Inproceedings,
Experimental Investigation Of Three Distributed Genetic Programming Models
Parallel Problem Solving from Nature - PPSN VII, 2439, page 641--650. Granada, Spain, Springer-Verlag, (7-11 September 2002)
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.
