Artificial Development of Biologically Plausible Neural-Symbolic Networks

Joe Townsend, Ed Keedwell, and Antony Galton

Cognitive Computation,Volume 6, Number 1, 2014, pages 18-34.

Published online 13th April 2013.

Abstract

Neural-symbolic networks are neural networks designed for the purpose of representing logic programs. One of the motivations behind this is to work towards a biologically plausible model of knowledge representation in the brain. This paper reviews work in this direction and suggests that a new direction to take would be to evolve neural-symbolic networks using artificial development, which also has some biological plausibility. This idea is supported by a review of artificial development, followed by some initial results in using artificial development to evolve a neural-symbolic SHRUTI network in order to demonstrate how the fields of neural-symbolic integration and artificial development may be integrated. The experiments were successful in evolving genomes which could develop connections between neurons in working SHRUTI networks.

Antony Galton

Last modified: Fri May 9 13:01:54 BST 2014