A Distributed Genetic Algorithm for Neural Network Design and Training
S. Oliker
M. Furst
Dept. of Electrical Engineering--Systems,
Faculty of Engineering, Tel Aviv University,
Ramat Aviv 69978, Israel
O. Maimon
Dept. of Industrial Engineering,
Faculty of Engineering, Tel Aviv University,
Ramat Aviv 69978, Israel
Abstract
A new approach for designing and training neural networks is developed using a distributed genetic algorithm. A search for the optimal architecture and weights of a neural network comprising binary, linear threshold units is performed. For each individual unit, we look for the optimal set of connections and associated weights under the restriction of a feedforward network structure. This is accomplished with the modified genetic algorithm, using an objective function---fitness---that considers, primarily, the overall network error; and, secondarily, using the unit's possible connections and weights that are preferable for continuity of the convergence process. Examples are given showing the potential of the proposed approach.
