Nonlinear feature extraction with radial basis functions using a weighted multidimensional scaling stress measure

We investigate radial basis functions for nonlinear feature extraction. The parameters of the transformation are determined by minimising a loss term (similar to stress in multidimensional scaling) that weights components of the loss by a nonlinear function of the dissimilarities. Several forms for the nonlinear function are considered and an optimisation scheme based on iterative majorisation is used to determine the parameter values. The technique is illustrated on two data sets