Rooted-Tree Schemata in Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.4420

@InCollection{rosca:1999:aigp3,
  author =       "Justinian P. Rosca and Dana H. Ballard",
  title =        "Rooted-Tree Schemata in Genetic Programming",
  booktitle =    "Advances in Genetic Programming 3",
  publisher =    "MIT Press",
  year =         "1999",
  editor =       "Lee Spector and William B. Langdon and 
                 Una-May O'Reilly and Peter J. Angeline",
  chapter =      "11",
  pages =        "243--271",
  address =      "Cambridge, MA, USA",
  month =        jun,
  keywords =     "genetic algorithms, genetic programming",
  ISBN =         "0-262-19423-6",
  URL =          "http://www.cs.bham.ac.uk/~wbl/aigp3/ch11.pdf",
  abstract =     "we present a novel way of addressing the issue of
                 variable complexity of evolved solutions and a revised
                 interpretation of how Genetic Programming (GP)
                 constructs solutions, based on the rooted-tree schema
                 concept.

                 A rooted-tree schema is a simple relation on the space
                 of tree-shaped structures which provides a quantifiable
                 partitioning of the search space. Formal manipulation
                 of rooted-tree schemata allows: (1) The role of the
                 size in the selection and survival of evolved
                 expressions to be made explicit; (2) The
                 interrelationship between parsimony penalty, size, and
                 fitness of evolved expressions to be clarified and
                 better understood; (3) The introduction of alternative
                 approaches to evolving parsimonious solutions by
                 preventing rooted-tree schema from bloating.

                 The rooted-tree schema concept provides a top-down
                 perspective of how program expressions are evolved,
                 contrary to the common belief that small pieces of
                 code, or building blocks, are gradually assembled to
                 create solutions. Analysis shows that GP, while it
                 improves solutions, combines both bottom-up and
                 top-down refinement strategies.",
  notes =        "AiGP3 See http://cognet.mit.edu",
}

Genetic Programming entries for Justinian Rosca Dana H Ballard

Citations