Approximation problems in system identification with neural networks (Q1334039)

The capability of approximating continuous functions of several variables by neural nets is investigated. Starting from a result by Cybenko concerning the denseness of linear combinations of continuous sigmoidal functions in \(L^ p (K)\), the author finally proves -- in the framework of \(L^ p\)-Tauber-Wiener functions -- that by composition of some functions of one variable, one can approximate continuous functionals defined on compact \(L^ p (K)\), and continuous operators from compact \(L^ p (K_ 1)\) to \(L^ p (K_ 2)\).