On the parallel speed-up of Estimation of Multivariate Normal Algorithm and Evolution Strategies

Fabien Teytaud (1, 2) , Olivier Teytaud (3, 4, 5)

1 INRIA Futurs
2 UP11 UFR Sciences - Université Paris-Sud - Paris 11 - Faculté des Sciences
3 TANC - Algorithmic number theory for cryptology
4 LRI - Laboratoire de Recherche en Informatique
5 TAO - Machine Learning and Optimisation

Motivated by parallel optimization, we experiment EDA-like adaptation-rules in the case of $\lambda$ large. The rule we use, essentially based on estimation of multivariate normal algorithm, is (i) compliant with all families of distributions for which a density estimation algorithm exists (ii) simple (iii) parameter-free (iv) better than current rules in this framework of $\lambda$ large. The speed-up as a function of $\lambda$ is consistent with theoretical bounds.

Domaines: Optimisation et contrôle [math.OC]

Fabien Teytaud, Olivier Teytaud. On the parallel speed-up of Estimation of Multivariate Normal Algorithm and Evolution Strategies. EvoNum (evostar workshop), 2009, Tuebingen, Germany. ⟨inria-00369781⟩