Bayesian approach to regularization for training task of radial basic functions network

The problem of scalar multivariable function approximation using the set of the radial basic functions is discussed. To manage the ill-posedness of the approximation problem the Tikhonov‘s regularization method is applied. The Bayesian approach is used to find the regularization coefficient. The considered algorithm makes it possible to decrease the calculation costs significantly due to replacement of the expensive iteration procedure of target function extremes search with analytic solution.