Active Covariance Matrix Adaptation for the (1+1)-CMA-ES

Dirk V. Arnold 1, * Nikolaus Hansen 2
* Corresponding author
2 TAO - Machine Learning and Optimisation
LRI - Laboratoire de Recherche en Informatique, UP11 - Université Paris-Sud - Paris 11, Inria Saclay - Ile de France, CNRS - Centre National de la Recherche Scientifique : UMR8623
Abstract : We propose a novel variant of the (1+1)-CMA-ES that updates the distribution of mutation vectors based on both successful and unsuccessful trial steps. The computational costs of the adaptation procedure are quadratic in the dimensionality of the problem, and the algorithm retains all invariance properties. Its performance on a set of standard test functions is compared with that of the original strategy that updates the distribution of mutation vectors in response to successful steps only. The new variant is not observed to be more than marginally slower on any function, and it is up to two times faster on some of the test problems.
Document type :
Conference papers
Genetic And Evolutionary Computation Conference, Jul 2010, Portland, United States. pp.385-392, 2010, <10.1145/1830483.1830556>
Liste complète des métadonnées


https://hal.archives-ouvertes.fr/hal-00503250
Contributor : Nikolaus Hansen <>
Submitted on : Sunday, July 18, 2010 - 3:13:34 PM
Last modification on : Thursday, February 9, 2017 - 3:59:16 PM
Document(s) archivé(s) le : Thursday, June 30, 2011 - 1:18:40 PM

File

arnold2010active.pdf
Files produced by the author(s)

Identifiers

Citation

Dirk V. Arnold, Nikolaus Hansen. Active Covariance Matrix Adaptation for the (1+1)-CMA-ES. Genetic And Evolutionary Computation Conference, Jul 2010, Portland, United States. pp.385-392, 2010, <10.1145/1830483.1830556>. <hal-00503250>

Share

Metrics

Record views

606

Document downloads

381