G-Prop-III: global optimization of Multilayer Perceptrons using an evolutionary algorithm
Pedro Ángel Castillo, Víctor M. Rivas, Juan Julián Merelo, Jesús González, A. Prieto, G. Romero · 1999
This paper proposes a new version of a method (G-Prop-III, genetic backpropagation) that attempts to solve the problem of finding appropriate initial weights and learning parameters for a single hidden layer Multilayer Perceptron (MLP) by combining a genetic algorithm (GA) and backpropagation (BP). The GA selects the initial weights and the learning rate of the network, and changes the number of neurons in the hidden layer through the application of specific genetic operators. Besides, this new version of the algorithm includes BP training as a mutation operator. G-Prop-III combines the advantages of the global search performed by the GA over the MLP parameter space and the local search of the BP algorithm. The application of the G-Prop-III algorithm to several realworld and benchmark problems shows that MLPs evolved using G-Prop-III are smaller and achieve a higher level of generalization than other perceptron training algorithms, such as QuickPropagation or RPROP, and other evolutive...