Bias and variance in continuous EDA

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

1 TANC - Algorithmic number theory for cryptology
2 LRI - Laboratoire de Recherche en Informatique
3 TAO - Machine Learning and Optimisation

Estimation of Distribution Algorithms are based on statistical estimates. We show that when combining classical tools from statistics, namely bias/variance decomposition, reweighting and quasi-randomization, we can strongly improve the convergence rate. All modifications are easy, compliant with most algorithms, and experimentally very efficient in particular in the parallel case (large offsprings).

Domaines: Optimisation et contrôle [math.OC]

Fabien Teytaud, Olivier Teytaud. Bias and variance in continuous EDA. EA 09, Oct 2009, Strasbourg, France. ⟨inria-00451416⟩