Decomposition is a fundamental strategy for solving multiobjective optimization problems (MOPs), as it enables the simplification of MOPs into more manageable sub-problems [1]. This strategy not only makes the problem-solving process more efficient, but also provides a flexible way to manage the diversity of solutions. This page includes the representative MultiObjective Bayesian Optimization (MOBO) papers that utilize decomposition techniques.

Applie the standard single-objective acquisition function after random scalarization.

• 2006, TEVC. Joshua Knowles. "ParEGO: a Hybrid Algorithm with on-line Landscape Approximation for Expensive Multiobjective Optimization Problems". [PDF]
• 2020, UAI. Biswajit Paria, Kirthevasan Kandasamy, Barnabás Póczos. "A Flexible Framework for Multi-Objective Bayesian Optimization using Random Scalarizations". [PDF]
• 2020, ICML. Richard Zhang, Daniel Golovin. "Random Hypervolume Scalarizations for Provable Mmulti-objective Black Box Optimization". [PDF]

Following the MOEA/D framework, decompose a MOP in question into a number of sub-problems via a set of reference vectors and solve these sub-problems in a collaborative manner. It is based on the assumption that the optimal solutions of two subproblems should be similar when their reference vectors are close.

• 2010, TEVC. MOEA/D-EGO. Qingfu Zhang, Wudong Liu, Edward Tsang, and Botond Virginas. "Expensive Multiobjective Optimization by MOEA/D with Gaussian Process Model". [PDF] [code]
• 2022, NeurIPS. PSL-MOBO. Xi Lin, Zhiyuan Yang, Xiaoyuan Zhang, and Qingfu Zhang. "Pareto Set Learning for Expensive Multi-Objective Optimization". [PDF] [Supplementary] [code]
• 2024, TEVC. DirHV-EGO. Liang Zhao and Qingfu Zhang. "Hypervolume-Guided Decomposition for Parallel Expensive Multiobjective Optimization". [PDF] [Supplementary] [code]