A Heuristic Strategy of a Bayesian Optimization Algorithm for Evaluation-times Constrained Optimization (ECO) Problems

This paper defines a new class of black-box optimization problems in which the objective evaluation-times are constrained in advance. An algorithm for this optimization class must find a better solution in the constrained evaluation times. This concept is significant if the evaluation every time is expensive on time and/or money. We call this class "Evaluation times Constrained Optimization (ECO)". To solve ECO problems efficiently we focus on a Bayesian optimization algorithm and propose a simple heuristic strategy to make the algorithm suit the problems. The effectiveness of the proposed strategy was investigated through numerical experiments using some objective functions with different properties.