Parameter Tuning of a Parallel Hierarchical Island Based Model

Abstract

One of the major drawbacks of using Evolutionary Algo- rithms is the determination of the input parameters, since they have an important in uence in the e ectiveness of the search. When using Par- allel Evolutionary Algorithms, such drawbacks are magni ed, since they incorporates new parameters needed to con gure the inherent charac- teristics of the parallel model. Achieving an adequate con guration can mean an optimization problem itself. This research group has developed a parallel distributed model characterized by a hierarchy of processes com- munication organized in islands that cooperate in the search process. In this work we present a study of calibration for some input parameters that determines good results quality, with the aim of tuning them taking into account di erent con gurations applied globally to the model, or locally to each island.