Fast Computation of the Multi-points Expected Improvement with Applications in Batch Selection

Clément Chevalier (1, 2, 3) , David Ginsbourger (1)

1 IMSV - Institute of Mathematical Statistics and Actuarial Science [Bern]
2 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques
3 IRSN - Institut de Radioprotection et de Sûreté Nucléaire

The Multipoint Expected Improvement criterion (or q-EI) has recently been studied in batch-sequential Bayesian Optimization. This paper deals with the new way of computing q-EI, without usingMonte-Carlo simulations, through a new closed form formula. The latter allows a very fast computation of q-EI for reasonably low values of q (typically, less than 10). New parallel kriging-based optimization strategies, tested on various toy examples, show promising results.

Mots clés: Batch-sequential optimization; Kriging; EGO; Tallis formula

Domaines: Modélisation et simulation; Optimisation et contrôle [math.OC]; Machine Learning [stat.ML]; Méthodologie [stat.ME]

Clément Chevalier, David Ginsbourger. Fast Computation of the Multi-points Expected Improvement with Applications in Batch Selection. 2012. ⟨hal-00732512v2⟩