Embracing premature convergence: the Hypergamous Parallel Genetic Algorithm

Traditional genetic algorithm (GA) approaches have emphasized recombination as a dominant heuristic but have typically included mutation as a critical contributor to effective global search dynamics. Mutation has been widely regarded as an important “insurance policy” to avoid the well-known problem of premature convergence in the GA population. This paper introduces an alternative genetic algorithm that uses only recombination as its global search operator. Rather than attempting to avoid problematic premature convergence, the Hypergamous Parallel Genetic Algorithm (HPGA) embraces premature convergence as exponential-like convergence to some (probabilistic) local optimum. Local optima discovered in multiple subpopulations are transferred to “melting pots” in which new rounds of recombination are used to search for more global optima. In lieu of mutation, multiple randomly initialized subpopulations serve as the ultimate source of diversity within the HPGA population