A Constructive Algorithm for the Training of a Multilayer Perceptron Based on the Genetic Algorithm

In this paper, we propose a genetic algorithm for the training and construction of a multilayer perceptron. It is based on the genetic algorithm and works on a layer-bylayer basis. For each layer, it automatically chooses the number of neurons required, computes the synaptic weights between the present layer of neurons and the next layer, and gives a set of training patterns for the succeeding layer. The algorithm presented here constructs networks with neurons implementing a threshold activation function. This architecture is suitable for classification problems with a single binary output. The method is applied to the XOR problem, the n-bit parity problems, as well as the MONK's problems, and its performance is found to be comparable to that of other techniques. 1 Introduction Recently there has been much study into the training of multilayer perceptrons using constructive algorithms [1, 2, 3, 4]. This type of algorithm is useful, as there are still no practical methods for obtain..