Pareto, Population Partitioning, Price and Genetic
Programming
W. Langdon. Research Note, RN/95/29. University College London, Gower Street, London WC1E 6BT, UK, (April 1995)
Abstract
A description of a use of Pareto optimality in genetic
programming is given and an analogy with Genetic
Algorithm fitness niches is drawn. Techniques to either
spread the population across many pareto optimal
fitness values or to reduce the spread are described.
It is speculated that a wide spread may not aid Genetic
Programming. It is suggested that this might give
useful insight into many GPs whose fitness is composed
of several sub-objectives.
The successful use of demic populations in GP leads to
speculation that smaller evolutionary steps might aid
GP in the long run.
An example is given where Price's covariance theorem
helped when designing a GP fitness function.
%0 Report
%1 Langdon:1995:ppp
%A Langdon, W. B.
%C Gower Street, London WC1E 6BT, UK
%D 1995
%K Artificial Automatic Demes Evolution, Learning, Machine Pareto Programming, algorithms, fitness, genetic programming,
%N RN/95/29
%T Pareto, Population Partitioning, Price and Genetic
Programming
%U http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/WBL_aaai-pppGP.ps
%X A description of a use of Pareto optimality in genetic
programming is given and an analogy with Genetic
Algorithm fitness niches is drawn. Techniques to either
spread the population across many pareto optimal
fitness values or to reduce the spread are described.
It is speculated that a wide spread may not aid Genetic
Programming. It is suggested that this might give
useful insight into many GPs whose fitness is composed
of several sub-objectives.
The successful use of demic populations in GP leads to
speculation that smaller evolutionary steps might aid
GP in the long run.
An example is given where Price's covariance theorem
helped when designing a GP fitness function.
@techreport{Langdon:1995:ppp,
abstract = {A description of a use of Pareto optimality in genetic
programming is given and an analogy with Genetic
Algorithm fitness niches is drawn. Techniques to either
spread the population across many pareto optimal
fitness values or to reduce the spread are described.
It is speculated that a wide spread may not aid Genetic
Programming. It is suggested that this might give
useful insight into many GPs whose fitness is composed
of several sub-objectives.
The successful use of demic populations in GP leads to
speculation that smaller evolutionary steps might aid
GP in the long run.
An example is given where Price's covariance theorem
helped when designing a GP fitness function.},
added-at = {2008-06-19T17:35:00.000+0200},
address = {Gower Street, London WC1E 6BT, UK},
author = {Langdon, W. B.},
biburl = {https://www.bibsonomy.org/bibtex/231a4a9e730e1ad796403ceed0b8f5bfc/brazovayeye},
institution = {University College London},
interhash = {7b9b66eda60370aae697cf5dd540b138},
intrahash = {31a4a9e730e1ad796403ceed0b8f5bfc},
keywords = {Artificial Automatic Demes Evolution, Learning, Machine Pareto Programming, algorithms, fitness, genetic programming,},
month = {April},
notes = {Accepted by AAAI Fall 1995 Genetic Programming
Symposium but withdrawn due to time pressures},
number = {RN/95/29},
size = {11 pages},
timestamp = {2008-06-19T17:44:42.000+0200},
title = {Pareto, Population Partitioning, Price and Genetic
Programming},
type = {Research Note},
url = {http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/WBL_aaai-pppGP.ps},
year = 1995
}