Neurite Networks: the genetic programming of cellular automata based neural nets which GROW

Abstract

This paper proposes a new branch of neural networks, called "neurite networks", it is a neural network that grows, i.e. it has an embryological component. The artificial neurite network introduced is based on a cellular automata (CA) network whose branchings are genetically programmed (i.e. they are grown under the control of a genetic algorithm). A sequence of CA signals is sent down the middle of a CA "trail". When a signal hits the end of a trail, it can make the trail extend, turn left, turn right, branch left, branch right, split, etc., depending upon the state of the CA signal. These signal sequences are treated as the chromosomes of a genetic algorithm. Once the CA network is formed, a second set of CA state transition rules is switched on to make it behave like a neural network. The fitness of this CA based neural network is measured in terms of how well it controls some behavior of a biological robot.