資源描述：
《data mining neural networks with genetic algorithms》由會(huì)員上傳分享，免費(fèi)在線(xiàn)閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、DataminingneuralnetworkswithgeneticalgorithmsAjitNarayanan,EdwardKeedwellandDraganSavicSchoolofEngineeringandComputerScienceUniversityofExeterExeterEX44PTUnitedKingdomajit@dcs.ex.ac.uktel:(+)1392264064AbstractItisanopenquestionastowhatisthebestwaytoextractsymbolicrulesfromtrainedneuralnetwork
2、sindomainsinvolvingclassification.Previousapproachesbasedonanexhaustiveanalysisofnetworkconnectionandoutputvalueshavealreadybeendemonstratedtobeintractableinthatthescale-upfactorincreasesexponentiallywiththenumberofnodesandconnectionsinthenetwork.Anovelapproachusinggeneticalgorithmstosearchfo
3、rsymbolicrulesinatrainedneuralnetworkisdemonstratedinthispaper.Preliminaryexperimentsinvolvingclassificationarereportedhere,withtheresultsindicatingthatourproposedapproachissuccessfulinextractingrules.Whileitisacceptedthatfurtherworkisrequiredtoconvincinglydemonstratethesuperiorityofourapproa
4、choverothers,thereisneverthelesssufficientnoveltyintheseresultstojustifyearlydissemination.(Ifthepaperisaccepted,thelatestresultswillbereported,togetherwithsufficientinformationtoaidreplicabilityandverification.)IntroductionArtificialneuralnetworks(ANNs)areincreasinglyusedinproblemdomainsinvo
5、lvingclassification.Theyareadeptatfindingcommonalitiesinasetofseeminglyunrelateddataandforthisreasonareusedinagrowingnumberofclassificationtasks.Unfortunately,acommonlyperceivedproblemwithANNswhenusedforclassificationisthat,whileatrainedANNcanindeedclassifythedata,sometimeswithmoreaccuracytha
6、natraditional,symbolicmachinelearningapproach,thereasonsfortheirclassificationcannotbefoundeasily.TrainedANNsarecommonlyperceivedtobe‘blackboxes’whichmapinputdataontoaclassthroughanumberofmathematicallyweightedconnectionsbetweenlayersofneurons.WhiletheideaofANNsasblackboxesmaynotbeaproblemina
7、pplicationswherethereislittleinterestinthereasonsbehindclassification,thiscanbeamajorobstacleinapplicationswhereitisimportanttohavesymbolicrulesorotherformsofknowledgestructure,suchasidentificationordecisiontrees,whichareeasilyinterpretableby
