Comparison of a spatially-structured cellular evolutionary algorithm to an evolutionary algorithm with panmictic population

Evolutionary Algorithms are metaheuristic optimization algorithms which are based on a population of individual candidate solutions. These solutions are evolved with the aim to solve a given problem. We compare two types of Evolutionary Algorithms, one with a panmictic population and one with a spatially-structured population. Previous works indicate that evolutionary algorithms with a spatially-structured population perform better that those with a panmictic population. In this work we will examine whether this holds true for evolving Artificial Neural Networks. For comparison we use two test problems, a simple XOR calculation and a complex task requiring self-organization among a number of agents. Our findings show that for the evaluated tasks, the algorithm with a spatially-structured population performs better than an algorithm with panmictic population.