Title :
Size reduction of a radial basis function network
Author :
Lemarié, Bernard
Author_Institution :
Service de Recherche Tech. de la Poste, Nantes, France
Abstract :
We present a method of reducing the size of a radial basis function (RBF) network. The reduction is issued from the parallelism between RBF-networks and Parzen Windows estimator. We validate the method on a handwritten digit classification task. The resulting network is compared with the k-nearest neighbour statistical method. The results appear better for the network at a much lower computational cost.
Keywords :
backpropagation; character recognition; feedforward neural nets; least mean squares methods; optimisation; Gaussian centre model; Parzen Windows estimator; backpropagation; handwritten digit classification; hidden cells; mean square error criterion; radial basis function network; size reduction; Bayesian methods; Computational efficiency; Context modeling; Electronic mail; Interpolation; Neural networks; Parallel processing; Probability; Radial basis function networks; Statistical analysis;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713924
