Track: Doctoral Dissertation Competition

Abstract

Metaheuristic and heuristic methods have many design parameters, and a good selection of that parameter values can increase their performance. In this study, we proposed a sampling based parameter tuning approach and applied it to the Differential Ant-Stigmergy Algorithm (DASA)’s five control parameters. In this study, performance of large set of DASA parameter settings, obtained using Latin Hypercube Hammersley Sampling (LHHS) method, was evaluated on the Sphere function optimization problem with respect to two dimensions of this function. According to the results of our research, the best parameter vector obtained with the LHHS method found better results than the default parameter value of the DASA and also than other proposed tuned version DASA*. And the results demonstrated that three parameter configurations obtained with LHHS found better result than the best configuration obtained with Sobol Sequence Sampling method for function dimension 20, and five parameter configuration for function dimension 40. According to the results, it can be said that usage of LHHS for initialization of other state-of-art algorithm configuration methods instead of other sampling methods is worth investigating. Although some studies evaluated sampling methods for various configuration methods (such as SPO and irace), this is the first study that uses LHSS method as an individual algorithm configuration method.