Dynamic Routing Between Capsules
Sara Sabour, Nicholas Frosst, Geoffrey E Hinton
Neural Nets used for computer vision typically contain many parameters that must be trained accurately. In this work, Sabour et. al. introduce a new approach that utilizes "capsules", or groups of neurons in a layer that work in concert, to restructure network activations as vectors rather than simple scalar values. The results of this new approach are remarkable, achieving state of the art performance on computer vision problems with much shallower neural networks and a fraction of the typical number of trained parameters.
Matt Hardwick is a software engineer with 10+ years of experience building systems ranging from military simulations to high-scale web applications. His interests lie in machine learning, computer perception and control systems, self-driving vehicles, and other similar real-world data science problems. Currently employed by Skuid in Chattanooga as a full-stack software engineer, he spends his day building systems to accelerate the development and deployment of enterprise-scale web applications.