Regularised OLS algorithm with fast implementation for training multi-output radial basis function networks

The paper presents an approach for training multi-output radial basis function (RBF) networks by combining subset selection with regularisation. A regularised orthogonal least squares (ROLS) algorithm is derived, which is capable of constructing parsimonious networks that generalise well. A fast implementation of the ROLS algorithm further reduces computational requirements significantly. System identification is used as an example to demonstrate the effectiveness of this training algorithm