Scalable learning in genetic programming using automatic function definition

Created by W.Langdon from gp-bibliography.bib Revision:1.3973

@InCollection{Kinnear:Koza:1994:adf,
  author =       "John R. Koza",
  title =        "Scalable learning in genetic programming using
                 automatic function definition",
  booktitle =    "Advances in Genetic Programming",
  publisher =    "MIT Press",
  year =         "1994",
  editor =       "Kenneth E. {Kinnear, Jr.}",
  pages =        "99--117",
  chapter =      "5",
  address =      "Cambridge, MA, USA",
  keywords =     "genetic algorithms, genetic programming",
  URL =          "http://www.genetic-programming.com/jkpdf/aigp1994lawn.pdf",
  URL =          "http://cognet.mit.edu/sites/default/files/books/9780262277181/pdfs/9780262277181_chap5.pdf",
  abstract =     "This chapter uses three differently sized versions of
                 an illustrative problem that has considerable
                 regularity, symmetry, and homogeneity in its problem
                 environment to compare genetic programming with and
                 without the newly developed mechanism of automatic
                 function definition. Genetic programming with automatic
                 function definition can automatically decompose a
                 problem into simpler subproblems, solve the
                 subproblems, and assemble the solutions to the
                 subproblems into a solution to the original overall
                 problem. The solutions to the problem produced by
                 genetic programming with automatic function definition
                 are more parsimonious than those produced without it.
                 Genetic programming requires fewer fitness evaluations
                 to yield a solution to the problem with 99percent
                 probability with automatic function definition than
                 without it. When we consider the three differently
                 sized versions of the problem we find that the size of
                 the solutions produced without automatic function
                 definition can be expressed as a direct multiple of
                 problem size. In contrast, the average size of
                 solutions with automatic function definition is
                 expressed as a certain minimum size representing the
                 overhead associated with automatic function definition;
                 however, there is only a very slight increase in the
                 average size of the solutions with problem size.
                 Moreover, the number of fitness evaluations required to
                 yield a solution to the problem with a 99percent
                 probability grows very rapidly with problem size
                 without automatic function definition, but this same
                 measure grows only linearly with problem size with
                 automatic function definition.",
  size =         "20 pages",
}

Genetic Programming entries for John Koza

Citations