Approximation properties of some two-layer feedforward neural networks

Abstract

In this article, we present a multiyariate two-layer feedforward neural networks that approximate continuos functions defined on $[0,1]^d$. We show that the $L_1$ error of approximation is asymptotically proportional to the modulus of continuity of the underlying function taken at $\sqrt{d}/n$, where n is the number of function values used.