A fuzzy self-organized backpropagation using nervous system

We consider a fuzzy multilayer neural network with a single hidden layer. A conventional backpropagation learning algorithm used widely in multilayer neural network has a possibility of local minima due to the inadequate initial weights and the insufficient number of hidden nodes. So we propose a fuzzy self-organized backpropagation algorithm that self-generates hidden nodes by a compound method of supervised learning and unsupervised learning. From the input layer to the hidden layer, a modified fuzzy ART is used to produce nodes. A winner-take-all method is adopted to the connection weight adaption, so that stored pattern for some pattern becomes updated. In simulation results, our method reduces the possibility of local minima and improves learning speed and paralysis than conventional error backpropagation.