Reset Networks: Emergent Topography by Composition of Convolutional Neural Networks

Abstract

We introduce the Reset model, a composition of neural networks - typically several levels of convolutional neural networks - whose outputs at one level are gathered and reshaped into a spatial input for the next level. We show that units in Reset networks self-organize into clusters when trained on MNIST, Fashion MNIST, CIFAR-10 and CIFAR-100. We then show that a stronger type of self-organization, reminiscent of the topography found for numbers in parietal cortex, arises when number images are mapped onto developmentally realistic number codes. We outline the implications of this model for theories of the cortex and developmental neuroscience.