Discovery of a main program and reusable subroutines
using genetic programming
J. Koza. Proceedings of the Fifth Workshop on Neural Networks:
An International Conference on Computational
Intelligence: Neural Networks, Fuzzy Systems,
Evolutionary Programming, and Virtual Reality, page 109--118. (1993)
Abstract
This paper describes an approach for automatically
decomposing a problem into subproblems, automatically
creating reusable subroutines to solve the subproblems,
and automatically assembling the results produced by
the subroutines in order to solve the problem. The
approach uses genetic programming with the recently
developed additional facility of automatic function
definition. Genetic programming provides a way to
genetically breed a computer program to solve a problem
and automatic function definition enables genetic
programming to create reusable subroutines dynamically
during a run. The approach is applied to an
illustrative problem containing a considerable amount
of regularity. Solutions to the problem produced using
automatic function definition are considerably smaller
in size and require processing of considerably fewer
individuals than is the case without automatic function
definition. Specifically, the average program size for
a solution to the problem without using automatic
function definition is 3.65 times larger than the size
for a solution when using automatic function
definition. The number of individuals required to be
processed to yield a solution with 99% probability
without automatic function definition is 9.09 times
larger than the equivalent number required with
automatic function definition.
Proceedings of the Fifth Workshop on Neural Networks:
An International Conference on Computational
Intelligence: Neural Networks, Fuzzy Systems,
Evolutionary Programming, and Virtual Reality
%0 Conference Paper
%1 Koza:1993:mprsGP
%A Koza, John R.
%B Proceedings of the Fifth Workshop on Neural Networks:
An International Conference on Computational
Intelligence: Neural Networks, Fuzzy Systems,
Evolutionary Programming, and Virtual Reality
%D 1993
%K ADF algorithms, genetic programming,
%P 109--118
%T Discovery of a main program and reusable subroutines
using genetic programming
%U http://www.genetic-programming.com/jkpdf/simtec1993.pdf
%X This paper describes an approach for automatically
decomposing a problem into subproblems, automatically
creating reusable subroutines to solve the subproblems,
and automatically assembling the results produced by
the subroutines in order to solve the problem. The
approach uses genetic programming with the recently
developed additional facility of automatic function
definition. Genetic programming provides a way to
genetically breed a computer program to solve a problem
and automatic function definition enables genetic
programming to create reusable subroutines dynamically
during a run. The approach is applied to an
illustrative problem containing a considerable amount
of regularity. Solutions to the problem produced using
automatic function definition are considerably smaller
in size and require processing of considerably fewer
individuals than is the case without automatic function
definition. Specifically, the average program size for
a solution to the problem without using automatic
function definition is 3.65 times larger than the size
for a solution when using automatic function
definition. The number of individuals required to be
processed to yield a solution with 99% probability
without automatic function definition is 9.09 times
larger than the equivalent number required with
automatic function definition.
@inproceedings{Koza:1993:mprsGP,
abstract = {This paper describes an approach for automatically
decomposing a problem into subproblems, automatically
creating reusable subroutines to solve the subproblems,
and automatically assembling the results produced by
the subroutines in order to solve the problem. The
approach uses genetic programming with the recently
developed additional facility of automatic function
definition. Genetic programming provides a way to
genetically breed a computer program to solve a problem
and automatic function definition enables genetic
programming to create reusable subroutines dynamically
during a run. The approach is applied to an
illustrative problem containing a considerable amount
of regularity. Solutions to the problem produced using
automatic function definition are considerably smaller
in size and require processing of considerably fewer
individuals than is the case without automatic function
definition. Specifically, the average program size for
a solution to the problem without using automatic
function definition is 3.65 times larger than the size
for a solution when using automatic function
definition. The number of individuals required to be
processed to yield a solution with 99% probability
without automatic function definition is 9.09 times
larger than the equivalent number required with
automatic function definition.},
added-at = {2008-06-19T17:35:00.000+0200},
author = {Koza, John R.},
biburl = {https://www.bibsonomy.org/bibtex/27d8a6c7b7901c392e0548836fc2db582/brazovayeye},
booktitle = {Proceedings of the Fifth Workshop on Neural Networks:
An International Conference on Computational
Intelligence: Neural Networks, Fuzzy Systems,
Evolutionary Programming, and Virtual Reality},
interhash = {039839a9d32be4f20002213e324d1714},
intrahash = {7d8a6c7b7901c392e0548836fc2db582},
keywords = {ADF algorithms, genetic programming,},
organisation = {The Society for Computer Simulation},
pages = {109--118},
publisher_address = {San Diego, CA, USA},
size = {11 pages},
timestamp = {2008-06-19T17:43:51.000+0200},
title = {Discovery of a main program and reusable subroutines
using genetic programming},
url = {http://www.genetic-programming.com/jkpdf/simtec1993.pdf},
year = 1993
}