On L2 convergence rates of radial basis function networks and kernel regression estimators

The paper generalises the rates of L₂ convergence for RBF nets based on the kernel regression estimates (KRE) obtained by optimising the empirical error with respect to the weight vector and the receptive field size. The centers of the radial functions are placed at the points sampled with replacement from the learning sequence. The bounded output convergence and the rate of convergence for the RBF net have been obtained for radial functions with noncompact support. New results have been obtained for the L₂ convergence rates of KRE and RBF nets in the case of unbounded outputs