Pruning RBF networks with QLP decomposition

Abstract:- The radial basis function (RBF) network is the main practical alternative to the multi-layer perceptron for non-linear modeling. This paper describes a methodology to adjust predictions models, calculated from experimental data using regression with Gaussian basis functions reduced by QLP decomposition. After introducing the concepts of linear basis function models and matrix design reduced by QLP decomposition, the method is applied to RBF networks with different choices of the hidden basis function. The QLP method is effective for reducing the network size by pruning hidden nodes, resulting in a parsimonious model which accurate out-of-sample prediction for a sinusoidal test function. Simulation results showed that Gaussian basis functions produced the best results for this bench mark problem