A note on error bounds for function approximation using nonlinear networks

For a variety of problems concerning classification, compensation, adaptivity, identification or signal processing, results concerning the representation and approximation of nonlinear functions can be of particular interest to engineers. Here we consider a large class of functions f that map Rⁿ into the set of real or complex numbers, and we give bounds on the number of parameters needed so that f is approximated to within a prescribed degree of accuracy using a certain approximation network. Related work in the neural networks literature is also described.