Generalization bounds for function approximation from scattered noisy data (Q1284139)

This paper concerns problems of providing error bounds for approximation of a unknown function from scattered, noisy data using certain kinds of approximating schemes. Since the data are randomly generated and noisy, the total (generalization) error necessarily involves statistical considerations in addition to the traditional approximation theoretic ones. The authors show how the generalization error can be decomposed into two parts: an approximation part that is due to the finite number of parameters of the approximation scheme used; and an estimation part that is due to the finite number of data available. They obtain general bounds of these two parts in the case where the approximating family \(H\) can be represented as a linear combination of functions drawn from some finite-dimensional class \(G\). In particular, such class \(H\) include radial basis functions with free centers and multilayer perceptrons.