A conditional gaussian martingale algorithm for global optimization

Authors:
Manuel L. Esquível
Affiliations:
Departamento de Matemática and CMA, FCT/UNL, Quinta da Torre, Caparica, Portugal
Venue:
ICCSA'06 Proceedings of the 2006 international conference on Computational Science and Its Applications - Volume Part III
Year:
2006

Citing 5
Cited 0

Global optimization requires global information

Journal of Optimization Theory and Applications
Testing Unconstrained Optimization Software

ACM Transactions on Mathematical Software (TOMS)
Rates of Convergence for a Class of Global Stochastic Optimization Algorithms

SIAM Journal on Optimization
Introduction to Stochastic Search and Optimization

Introduction to Stochastic Search and Optimization
On Accelerated Random Search

SIAM Journal on Optimization

Quantified Score

Hi-index	0.00

Visualization

Abstract

A new stochastic algorithm for determination of a global minimum of a real valued continuous function defined on K, a compact set of ℝn, having an unique global minimizer in K is introduced and studied, a context discussion is presented and implementations are used to compare the performance of the algorithm with other algorithms. The algorithm may be thought to belong to the random search class but although we use Gaussian distributions, the mean is changed at each step to be the intermediate minimum found at the preceding step and the standard deviations, on the diagonal of the covariance matrix, are halved from one step to the next. The convergence proof is simple relying on the fact that the sequence of intermediate random minima is an uniformly integrable conditional Gaussian martingale.