Bayesian CMA-ES: a new approach - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

Bayesian CMA-ES: a new approach

Résumé

This paper introduces a novel theoretically sound approach for the celebrated CMA-ES algorithm. Assuming the parameters of the multi variate normal distribution for the minimum follow a conjugate prior distribution, we can derive the optimal update at each iteration step thanks to Bayesian statistics. Update formulae are very similar to the vanilla (μ/λ) CMA-ES. We also use variance contraction and dilatation to accommodate for local and global search. As a result, this Bayesian framework provides a justification for the update of the CMA-ES algorithm and gives a new version of CMA-ES assuming normal-Inverse Wishart prior that has similar convergence speed and efficiency as the vanilla (μ/λ) CMA-ES on test functions ranging from cone, Rastrigin to Schwefel 1 and 2 functions.
Fichier principal
Vignette du fichier
main_hal.pdf (822.11 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03814858 , version 1 (18-04-2023)

Identifiants

Citer

Eric Benhamou, David Saltiel, Sébastien Verel, Fabien Teytaud. Bayesian CMA-ES: a new approach. GECCO '20: Genetic and Evolutionary Computation Conference, Jul 2020, Cancún, Mexico. pp.203-204, ⟨10.1145/3377929.3389913⟩. ⟨hal-03814858⟩
49 Consultations
17 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More