Second order-response surface model for the automated parameter tuning problem

Several automated parameter tuning procedures/configurators have been proposed in order to find the best parameter setting for a target algorithm. These configurators can generally be classified into model-free and model-based approaches. We introduce a recent approach which is based on the hybridization of both approaches. It combines the Design of Experiments (DOE) and Response Surface Methodology (RSM) with prevailing model-free techniques. DOE is mainly used for determining the importance of parameters. A First Order-RSM is initially employed to define the promising region for the important parameters. A Second Order-RSM is then built to approximate the center point as well as the final promising ranges of parameter values. We show how our approach can be embedded with existing model-free techniques, namely ParamILS and Randomized Convex Search, to tune target algorithms and demonstrate that our proposed methodology leads to improvements in terms of the quality of the solutions compared against the earlier work.