Feature selection is the process of finding the set of inputs to a
machine learning algorithm that will yield the best performance.
Developing a way to solve this problem automatically would make current
machine learning methods much more useful. Previous efforts to automate
feature selection rely on expensive meta-learning or are applicable
only when labeled training data is available. This paper presents
a novel method called FS-NEAT which extends the NEAT neuroevolution
method to automatically determine an appropriate set of inputs for
the networks it evolves. By learning the network's inputs, topology,
and weights simultaneously, FS-NEAT addresses the feature selection
problem without relying on meta-learning or labeled data. Initial
experiments in an autonomous car racing simulation demonstrate that
FS-NEAT can learn better and faster than regular NEAT. In addition,
the networks it evolves are smaller and require fewer inputs. Furthermore,
FS-NEAT's performance remains robust even as the feature selection
task it faces is made increasingly difficult.
%0 Conference Paper
%1 Whiteson2005
%A Whiteson, Shimon
%A Stone, Peter
%A Stanley, Kenneth O.
%A Miikkulainen, Risto
%A Kohl, Nate
%B GECCO '05: Proceedings of the 2005 conference on Genetic and evolutionary
computation
%C New York, NY, USA
%D 2005
%I ACM
%K imported
%P 1225--1232
%R http://doi.acm.org/10.1145/1068009.1068210
%T Automatic feature selection in neuroevolution
%X Feature selection is the process of finding the set of inputs to a
machine learning algorithm that will yield the best performance.
Developing a way to solve this problem automatically would make current
machine learning methods much more useful. Previous efforts to automate
feature selection rely on expensive meta-learning or are applicable
only when labeled training data is available. This paper presents
a novel method called FS-NEAT which extends the NEAT neuroevolution
method to automatically determine an appropriate set of inputs for
the networks it evolves. By learning the network's inputs, topology,
and weights simultaneously, FS-NEAT addresses the feature selection
problem without relying on meta-learning or labeled data. Initial
experiments in an autonomous car racing simulation demonstrate that
FS-NEAT can learn better and faster than regular NEAT. In addition,
the networks it evolves are smaller and require fewer inputs. Furthermore,
FS-NEAT's performance remains robust even as the feature selection
task it faces is made increasingly difficult.
%@ 1-59593-010-8
@inproceedings{Whiteson2005,
abstract = {Feature selection is the process of finding the set of inputs to a
machine learning algorithm that will yield the best performance.
Developing a way to solve this problem automatically would make current
machine learning methods much more useful. Previous efforts to automate
feature selection rely on expensive meta-learning or are applicable
only when labeled training data is available. This paper presents
a novel method called FS-NEAT which extends the NEAT neuroevolution
method to automatically determine an appropriate set of inputs for
the networks it evolves. By learning the network's inputs, topology,
and weights simultaneously, FS-NEAT addresses the feature selection
problem without relying on meta-learning or labeled data. Initial
experiments in an autonomous car racing simulation demonstrate that
FS-NEAT can learn better and faster than regular NEAT. In addition,
the networks it evolves are smaller and require fewer inputs. Furthermore,
FS-NEAT's performance remains robust even as the feature selection
task it faces is made increasingly difficult.},
added-at = {2009-09-12T19:19:34.000+0200},
address = {New York, NY, USA},
author = {Whiteson, Shimon and Stone, Peter and Stanley, Kenneth O. and Miikkulainen, Risto and Kohl, Nate},
biburl = {https://www.bibsonomy.org/bibtex/222ec46fc7d7a5fd0b93b281572793a43/mozaher},
booktitle = {GECCO '05: Proceedings of the 2005 conference on Genetic and evolutionary
computation},
doi = {http://doi.acm.org/10.1145/1068009.1068210},
file = {:Whiteson2005.pdf:PDF},
interhash = {7c80b6bcfa9c4055da6146fc1afb0e25},
intrahash = {22ec46fc7d7a5fd0b93b281572793a43},
isbn = {1-59593-010-8},
keywords = {imported},
location = {Washington DC, USA},
owner = {Mozaher},
pages = {1225--1232},
publisher = {ACM},
timestamp = {2009-09-12T19:19:43.000+0200},
title = {Automatic feature selection in neuroevolution},
year = 2005
}