Distributed Genetic Programming- Empirical Study and Analysis

The MIT Press eBooks · 1996

This paper introduces a distributed parallel implementation of GP (Genetic Programming), DGP. We implemented our DGP on the parallel system AP-1000+ of Fujitsu, which consisting of 32 processors. Several experimental results are shown comparing three architectures of DGP, i.e. the ring type, the one-way torus, and the two-way torus. We provide evidence for the effectiveness of our DGP by comparative experiments with the partitioned GP, i.e. a DGP without migration, and the canonical GP. In order to analyze the performance, we discuss the change of the structural diversity with generations for solving the deceptive problem.

Read the paper · More papers on PaperTik