Activity-based pruning in developmental artificial neural networks

The expression of development programmes encoded in genes leads to the wide range of structures and functionality found in biology. Neural development is a highly adaptive form of self-organisation. This arises from the ability of neural systems to fine-tune their structure to both internal and external environments after an initial over-production of neural elements, Such structural refinement is a necessary part of self-organisation. This paper reports a biologically motivated model that regulates and refines the growth of neuron-to-neuron interconnections in a 3D artificial neural network. Our inspiration is spontaneous neural activity and the multiple roles of neurotrophic factors in biological systems, The model implements self-regulating growth of neurons, and competitive mechanisms which remove complete neuronal trees and differentiate between individual synapses.