ABSTRACT

This paper recalls the stepwise uncertainty reduction algorithm (SUR) proposed in Vazquez & Bect (2009), which is in essence a Bayesian sequential search algorithm similar to the algorithm EGO in Jones et al. (1998) for finding a global optimum of a function. More precisely, the SUR strategy is based on the sequential maximization of the one-step lookahead expected mean square error J fSUR n n) ) | }

ˆ F 1 where αˆn is an estimator of α( f ), and Fn−1 represents the information that has been learned about f after n−1 evaluations. Implementation details are provided in the full paper.