Learning algorithm for multilayer morphological perceptron using cartesian genetic programming

This article describes an algorithm that uses Cartesian Genetic Programming (CGP) to train Multilayer Morphological Perceptron (MLMP). The resulting model consists of the morphological neural network itself, which is able to classify patterns received as the inputs for the nodes. CGP is used to evolve a graph of nodes, where the resulting graph represents the network connectivity, and node functions consist of the computational model of the morphological perceptron. The proposed algorithm evolves topology and connection weights simultaneously producing complex neural networks whose decision boundaries are able to classify patterns in open regions, as well as patterns in clustered regions. These results in a better classification of non-clustered patterns, as well as clustered patterns, over previously presented evolutionary algorithms proposed for morphological neural networks.