Artificial evolution: modelling the development of the retina

Abstract

The evolution of neural systems relies on the repeated modification of developmental programmes contained within genes. This paper proposes that to efficiently investigate artificial evolution, developmental processes must first be characterised and encoded. The development processes themselves are inherently simple, representing a form of emergent self-organisation. A model of artificial neural structure self-organisation is presented, where development occurs across different levels of scale and development processes are divided into phases. Within this framework, a Developmental Artificial Neural Network (DANN) has been implemented to model the growth of neuron-neuron connections in both 2D and 3D. The development of a biologically inspired, artifcial retina is being investigated. This is believed to be the first reported 3D DANN.