Multi-Criteria Optimization for Hard Problems under Limited Budgets

Many relevant industrial optimization tasks feature more than just one quality criterion. State-of-the-art multi-criteria optimization algorithms require a relatively large number of function evaluations (usually more than 10^5) to approximate Pareto fronts. Due to high cost or time consumption this large amount of function evaluations is not always available. Therefore, it is obvious to combine techniques such as Sequential Parameter Optimization (SPO), which need a very small number of function evaluations only, with techniques from evolutionary multi-criteria optimization (EMO). In this paper, we show how EMO techniques can be efficiently integrated into the framework of the SPO Toolbox (SPOT). We discuss advantages of this approach in comparison to state-of-the-art optimizers. Moreover, with the resulting capability to allow competing objectives, the opportunity arises to not only aim for the best, but also for the most robust solution. Herein we present an approach to optimize not only the quality of the solution, but also its robustness, taking these two goals as objectives for multi-criteria optimization into account.