Digital Talk

Abstract

Modern Convolutional Neural Networks (CNN's) are learning the weights of vast numbers of convolutional operators. In this talk, we raise the fundamental question if this is actually necessary. We show that even in the extreme case of only randomly initializing and never updating spatial filters, certain CNN architectures can be trained to surpass the accuracy of standard training. Additionally, we propose a novel weight sharing mechanism, which allows sharing of a single weight tensor between all spatial convolution layers to massively reduce the number of weights in these settings.