Definición de una red neuronal para clasificación por medio de un programa evolutivo
Artificial neural nets called Multi-layer Perceptrons (MP) are an extremely useful tool for solving classification problems. It has been shown that MPs with a single hidden layer can satisfactorily separate the classes involved in a given problem: however, the number of hidden units required is unknown as there is no exact method for calculating them. The present work describes a way, using an evolutionary program (EP) with variable length chromosomes, of finding a suitable number of hidden units, as well as the weights on all the connections, thereby defining the architecture of a MP for a given classification problem. Only the number of input and output units are considered given. The operators used by the EP are the selection and two forms of mutation, and these are specified in the paper.
Copyright (c) 2001 Revista Mexicana de Ingeniería Biomédica
This work is licensed under a Creative Commons Attribution 4.0 International License.
Upon acceptance of an article in the RMIB, corresponding authors will be asked to fulfill and sign the copyright and the journal publishing agreement, which will allow the RMIB authorization to publish this document in any media without limitations and without any cost. Authors may reuse parts of the paper in other documents and reproduce part or all of it for their personal use as long as a bibliographic reference is made to the RMIB and a copy of the reference is sent. However written permission of the Publisher is required for resale or distribution outside the corresponding author institution and for all other derivative works, including compilations and translations.