A brief comparison of some evolutionary optimization methods
Andy J. Keane · ePrints Soton (University of Southampton) · 1996
The subject of evolutionary computing is a rapidly developing one where many new search methods are being proposed all the time. Inevitably, some of these new methods will not still be in current use in a few years as a small subset becomes the preferred choice of the optimization community. A key part in this process is the back-to-back testing of competing methods on test problems that simulate real problems. This brief paper presents results for such a test, focusing on the robustness of four methods when applied to a pair of high dimensional test functions. The work presented shows that some modern search methods can be highly sensitive to mistuning of their control parameters and, moreover, that different problems may well require radically different settings of such parameters if the searches are to be effective.