Quantum-inspired evolutive algorithms represent one of the most recent
advances in evolutionary computation. They have been used to evolve artificial neural
networks in contrast to traditional training methods based on the decreasing gradient
and backpropagation. The algorithm used in this work was to Quantum-Inspired
Evolucionary algorithm for Neuro-evolution with binary-real representation (QIENBR)
developed for modeling neural networks of multilayer perceptron and used to
problems of pattern classification. The regularization technique called weight decay aims to make the weights who have little or no influence on the network, take values close to zero, with this reducing
network complexity and improving their ability to generalize. This work implements weight decay in the algorithm QIEN-BR and to evaluates the performance of the new algorithm, we will use three benchmark cases of classification. The results will be compared with those obtained with the original algorithm.
|