Wavelet Neural Networks for Function Learning - 1995

《Wavelet Neural Networks for Function Learning - 1995》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。

1、IEEETRANSACTIONSONSIGNALPROCESSING,VOL.43,NO.6.JUNE19951485WaveletNeuralNetworksforFunctionLearningJunZhang,Member,IEEE,GilbertG.Walter,YuboMiao,andWanNgaiWayneLee,Member,IEEEAbshurct-Inthispaper,awavelet-basedneuralnetworkisnetworkscanbemucheasierthanMPLnetworks.Hence,describe

2、d.Thestructureofthisnetworkissimilartothatofthetherehasbeenconsiderableinterestintheimplementationofradialbasisfunction(RBF)network,exceptthatheretheradialRBFnetworks[3]-[5](alsoseethereferencesin[8])andthebasisfunctionsarereplacedbyorthonormalscalingfunctionsthattheoreticalana

3、lysisoftheirproperties,suchasapproximationarenotnecessarilyradial-symmetric.Theefficacyofthistypeofnetworkinfunctionlearningandestimationisdemonstratedabilityandconvergencerates[6]-[ti].throughtheoreticalanalysisandexperimentalresults.Inpartic-Fromthepointofviewoffunctionrepres

4、entation,anRBFular,ithasbeenshownthatthewaveletnetworkhasuniversalnetworkisaschemethatrepresentsafunctionofinterestbyandL2approximationpropertiesandisaconsistentfunctionusingmembersofafamilyofcompactly(orlocally)supportedestimator.Convergenceratesassociatedwiththesepropertiesba

5、sisfunctions.Thelocalityofthebasisfunctionsmakesareobtainedforcertainfunctionclasseswheretheratesavoidthe“curseofdimensionality.”Intheexperiments,thewavelettheIU3FnetworkmoresuitableinlearningfunctionswithnetworkperformedwellandcomparedfavorablytotheMLPlocalvariationsanddiscont

6、inuities.Furthermore,theRBFandRBFnetworks.networkscanrepresentanyfunctionthatisinthespacespannedbythefamilyofbasisfunctions.However,thebasisI.INTRODUCTIONfunctionsinthefamilyaregenerallynotorthogonalandareredundant.Thismeansthatforagivenfunction,itsRBFEVELOPINGMODELSfromobserve

7、ddata,orfunc-Dnetworkrepresentationisnotuniqueandisprobablynottionlearning,isafundamentalprobleminmanyfields,themostefficient.Inthiswork,wereplacethefamilyofsuchasstatisticaldataanalysis,signalprocessing,control,basisfunctionsfortheRBFnetworkbyanorthonormalbasis,forecasting,and

8、artificialintelligence.Thisproblemisalsonamely,thescal