Complexity of Gaussian-radial-basis networks approximating smooth functions (Q998978): Difference between revisions

The authors approximate smooth functions on \(\mathbb R^d\) by Gaussian radial-basis function networks with units of varying widths, using \(L^2\)-distance with respect to the Lebesgue measure and derive upper bounds of rates of approximation in terms of the Bessel or Sobolev norm of the functions which are approximated. The estimates hold for all numbers of hidden unities and all degrees \(r>d/2\) of Bessel potentials. Moreover, consequences for tractability of multivariable approximation by these networks are discussed.