Optimizing deceptive functions with the SG-Clans algorithm

Starting from a different view of natural evolution, namely that of English biologist R. Dawkins, called the selfish gene theory, a new evolutionary computation approach can be developed, the selfish gene (SG) algorithm. This paper presents a significant improvement to the SG algorithm that is able to find and exploit linkages among different genes thanks to the evolution of isolated groups called clans. The resulting SG-Clans algorithm is shown to be able to find the absolute maximum of Holland Royal Road functions, which were specifically designed to create insurmountable difficulties for a wide class of hill-climbing approaches. We support experimental evidence that SG-Clans shares the speed of a hill-climber with the ability of broadly exploring the search space.