%0 Conference Paper %1 Whiteson:2005:gecco %A Whiteson, Shimon %A Stone, Peter %A Stanley, Kenneth O. %A Miikkulainen, Risto %A Kohl, Nate %B GECCO'05: Proc. 7th Genetic and Evolutionary Computation Conf. %C Washington, DC %D 2005 %E Beyer, Hans-Georg %E O'Reilly, Una-May %E Arnold, Dirk V. %E Banzhaf, Wolfgang %E Blum, Christian %E Bonabeau, Eric W. %E Cantu-Paz, Erick %E Dasgupta, Dipankar %E Deb, Kalyanmoy %E Foster, James A. %E de Jong, Edwin D. %E Lipson, Hod %E Llora, Xavier %E Mancoridis, Spiros %E Pelikan, Martin %E Raidl, Guenther R. %E Soule, Terence %E Tyrrell, Andy M. %E Watson, Jean-Paul %E Zitzler, Eckart %I ACM Press %K A-Life, Adaptive Algorithms, Ant Applications, Artificial Behaviour, Biological Coevolution, Colony Combinatorial Distribution Engineering Estimation Evolutionary Evolvable Hardware, Immune Intelligence, Local Meta-heuristics Multi-objective Optimisation Optimisation, Optimization, Programming, Real Robotics Search, Search-based Software Strategies, Swarm Systems, World algorithms, and genetic of programming, thesis %P 1225--1232 %R 10.1145/1068009.1068210 %T Automatic Feature Selection in Neuroevolution %U http://portal.acm.org/citation.cfm?id=1068009&jmp=cit&coll=GUIDE&dl=GUIDE&CFID=48779769&CFTOKEN=55479664#supp %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{Whiteson:2005:gecco, 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 = {2017-03-16T11:50:55.000+0100}, address = {Washington, DC}, author = {Whiteson, Shimon and Stone, Peter and Stanley, Kenneth O. and Miikkulainen, Risto and Kohl, Nate}, biburl = {https://www.bibsonomy.org/bibtex/2eaf541ed8f056b6b6aa0d6145d3025dc/krevelen}, booktitle = {GECCO'05: Proc. 7th Genetic and Evolutionary Computation Conf.}, crossref = {gecco:2005}, doi = {10.1145/1068009.1068210}, editor = {Beyer, Hans-Georg and O'Reilly, Una-May and Arnold, Dirk V. and Banzhaf, Wolfgang and Blum, Christian and Bonabeau, Eric W. and Cantu-Paz, Erick and Dasgupta, Dipankar and Deb, Kalyanmoy and Foster, James A. and de Jong, Edwin D. and Lipson, Hod and Llora, Xavier and Mancoridis, Spiros and Pelikan, Martin and Raidl, Guenther R. and Soule, Terence and Tyrrell, Andy M. and Watson, Jean-Paul and Zitzler, Eckart}, interhash = {7c80b6bcfa9c4055da6146fc1afb0e25}, intrahash = {eaf541ed8f056b6b6aa0d6145d3025dc}, isbn = {1-59593-010-8}, keywords = {A-Life, Adaptive Algorithms, Ant Applications, Artificial Behaviour, Biological Coevolution, Colony Combinatorial Distribution Engineering Estimation Evolutionary Evolvable Hardware, Immune Intelligence, Local Meta-heuristics Multi-objective Optimisation Optimisation, Optimization, Programming, Real Robotics Search, Search-based Software Strategies, Swarm Systems, World algorithms, and genetic of programming, thesis}, notes = {GECCO-2005 A joint meeting of the fourteenth international conference on genetic algorithms (ICGA-2005) and the tenth annual genetic programming conference (GP-2005). ACM Order Number 910052}, organisation = {ACM SIGEVO (formerly ISGEC)}, owner = {Rick}, pages = {1225--1232}, publisher = {ACM Press}, publisher_address = {New York}, timestamp = {2017-03-16T11:54:14.000+0100}, title = {Automatic Feature Selection in Neuroevolution}, url = {http://portal.acm.org/citation.cfm?id=1068009&jmp=cit&coll=GUIDE&dl=GUIDE&CFID=48779769&CFTOKEN=55479664#supp}, year = 2005 }