Flocking of subpopulations in distributed genetic programming

The distribution of the genetic programming algorithm improves the efficiency of the search for the solution, but additional parameters of this distribution are undesirable. This paper presents the analysis of early experimental results of using flocking to control interactions among the distributed subpopulations so less human intervention is needed The possibility to set up migration parameters dynamically at the run time brings the distributed genetic programming algorithm to the same level of automation as standard genetic programming while keeping the increased performance of the distributed GP. The paper discusses the nature of the required additional computations of the GP algorithm when adapting flocking for migration control. The positive empirical results support the idea of mixing both search techniques together.