TREGO: a trust-region framework for efficient global optimization

Efficient global optimization (EGO) is the canonical form of Bayesian that has been successfully applied to solve expensive-to-evaluate black-box problems. However, EGO struggles scale with dimension, and offers limited theoretical guarantees. In this work, a trust-region framework for (TREGO) proposed analyzed. TREGO alternates between regular steps local within trust region. By following classical scheme region (based on sufficient decrease condition), algorithm enjoys convergence properties, while departing from only subset steps. Using extensive numerical experiments based well-known COCO bound constrained problems, we first analyze sensitivity its own parameters, then show resulting consistently outperforming getting competitive other state-of-the-art methods.