Optimally Weighted Ensembles for Efficient Multi-objective Optimization

The process of industrial design engineering is often involved with the simultaneous optimization of multiple expensive objectives. The surrogate assisted multi-objective S-Metric Selection - Efficient Global Optimization (SMS-EGO) algorithm is one of the most popular algorithms to solve these kind of problems. We propose an extension of the SMS-EGO algorithm with optimally weighted, linearly combined ensembles of regression models to improve its objective modelling capabilities. Multiple (different) surrogates are combined into one optimally weighted ensemble per objective using a model agnostic uncertainty quantification method to balance between exploration and exploitation. The performance of the proposed algorithm is evaluated on a diverse set of benchmark problems with a small initial sample and an additional budget of 25 evaluations of the real objective functions. The results show that the proposed Ensemble-based - S-Metric Selection - Efficient Global Optimization (E-SMS-EGO) algorithm outperforms the state-of-the-art algorithms in terms of efficiency, robustness and spread across the objective space.