TREGO: a trust-region framework for efficient global optimization

• Published in: Journal of global optimization Vol. 86; no. 1; pp. 1 - 23
• Main Authors: Diouane, Youssef, Picheny, Victor, Riche, Rodolophe Le, Perrotolo, Alexandre Scotto Di
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2023; Springer; Springer Nature B.V; Springer Verlag
• Subjects: Algorithms; Analysis; Black boxes; Canonical forms; Computational Engineering, Finance, and Science; Computer Science; Design of experiments; Engineering Sciences; Exploitation; Global optimization; Mathematics; Mathematics and Statistics; Mechanical engineering; Mechanics; Operations Research/Decision Theory; Optimization; Optimization and Control; Parameter sensitivity; Real Functions; Trust-region; Bayesian optimization; Non-linear optimization; Black-box optimization; Gaussian processes; trust-region; nonlinear optimization
• ISSN: 0925-5001; 1573-2916; 1573-2916
• DOI: 10.1007/s10898-022-01245-w

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