Evolutionary Learning Algorithm for Morphological Perceptrons

J.L. Ortiz and R. Piñeiro (Puerto Rico)


Morphological Perception, Neural Networks, Genetic Algorithms, Learning Algorithms and Training


This paper presents a method based on evolutionary computation to train and implement multilayer morphological neural networks (MNN). The algorithm calculates network parameters such as its weights, presynaptic and post-synaptic values, and other MNN values. Morphological perceptron is a new type of feed-forward artificial neural network based on lattice algebra that can be used as a single layer or as multiple layer netwrok in which the activation function is hard limit. The network is represented using a tree structure format allowing the algorithm to perform operations such as crossover replacing or switching whole neurons between parent networks. Adaptive mutation is used as the genetic algorithm approaches convergence to fine tune network parameters final values. The algorithm uses a special fitness function based on the mean square error of the classification patterns and introduces a type of penalty function to reduce the number of redundant neurons in the solution. Several multidimensional cases were conducted and results are shown.

Important Links:

Go Back