Developmental artificial neural networks for shape recognition: a model of the retina

Abstract

There has been recent interest in mimicking the self-organising processes of biological development to design artificial neural networks. An a priori decision must however be made as to the degree of biological detail modelled (ie neuron vs molecule). This report therefore presents a theoretical framework aimed at investigating biological self-organisation. A model of self-organisation of structure is proposed in terms of development across different levels of scale and the division of self-organising processes into phases. This initial study focuses on the design of a Developmental Artificial Neural Network (DANN) to investigate the growth of neuron-neuron connections. A model of a biological retina ia used as the target structure for the self-organising process. The development of general 2D and 3D retina-like structures are reported. This is believed to be the first report of a 3D DANN.