Developmental neural networks for shape recognition: motivation and review

Abstract

Artificial Neural Networks (ANNs) are widely exploited in Artificial Intelligence applications. However, at present, there is a wide gap in functionality between artificial and biological neural systems. Appropriate neuron functionality, network topology and multi-network system organisation in ANNs are the subjects of much debate. This report presents arguments for modelling the developmental processes in biological neural systems as a stepping stone towards better understanding of biological neural architectures and self organisation. The authors are interested in shape recognition systems. The early vision system in mammals is proposed as an appropriate test bed for developmental modelling. The report presents an overview of biological and ANN shape processing systems. A review of developmental processes in the mammalian visual system is followed by an analysis of current research in developmental neural networks.