Experimental investigation of three distributed genetic programming models

Abstract

Three models of distributed Genetic Programming are presented comprising synchronous and asynchronous communication. These three models are compared with each other and with the standard panmictic model on three well known Genetic Programming benchmarks. The measures used are the computational effort, the phenotypic entropy of the populations, and the execution time. We find that all the distributed models are better than the sequential one in terms of effort and time. The differences among the distributed models themselves are rather small in terms of effort but one of the asynchronous models turns out to be significantly faster. The entropy confirms that migration helps in conserving some phenotypic diversity in the populations.