ERNEAD: Training of Artificial Neural Networks Based on a Genetic Algorithm and Finite Automata Theory
No hay miniatura disponible
Fecha
2018
Autores
Ruiz-Rangel, Jonathan
Ardila Hernandez, Carlos Julio
Maradei Gonzalez, Luis
Jabba Molinares, Daladier
Título de la revista
ISSN de la revista
Título del volumen
Editor
Editorial Board
Resumen
This paper presents a variation in the algorithm EMODS (Evolutionary Metaheuristic of Deterministic Swapping), at the level of its mutation stage in order to train algorithms for
each problem. It should be noted that the EMODS metaheuristic is a novel framework that allows multi-objective optimization of combinatorial problems. The proposal for the
training of neural networks will be named ERNEAD (training of Evolutionary Neural Networks through Evolutionary Strategies and Finite Automata). The selection process consists of five phases: the initial population generation phase, the forward feeding phase of the network, the EMODS search phase, the crossing and evaluation phase, and finally the verification phase. The application of the process in the neural networks will generate sets of networks with optimal weights for a particular problem. ERNEAD algorithm was applied to two typical problems: breast cancer and flower classification, the solution of the problems were compared with solutions obtained by applying the classical Backpropagation, Conjugate Gradient and Levenberg-Marquardt algorithms. The analysis of the results indicated that ERNEAD produced more precise solutions than the ones thrown by the classic algorithms.
Descripción
Palabras clave
Finite Deterministic Automaton, Artificial Neural Networks, Genetic Algorithm, EMODS, Backpropagation Algorithm, Conjugate Gradient Algorithm, Levenberg-Marquardt Algorithm