On L₂ convergence rates of radial basis functionnetworks and kernel regression estimators

摘要

The paper generalises the rates of L₂ convergence forRBF nets based on the kernel regression estimates (KRE) obtained byoptimising the empirical error with respect to the weight vector and thereceptive field size. The centers of the radial functions are placed atthe points sampled with replacement from the learning sequence. Thebounded output convergence and the rate of convergence for the RBF nethave been obtained for radial functions with noncompact support. Newresults have been obtained for the L₂ convergence rates ofKRE and RBF nets in the case of unbounded outputs