A Schema Theory Analysis of Mutation Size Biases in
Genetic Programming with Linear Representations
N. McPhee, R. Poli, and J. Rowe. Proceedings of the 2001 Congress on Evolutionary
Computation CEC2001, page 1078--1085. COEX, World Trade Center, 159 Samseong-dong,
Gangnam-gu, Seoul, Korea, IEEE Press, (27-30 May 2001)
Abstract
Understanding operator bias in evolutionary
computation is important because it is possible for the
operator's biases to work against the intended biases
induced by the fitness function. In recent work we
showed how developments in GP schema theory can be used
to better understand the biases induced by the standard
subtree crossover when genetic programming is applied
to variable length linear structures. We use the schema
theory to better understand the biases induced on
linear structures by two common GP subtree mutation
operators: FULL and GROW mutation. In both cases we
find that the operators do have quite specific biases
and typically strongly oversample shorter strings.
COEX, World Trade Center, 159 Samseong-dong,
Gangnam-gu, Seoul, Korea
booktitle
Proceedings of the 2001 Congress on Evolutionary
Computation CEC2001
year
2001
month
27-30 May
pages
1078--1085
publisher
IEEE Press
organisation
IEEE Neural Network Council (NNC), Evolutionary
Programming Society (EPS), Institution of Electrical
Engineers (IEE)
publisher_address
445 Hoes Lane, P.O. Box 1331, Piscataway, NJ
08855-1331, USA
isbn
0-7803-6658-1
notes
CEC-2001 - A joint meeting of the IEEE, Evolutionary
Programming Society, Galesia, and the IEE.
IEEE Catalog Number = 01TH8546C,
Library of Congress Number = .
linear (unary) tree schemata. flat fitness landscape.
biases of full mutation, grow mutation,
No fitness. Full(unary) average length = 2*D-1.
Limiting size distribution: 0 for size < D, flat region
size < 2D, rapid falling size>=2D. Similar to subtree
crossover. Grow(unary) discrete gamma distribution (cf.
Rowe01 ) cf subtree crossover.
önes then zeros" unary problem. Subtree crossover
bloat (at least to 75 generations). full no bloat,
actually as with no fitness, ärtifact of this
particular problem". Grow similar to no fitness.
%0 Conference Paper
%1 mcphee:2001:astamsbgplr
%A McPhee, Nicholas Freitag
%A Poli, Riccardo
%A Rowe, Jonathan E.
%B Proceedings of the 2001 Congress on Evolutionary
Computation CEC2001
%C COEX, World Trade Center, 159 Samseong-dong,
Gangnam-gu, Seoul, Korea
%D 2001
%I IEEE Press
%K algorithms, bias genetic linear mutation, programming, representation, schema size theory,
%P 1078--1085
%T A Schema Theory Analysis of Mutation Size Biases in
Genetic Programming with Linear Representations
%U http://citeseer.ist.psu.edu/501380.html
%X Understanding operator bias in evolutionary
computation is important because it is possible for the
operator's biases to work against the intended biases
induced by the fitness function. In recent work we
showed how developments in GP schema theory can be used
to better understand the biases induced by the standard
subtree crossover when genetic programming is applied
to variable length linear structures. We use the schema
theory to better understand the biases induced on
linear structures by two common GP subtree mutation
operators: FULL and GROW mutation. In both cases we
find that the operators do have quite specific biases
and typically strongly oversample shorter strings.
%@ 0-7803-6658-1
@inproceedings{mcphee:2001:astamsbgplr,
abstract = {Understanding operator bias in evolutionary
computation is important because it is possible for the
operator's biases to work against the intended biases
induced by the fitness function. In recent work we
showed how developments in GP schema theory can be used
to better understand the biases induced by the standard
subtree crossover when genetic programming is applied
to variable length linear structures. We use the schema
theory to better understand the biases induced on
linear structures by two common GP subtree mutation
operators: FULL and GROW mutation. In both cases we
find that the operators do have quite specific biases
and typically strongly oversample shorter strings.},
added-at = {2008-06-19T17:35:00.000+0200},
address = {COEX, World Trade Center, 159 Samseong-dong,
Gangnam-gu, Seoul, Korea},
author = {McPhee, Nicholas Freitag and Poli, Riccardo and Rowe, Jonathan E.},
biburl = {https://www.bibsonomy.org/bibtex/23653952c4ceef16ff2f40ccfede84c34/brazovayeye},
booktitle = {Proceedings of the 2001 Congress on Evolutionary
Computation CEC2001},
interhash = {b90aa8ad5282b9ebd6cc9a4f62e66219},
intrahash = {3653952c4ceef16ff2f40ccfede84c34},
isbn = {0-7803-6658-1},
keywords = {algorithms, bias genetic linear mutation, programming, representation, schema size theory,},
month = {27-30 May},
notes = {CEC-2001 - A joint meeting of the IEEE, Evolutionary
Programming Society, Galesia, and the IEE.
IEEE Catalog Number = 01TH8546C,
Library of Congress Number = .
linear (unary) tree schemata. flat fitness landscape.
biases of full mutation, grow mutation,
No fitness. Full(unary) average length = 2*D-1.
Limiting size distribution: 0 for size < D, flat region
size < 2D, rapid falling size>=2D. Similar to subtree
crossover. Grow(unary) discrete gamma distribution (cf.
\cite{Rowe01} ) cf subtree crossover.
{"}ones then zeros{"} unary problem. Subtree crossover
bloat (at least to 75 generations). full no bloat,
actually as with no fitness, {"}artifact of this
particular problem{"}. Grow similar to no fitness.},
organisation = {IEEE Neural Network Council (NNC), Evolutionary
Programming Society (EPS), Institution of Electrical
Engineers (IEE)},
pages = {1078--1085},
publisher = {IEEE Press},
publisher_address = {445 Hoes Lane, P.O. Box 1331, Piscataway, NJ
08855-1331, USA},
timestamp = {2008-06-19T17:46:47.000+0200},
title = {A Schema Theory Analysis of Mutation Size Biases in
Genetic Programming with Linear Representations},
url = {http://citeseer.ist.psu.edu/501380.html},
year = 2001
}