A gradient method for unconstrained optimization in noisy environment

Authors:
NatašA Krejić;Zorana Luanin;Irena Stojkovska
Affiliations:
Department of Mathematics and Informatics, Faculty of Science, University of Novi Sad, Trg Dositeja Obradovića 4, 21000 Novi Sad, Serbia;Department of Mathematics and Informatics, Faculty of Science, University of Novi Sad, Trg Dositeja Obradovića 4, 21000 Novi Sad, Serbia;Department of Mathematics, Faculty of Natural Sciences and Mathematics, Ss. Cyril and Methodius University, Gazi Baba b.b., 1000 Skopje, Macedonia
Venue:
Applied Numerical Mathematics
Year:
2013

Citing 8
Cited 0

Stochastic algorithms with Armijo stepsizes for minimization of functions

Journal of Optimization Theory and Applications
A scaled stochastic approximation algorithm

Management Science
Almost surely convergent global optimization algorithm using noise-corrupted observations

Journal of Optimization Theory and Applications
Testing Unconstrained Optimization Software

ACM Transactions on Mathematical Software (TOMS)
The Barzilai and Borwein Gradient Method for the Large Scale Unconstrained Minimization Problem

SIAM Journal on Optimization
A Derivative-Free Algorithm for Bound Constrained Optimization

Computational Optimization and Applications
Introduction to Stochastic Search and Optimization

Introduction to Stochastic Search and Optimization
Improved Simultaneous Perturbation Stochastic Approximation and Its Application in Reinforcement Learning

CSSE '08 Proceedings of the 2008 International Conference on Computer Science and Software Engineering - Volume 01

Quantified Score

Hi-index	0.00

Visualization

Abstract

A gradient method for solving unconstrained minimization problems in noisy environment is proposed and analyzed. The method combines line-search technique with Stochastic Approximation (SA) method. A line-search along the negative gradient direction is applied while the iterates are far away from the solution and upon reaching some neighborhood of the solution the method switches to SA rule. The main issue is to determine the switching point and that is resolved both theoretically and practically. The main result is the almost sure convergence of the proposed method due to a finite number of line-search steps followed by infinitely many SA consecutive steps. The numerical results obtained on a set of standard test problems confirm theoretical expectations and demonstrate the efficiency of the method.