Shortcomings with Tree-Structured Edge Encodings for
Neural Networks
G. Hornby. Genetic and Evolutionary Computation -- GECCO-2004,
Part II, volume 3103 of Lecture Notes in Computer Science, page 495--506. Seattle, WA, USA, Springer-Verlag, (26-30 June 2004)
DOI: doi:10.1007/b98645
Abstract
In evolutionary algorithms a common method for
encoding neural networks is to use a tree-structured
assembly procedure for constructing them. Since node
operators have difficulties in specifying edge weights
and these operators are execution-order dependent, an
alternative is to use edge operators. Here we identify
three problems with edge operators: in the
initialisation phase most randomly created genotypes
produce an incorrect number of inputs and outputs;
variation operators can easily change the number of
input/output (I/O) units; and units have a connectivity
bias based on their order of creation. Instead of
creating I/O nodes as part of the construction process
we propose using parameterised operators to connect to
pre-existing I/O units. Results from experiments show
that these parameterized operators greatly improve the
probability of creating and maintaining networks with
the correct number of I/O units, remove the
connectivity bias with I/O units and produce better
controllers for a goal-scoring task.
Genetic and Evolutionary Computation -- GECCO-2004,
Part II
year
2004
month
26-30 June
pages
495--506
publisher
Springer-Verlag
series
Lecture Notes in Computer Science
volume
3103
issn
0302-9743
organisation
ISGEC
publisher_address
Heidelberg
size
12 pages
isbn
3-540-22343-6
notes
GECCO-2004 A joint meeting of the thirteenth
international conference on genetic algorithms
(ICGA-2004) and the ninth annual genetic programming
conference (GP-2004)
football
%0 Conference Paper
%1 Hornby:SwT:gecco2004
%A Hornby, Gregory S.
%B Genetic and Evolutionary Computation -- GECCO-2004,
Part II
%C Seattle, WA, USA
%D 2004
%E Deb, Kalyanmoy
%E Poli, Riccardo
%E Banzhaf, Wolfgang
%E Beyer, Hans-Georg
%E Burke, Edmund
%E Darwen, Paul
%E Dasgupta, Dipankar
%E Floreano, Dario
%E Foster, James
%E Harman, Mark
%E Holland, Owen
%E Lanzi, Pier Luca
%E Spector, Lee
%E Tettamanzi, Andrea
%E Thierens, Dirk
%E Tyrrell, Andy
%I Springer-Verlag
%K algorithms, genetic graphs, networks, neural programming, representation
%P 495--506
%R doi:10.1007/b98645
%T Shortcomings with Tree-Structured Edge Encodings for
Neural Networks
%U http://link.springer.de/link/service/series/0558/papers/3103/31030495.pdf
%V 3103
%X In evolutionary algorithms a common method for
encoding neural networks is to use a tree-structured
assembly procedure for constructing them. Since node
operators have difficulties in specifying edge weights
and these operators are execution-order dependent, an
alternative is to use edge operators. Here we identify
three problems with edge operators: in the
initialisation phase most randomly created genotypes
produce an incorrect number of inputs and outputs;
variation operators can easily change the number of
input/output (I/O) units; and units have a connectivity
bias based on their order of creation. Instead of
creating I/O nodes as part of the construction process
we propose using parameterised operators to connect to
pre-existing I/O units. Results from experiments show
that these parameterized operators greatly improve the
probability of creating and maintaining networks with
the correct number of I/O units, remove the
connectivity bias with I/O units and produce better
controllers for a goal-scoring task.
%@ 3-540-22343-6
@inproceedings{Hornby:SwT:gecco2004,
abstract = {In evolutionary algorithms a common method for
encoding neural networks is to use a tree-structured
assembly procedure for constructing them. Since node
operators have difficulties in specifying edge weights
and these operators are execution-order dependent, an
alternative is to use edge operators. Here we identify
three problems with edge operators: in the
initialisation phase most randomly created genotypes
produce an incorrect number of inputs and outputs;
variation operators can easily change the number of
input/output (I/O) units; and units have a connectivity
bias based on their order of creation. Instead of
creating I/O nodes as part of the construction process
we propose using parameterised operators to connect to
pre-existing I/O units. Results from experiments show
that these parameterized operators greatly improve the
probability of creating and maintaining networks with
the correct number of I/O units, remove the
connectivity bias with I/O units and produce better
controllers for a goal-scoring task.},
added-at = {2008-06-19T17:35:00.000+0200},
address = {Seattle, WA, USA},
author = {Hornby, Gregory S.},
biburl = {https://www.bibsonomy.org/bibtex/2a7f58a14937f88cc2bdb33b3b1cc3b85/brazovayeye},
booktitle = {Genetic and Evolutionary Computation -- GECCO-2004,
Part II},
doi = {doi:10.1007/b98645},
editor = {Deb, Kalyanmoy and Poli, Riccardo and Banzhaf, Wolfgang and Beyer, Hans-Georg and Burke, Edmund and Darwen, Paul and Dasgupta, Dipankar and Floreano, Dario and Foster, James and Harman, Mark and Holland, Owen and Lanzi, Pier Luca and Spector, Lee and Tettamanzi, Andrea and Thierens, Dirk and Tyrrell, Andy},
interhash = {39fc5d314732c7eb52de2498f43f4546},
intrahash = {a7f58a14937f88cc2bdb33b3b1cc3b85},
isbn = {3-540-22343-6},
issn = {0302-9743},
keywords = {algorithms, genetic graphs, networks, neural programming, representation},
month = {26-30 June},
notes = {GECCO-2004 A joint meeting of the thirteenth
international conference on genetic algorithms
(ICGA-2004) and the ninth annual genetic programming
conference (GP-2004)
football},
organisation = {ISGEC},
pages = {495--506},
publisher = {Springer-Verlag},
publisher_address = {Heidelberg},
series = {Lecture Notes in Computer Science},
size = {12 pages},
timestamp = {2008-06-19T17:41:45.000+0200},
title = {Shortcomings with Tree-Structured Edge Encodings for
Neural Networks},
url = {http://link.springer.de/link/service/series/0558/papers/3103/31030495.pdf},
volume = 3103,
year = 2004
}