Macroevolutionary Algorithms: A New Optimization Method on Fitness Landscapes

In this paper we introduce a new approach to optimization problems based on a previous theoretical work on extinction patterns in macroevolution. We name them Macroevolutionary Algorithms (MA). Unlike population-level evolution, which is employed in standard genetic algorithms, evolution at the level of higher taxa is used as the underlying metaphor. The model exploits the presence of links between "species" which represent candidate solutions to the optimization problem. In order to test its effectiveness, we compare the performance of MAs versus genetic algorithms (GA) with tournament selection. The method is shown to be a good alternative to standard GAs, showing a fast monotonous search over the solution space even for very small population sizes. A mean field theoretical approach is presented, showing that the basic dynamics of MAs is close to an ecological model of multispecies competition. <p> <p> Submitted to IEEE Transactions on Evolutionary Computation.