Trivial Geography in Genetic Programming

Geographical distribution is widely held to be a major determinant of evolutionary dynamics. Correspondingly, genetic programming theorists and practitioners have long developed, used, and studied systems in which populations are structured in quasi-geographicalways. Herewe show that a remarkably simple version of this idea produces surprisingly dramatic improvements in problem-solving performance on a suite of test problems. The scheme is trivial to implement, in some cases involving little more than the addition of a modulus operation in the population access function, and yet it provides significant benefits on all of our test problems (ten symbolic regression problems and a quantum computing problem). We recommend the broader adoption of this form of “trivial geography” in genetic programming systems.