Adaptive Meta-Network Design Employing Genetic Algorithm Techniques

Ben B. McElroy and Gareth Howells


Weightless Neural Network, Meta-networks, Genetic Algorithm, Neural Architectures


This paper investigates the employment of a Genetic Algorithm to optimally configure the parameters of a neural network architecture. Specifically, the issue of optimising the input connectivity or “connectivity map” for a class of weightless artificial neural networks is investigated by employing a genetic algorithm to intelligently ‘search’ the significantly large problem space, thereby reducing the potential for poor results based solely on a bad network layer group. This study takes as its example the mapping of independently produced solutions to programming exercises from a diverse range of specifications to the specification from which the implementation was derived. An important component of this technique lies in the crossover and mutation algorithms within the genetic algorithm, which have been specifically designed for this system. The results show significant promise for what is a novel and challenging problem domain for artificial neural systems.

Important Links:

Go Back