A Fine-Grained View of Phenotypes and Locality in Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.4549

  author =       "James McDermott and Edgar Galvan-Lopez and 
                 Michael O'Neill",
  title =        "A Fine-Grained View of Phenotypes and Locality in
                 Genetic Programming",
  booktitle =    "Genetic Programming Theory and Practice IX",
  year =         "2011",
  editor =       "Rick Riolo and Ekaterina Vladislavleva and 
                 Jason H. Moore",
  series =       "Genetic and Evolutionary Computation",
  address =      "Ann Arbor, USA",
  month =        "12-14 " # may,
  publisher =    "Springer",
  chapter =      "4",
  pages =        "57--76",
  keywords =     "genetic algorithms, genetic programming, Evolutionary
                 computation, fitness landscape, problem difficulty,
                 phenotype, locality, artificial ant, Boolean problems",
  isbn13 =       "978-1-4614-1769-9",
  DOI =          "doi:10.1007/978-1-4614-1770-5_4",
  abstract =     "The locality of the mapping from genotype to phenotype
                 is an important issue in the study of landscapes and
                 problem difficulty in evolutionary computation. In
                 tree-structured Genetic Programming (GP), the locality
                 approach is not generally applied because no explicit
                 genotype-phenotype mapping exists, in contrast to some
                 other GP encodings. we define GP phenotypes in terms of
                 semantics or behaviour. For a given problem, a model of
                 one or more phenotypes and mappings between them may be
                 appropriate e.g. g -> p_0, where g is the genotype, p_i
                 are distinct types of phenotypes and f is fitness.
                 Thus, the behaviour of each component mapping can be
                 studied separately. The locality of the
                 genotype-phenotype mapping can also be decomposed into
                 the effects of the encoding and those of the operator's
                 genotypic step-size. Two standard benchmark problem
                 classes, Boolean and artificial ant, are studied in a
                 principled way using this fine-grained view of
                 locality. The method of studying locality with
                 phenotypes seems useful in the case of the artificial
                 ant, but Boolean problems provide a counter-example.",
  notes =        "part of \cite{Riolo:2011:GPTP}",
  affiliation =  "Evolutionary Design and Optimization, CSAIL, MIT,
                 Cambridge, USA",

Genetic Programming entries for James McDermott Edgar Galvan Lopez Michael O'Neill