Model selection for RBF neural networks using distorter

We propose an unbiasedness criterion using distorter (UCD) which is a heuristic model selection criterion, and apply it to determining the number of hidden units of RBF networks. The criterion is defined as the difference between outputs of two RBF networks with the same architecture, the one is trained to minimize the ordinary training error and the other is trained to minimize the error between the training data and output of the network transformed by the nonlinear function called "distorter". In order to compare the performance of proposed criterion with other criteria such as AIC and NIC, we carried out some numerical simulations of the model selection.