Using a cooperative genetic algorithm population to train hidden neural network weights

Author

Sean Lane Fuller

Date of Award

5-1993

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Bruce Whitehead

Committee Members

Alfonso Pujol, Dinesh Mehta

Abstract

In a cooperative genetic neural network the genetic algorithm optimizes the hidden weights in a 3 layer feed-forward neural network. Perceptron training optimizes the output layer of the network and the trained weights of the output layer are used as a measure of fitness for each weight vector in the hidden layer. The hidden weight vectors compete through genetic evolution to produce an optimal transformation for the output layer. Common weight vectors in the output layer provide a sharing function between the hidden weight vectors. This produces a cooperative genetic environment that allows the total population to find the optimal hidden weight matrices. This approach is simulated using different benchmark datasets. This neural network and genetic algorithm hybrid achieves performance during testing comparable to the backpropagation algorithm on these datasets in terms of accuracy and elapsed time.

Recommended Citation

Fuller, Sean Lane, "Using a cooperative genetic algorithm population to train hidden neural network weights. " Master's Thesis, University of Tennessee, 1993.
https://trace.tennessee.edu/utk_gradthes/11532