjoy reading his papers as they are heavily inspired by biology. This paper is nothing short of amazing! I hope you enjoy this episode.

Link to the paper: https://arxiv.org/pdf/1710.09829.pdf Link to Max Pechyonkin’s blog: https://pechyonkin.me/capsules-1/