Understanding High-Dimensional Bayesian Optimization
Papenmeier, Leonard; Poloczek, Matthias; Nardi, Luigi
Abstract
Recent work reported that simple Bayesian optimization (BO) methods perform well for high-dimensional real-world tasks, seemingly contradicting prior work and tribal knowledge. This paper investigates why. We identify underlying challenges that arise in high-dimensional BO and explain why recent methods succeed. Our empirical analysis shows that vanishing gradients caused by Gaussian process (GP) initialization schemes play a major role in the failures of high-dimensional Bayesian optimization (HDBO) and that methods that promote local search behaviors are better suited for the task. We find that maximum likelihood estimation (MLE) of GP length scales suffices for state-of-the-art performance. Based on this, we propose a simple variant of MLE called MSR that leverages these findings to achieve state-of-the-art performance on a comprehensive set of real-world applications. We present targeted experiments to illustrate and confirm our findings.
Keywords
Bayesian optimization; global optimization; Gaussian process; high-dimensional
Cite as
Papenmeier, L., Poloczek, M., & Nardi, L. (2025). Understanding High-Dimensional Bayesian Optimization. In Singh, A., Fazel, M., Hsu, D., Lacoste-Julien, S., Berkenkamp, F., Maharaj, T., Wagstaff, K., & Zhu, J. (Eds.),
Proceedings of Machine Learning Research (PMLR) (-, pp. 47902–47923). Proceedings of Machine Learning Research (PMLR): Vol. 267. Vancouver, Canada: MLResearchPress.
Details
Publication type
Research article in proceedings (conference)
Peer reviewed
Yes
Publication status
Published
Year
2025
Conference
Proceedings of the Forty-Second International Conference on Machine Learning
Venue
Vancouver Convention Center
Volume
267
Book title
Proceedings of Machine Learning Research (PMLR)
Editor
Singh, Aarti; Fazel, Maryam; Hsu, Daniel; Lacoste-Julien, Simon; Berkenkamp, Felix; Maharaj, Tegan; Wagstaff, Kiri; Zhu, Jerry
Start page
47902
End page
47923
Edition
-
Volume
267
Title of series
Proceedings of Machine Learning Research (PMLR)
Publisher
MLResearchPress
Place
Vancouver, Canada
Language
English
ISSN
2640-3498
ISBN
-
DOI
Full text
