Weighted Means in Stochastic Approximation of Minima

From MaRDI portal





Continuing their own work [Math. Methods Stat. 5, No. 1, 32-60 (1996; Zbl 0858.62070)] the authors derive optimal rates of convergence for various multi-dimensional stochastic approximation procedures of the Kiefer-Wolfowitz type which aim at finding the extremum of a multivariate response function. The procedures considered here are based on the averaging method introduced by Ruppert and Polyak for the Robbins-Monro situation which was one of the major developments in stochastic approximation during the last years. In contrast to the standard Kiefer-Wolfowitz process no prior knowledge is required for the curvature at the extremum. The method allows for a wide flexibility in choosing the estimator of the gradient which covers averaged versions of the \textsl{random direction} and the \textsl{simultaneous perturbation stochastic approximation}. A computation of the resulting asymptotic mean squared errors exhibits a stable behaviour for the present procedures based on the averaging method compared to standard competitors.




Cited in
(28)








This page was built for publication: Weighted Means in Stochastic Approximation of Minima

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4377400)