Using biology to solve a problem in automated machine learning

Created by W.Langdon from gp-bibliography.bib Revision:1.4221

  author =       "John R. Koza",
  title =        "Using biology to solve a problem in automated machine
  booktitle =    "Models of Action: Mechanisms for Adaptive Behavior",
  publisher =    "Lawrence Erlbaum Associates",
  year =         "1998",
  editor =       "Clive D. L. Wynne and John E. R. Staddon",
  chapter =      "5",
  pages =        "157--199",
  address =      "Hillsdale, NJ, USA",
  keywords =     "genetic algorithms, genetic programming",
  ISBN =         "0-8058-1597-X",
  URL =          "",
  abstract =     "This chapter describes how the biological theory of
                 gene duplication described in Susumu Ohno's provocative
                 book, Evolution by Means of Gene Duplication, was
                 brought to bear on a vexatious problem from the domain
                 of automated machine learning. The goal of automatic
                 programming is to create, in an automated way, a
                 computer program that enables a computer to solve a
                 problem. Ideally, an automatic programming system
                 should require that the user pre-specify little about
                 the problem environment. Genetic programming is a
                 domain-independent approach to automated machine
                 learning that attempts to evolve a computer program
                 that solves, or approximately solves, problems.
                 Starting with a primordial ooze of randomly generated
                 computer programs composed of the available
                 programmatic ingredients, genetic programming applies
                 the principles of animal husbandry (including Darwinian
                 selection and sexual recombination) to breed new (and
                 often improved) populations of computer programs. One
                 of the undesirable aspects of many techniques of
                 automated machine learning is that the user of the
                 technique may be required to specify the size and shape
                 (i.e., the architecture) of the ultimate solution to
                 his problem before he can begin to apply the technique
                 to his problem. Specification of the size and shape of
                 the solution often corresponds to discovering a way to
                 decompose the problem into useful subspaces (usually of
                 lower dimensionality) or to discovering a congenial
                 representation of the problem that facilitates solution
                 of the problem. Thus, in practice, for many problems of
                 interest, determining the size and shape of the
                 solution may be the problem (or at least a substantial
                 part of the problem). This chapter describes how
                 biology motivated a solution to the problem of
                 architecture discovery for genetic programming. The
                 resulting biologically-motivated approach enables
                 genetic programming to automatically discover the size
                 and shape of the solution at the same time as genetic
                 programming is evolving a solution to the problem. This
                 is accomplished using six new architecture-altering
                 operations that provide a way to automatically
                 discover, during a run of genetic programming, both the
                 architecture and the sequence of steps of a multi-part
                 computer program that will solve the given problem.",
  notes =        "",
  size =         "75 pages",

Genetic Programming entries for John Koza