Adaptive Migration for the Distributed Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.8051

@PhdThesis{Paulikas:thesis,

• author = "Giedrius Paulikas",
• title = "Adaptive Migration for the Distributed Genetic Programming",
• school = "Technological science, Informatics Engineering, Kaunas University of Technology",
• year = "2007",
• address = "Kaunas, Lithuania",
• keywords = "genetic algorithms, genetic programming",
• URL = "http://en.ktu.lt/sites/default/files/2007-09-07%20giedrius%20paulikas.pdf",
• URL = "http://www.libis.lt/showRecordDetails.do?recordNum=1&biId=216090158&catalog=false&resId=&previewUrl=undefined&epaveldas=f",
• abstract = "The work studies a genetic programming algorithm that is used for automatic generation of computer programs. The distributed version of this algorithm processes several large parts of the program population (subpopulations) separately, occasionally moving a small quantity of individuals among the subpopulations. The distributed genetic programming algorithm exhibits a higher search speed when compared to traditional sequential counterpart, but it has additional distribution parameters. The selection of these parameters can be performed automatically during algorithm run time, this also improves the effectiveness of the search. Migration among the subpopulations is controlled using the flocking algorithm. Flocking algorithm is used to move independent agents, where each agent acts according to a set of simple rules that help to keep a constant shape of the flock. The measure of individual location in the search space is required in order to be able to apply flocking rules to the flock of genetic programming subpopulations. Two different origins of location measures are examined, namely the genotype based and phenotype based locations. The latter, phenotype, or fitness, based measure is preferred. Topology of migration is formed by moving programs to the most distant neighbors of the subpopulation. The observation of the progress of algorithm run shows that adapted flocking rules keep the bigger distance among the flock mates and cover the larger portion of search space. When genetic programming with adaptive migration control is experimentally compared to traditional sequential and distributed versions of the algorithm, generally an increase of effectiveness of the search is recorded. The adaptive control also eliminates the necessity of analysis of the problem domain that is required to choose the optimal distribution parameters of the algorithm.",
• notes = "http://en.ktu.lt/sites/default/files/2007-09-07%20giedrius%20paulikas.pdf gives English language summary",

}

Genetic Programming entries for Giedrius Paulikas

Citations