An improved high-dimensional Bayesian optimization algorithm

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 55; no. 13; p. 915
• Main Authors: Guan, Juan, Wang, Yanhua
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Bayesian analysis; Clustering; Computer Science; Datasets; Decomposition; Design optimization; Global optimization; Machine learning; Machines; Manufacturing; Mechanical Engineering; Methods; Optimization algorithms; Principal components analysis; Processes; KPCA; High-dimensional Bayesian optimization; MIMIC algorithm; Variable interaction
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-025-06750-5

The Bayesian Optimization Algorithm, as an effective approach to addressing non-linear global optimization problems, is widely embraced in a myriad of machine learning application domains. With the development of big data, the presence of computational and statistical challenges in high-dimensional settings means that, despite the proposed improvements and enhancements, the applicability of the Bayesian Optimization Algorithm is still restricted to low-dimensional problems. Our algorithm (1) extracts an interesting nonlinear latent structure in the function by Kernal Principal Component Analysis(KPCA) to reduce the computational complexity, and (2) uses an improved Mutual-Information-Maximizing Input Clustering (MIMIC) algorithm to optimize only a low-dimensional subspace each iteration for more efficient and effective BO. The experiments demonstrate that the proposed algorithm can achieve a clear improvement in optimization accuracy and speed in high-dimensional space and can efficiently solve high-dimensional problems for Bayesian optimization algorithm.