scientific article; zbMATH DE number 4029264

From MaRDI portal
Publication:3770286





[For the entire collection see Zbl 0627.00024.] Stochastic approximation methods, e.g. stochastic gradient procedures, can be accelerated considerably by using deterministic (feasible) descent directions or more exact gradient estimations at certain iteration points. A further improvement of the convergence behavior is attained by an adaptive selection of the step sizes. Given a certain optimality criterion, a recursive procedure for the computation of optimal step sizes is derived. Several properties of the optimal step sizes are shown, and the convergence rate of the resulting optimal semi-stochastic approximation algorithm is calculated. Furthermore, the initial behavior of the optimal procedure is examined.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3770286)