Probabilistic Adaptive Mapping Developmental Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.8010

@PhdThesis{G_C_Wilson:thesis,

• author = "Garnett Carl Wilson",
• title = "Probabilistic Adaptive Mapping Developmental Genetic Programming",
• school = "Dalhousie University",
• year = "2007",
• address = "Halifax, Nova Scotia, Canada",
• month = mar,
• keywords = "genetic algorithms, genetic programming, developmental genetic programming, genetic code, cooperative coevolution, genotype-phenotype mapping, redundant representation, neutrality, recursion",
• URL = "https://web.cs.dal.ca/~mheywood/Thesis/PhD.html",
• URL = "http://hdl.handle.net/10222/54875",
• URL = "https://dalspace.library.dal.ca/bitstream/handle/10222/54875/NR27171.PDF",
• size = "247 pages",
• abstract = "Developmental Genetic Programming (DGP) algorithms explicitly enable the search space for a problem to be divided into genotypes and corresponding phenotypes. The two search spaces are often connected with a genotype-phenotype mapping (GPM) intended to model the biological genetic code, where current implementations of this concept involve evolution of the mappings along with evolution of the genotype solutions. This work presents the Probabilistic Adaptive Mapping DGP (PAM DGP) algorithm, a new developmental implementation that provides research contributions in the areas of GPMs and coevolution. The algorithm component of PAM DGP is demonstrated to overcome coevolutionary performance problems as identified and empirically benchmarked against the latest competing Adaptive Mapping algorithm with both algorithms using the same (non-redundant) mapping encoding process. Having established that PAM DGP provides a superior algorithmic framework given equivalent mapping and genotype structures for the individuals, a new adaptive redundant mapping is incorporated into PAM DGP for further performance enhancement and closer adherence to developmental modeling of the biological code. PAM DGP with two mapping types is then compared to the competing Adaptive Mapping algorithm and Traditional GP with respect to three regression benchmarks. PAM DGP using redundant mappings is then applied to two medical classification domains, where PAM DGP with redundant encodings is found to provide better classifier performance than the alternative algorithms. PAM DGP with redundant mappings is also given the task of learning three sequences of increasing recursion order given a function set consisting of general (not implicitly recursive) machine-language instructions; where it is found to more efficiently learn second and third order recursive Fibonacci functions than the related developmental systems and Traditional GP. PAM DGP using redundant encoding is also demonstrated to produce the semantically highest quality solutions for all three recursive functions considered (Factorial, second and third order Fibonacci). PAM DGP is shown for regression, medical classification, and recursive problems to have produced its solutions by evolving redundant mappings to emphasise appropriate members within relevant subsets of the problem's original function set.",
• notes = "Supervisor: Malcolm I. Heywood",

}

Genetic Programming entries for Garnett Carl Wilson

Citations