Orthogonal least squares algorithms for training multi-output radial basis function networks

A constructive learning algorithm for multioutput radial basis function networks is pre- sented. Unlike most network learning algorithms, which require a fixed network structure, this algo- rithm automatically determines an adequate radial basis function network structure during learning. By formulating the learning problem as a subset model selection, an orthogonal least- squares procedure is used to identify appropriate radial basis function centres from the network training data, and to estimate the network weights simultaneously in a very efficient manner. This algorithm has a desired property, that the selec- tion of radial basis function centres or network hidden nodes is directly linked to the reduction in the trace of the error covariance matrix. Nonlin- ear system modelling and the reconstruction of pulse amplitude modulation signals are used as two examples to demonstrate the effectiveness of this learning algorithm.