PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

{Dimension-independent convergence rate for non-isotropic (1,\lambda)-ES
Anne Auger, Claude LeBris and Marc Schoenauer
In: GECCO 2003, July 2003, Chicago.


Based on the theory of non-negative supermartingales, convergence results are proven for adaptive $(1,\lambda)-ES$ (i.e. with Gaussian mutations), and geometrical convergence rates are derived. In the $d$-dimensional case ($d > 1$), the algorithm studied here uses a different step-size update in each direction. However, the critical value for the step-size, and the resulting convergence rate do not depend on the dimension. Those results are discussed with respect to previous work. Finally, rigourous numerical investigations on some 1-dimensional functions validate the theoretical results.

Postscript - Requires a viewer, such as GhostView
EPrint Type:Conference or Workshop Item (Talk)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Theory & Algorithms
ID Code:613
Deposited By:Marc Schoenauer
Deposited On:29 December 2004