文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Global optimisation Evolutionary neural networks Evolutionary algorithms Genetic algorithms
侵权投诉

Global optimisation Evolutionary neural networks Evolutionary algorithms Genetic algorithms

developments and future directions of bio-inspired

10

ECOLOGICALINFORMATICS1(2006)9–22

obtainverydifferentfitnessvaluesindifferentpopulations.Coevolutionhasbeenstudiedextensivelyinrecentyearsasawaytodecomposingacomplexproblemautomatically(Dar-wenandYao,1997;Khareetal.,2004;Liuetal.,2001).Givenacomplexproblem,theideahereistousecoevolutiontoformspeciesautomatically,whereeachspeciesisspecialisedinsolvingonlypartsoftheentireproblem.Theentireproblemisthensolvedbycombiningalldifferentspeciestogether.Thiscanberegardedasanautomaticdivide-and-conquerapproachtocomplexproblemsolving.Wewillgiveoneexam-pleinoptimisationinthispapertoillustratehowproblemdecompositioncouldbebeneficial(Liuetal.,2001).

Anotherexampleofautomaticproblemdecompositionistheneuralnetworkensemblesthatwehaveproposed(Mon-irulIslametal.,2003;Liuetal.,2000;ChandraandYao,2004).Insteadofdesigningandtrainingamonolithicneuralnetworkforalargeandcomplexproblem,agroupofsim-plerneuralnetworkswereconstructedandtrained.Negativecorrelationandcoevolutionhavebeenusedtoincreasethediversitywithintheensembleofneuralnetworksandim-provethegeneralisationabilityoftheentireneuralnetworkensemble.Excellentresultshavebeenreportedintheliter-ature(MonirulIslametal.,2003;Liuetal.,2000;ChandraandYao,2004).

Theworkintroducedsofarcanberegardedastheapplica-tionofbio-inspiredapproaches(e.g.,evolutionarycomputa-tionandneuralnetworkensembles)toecoinformaticsproblems,wheretheinformationflowappearstobefromcomputationaltechniquestoproblemsrelatedtoecosystems.However,thereisanoppositedirectionofinformationflow,wheresomeideasandconceptsinbiologicalecosystemshaveinspiredresearcherstoexplorenewformsofcomputationandnewalgorithmsforproblemsolving.Thisisaveryimportantpartofecoinformaticsthatisunderexploredandunderstudiedatthemoment.Wehopeourpaperwillencouragemoreresearchersinmovingintothesenewresearchareas,whichrequireclosecollaborationbetweencomputerscientistsandecologists.

Therestofthispaperwillbeorganisedasfollows.Section2discussesourmodellingworkusingthetwostageevolution-aryapproaches(Lietal.,2004).Section3describesourworkoncoevolutionaryoptimisation(Liuetal.,2001).Section4pre-sentsneuralnetworkensemblesandnegativecorrelationlearning(ChandraandYao,2004).Section5outlinesthefutureresearchdirectionsforpossibleecosystem-inspiredcomputa-tionalmodelsandalgorithms.Finally,Section6concludesthepaperwithabriefsummary.

2.Data-drivenevolutionarymodelling

Therearemanyapproachesthatonecouldusefordata-drivenevolutionarymodelling(Lietal.,2002,2004).Wewillreviewoneexample(Lietal.,2004)herewheregeneticprogrammingwasusedtoevolvemodelstructures(i.e.,explicitfunctionforms)first.Thencommon“features”fromevolvedmodelswereidentifiedandabstracted.Finally,theabstractedmodelwascalibratedbyusingevolutionaryprogrammingtofine-tunemodelparameters.MoredetailsabouttheexamplecanbefoundinLietal.(2004).

2.1.Distributionoflightinellipticalgalaxies(Lietal.,

2004)

Inopticalimages,galaxieshaveawiderangeofshapesandsizes.Todevelopadeeperunderstandingoftheevolutionofgalaxies,itisimportanttoquantifytheirmorphologicalfea-turesasafirststageinanattempttolinkthemtothephysicsoftheirformation.

ThemostcommonlyusedfittingfunctionsforellipticalgalaxyprofilesaretheHubble'slawgivenbyIðrÞ¼

I0ðr=aþ1Þ;

ð1Þ

andthedeVaucouleursr1/4lawgivenbyIðrÞ¼IeexpfÀ3:33½ðr=reÞ1=4À1 g:

ð2Þ

Eachfunctionhastwoparameters,abrightnessscaleI0orIeandasizescaleaorre.I0isthesurfacebrightnessatr=0,i.e.,thecentralsurfacebrightness.Thescalelengthaisthedis-tanceatwhichthesurfacebrightnessfallstoaquarterofitscentralvalue.Incontrast,re,knownastheeffectiveradius,containshalfofthetotallightofthegalaxy,andIeisthebrightnessatre.Thetwolawsgiveareasonablygooddescrip-tionoftheluminosityprofilesofellipticalgalaxiesandthebulgesofsometypesofspiralgalaxies.However,theydonotdescribewelltheseprofilesbeyondcertainboundsofrorthebulgesofspiralgalaxieswithpronouncedspiralfeatures.Forthelatter,apopularmodelisanexponentialformIðrÞ¼5:36Ieexp½À1:68ðr=reÞ :

ð3Þ

NeitherEq.(2)norEq.(3)hasbeenformallylinkedtoaphysicalmodelofgalaxyformation,andtheyremainmereempiricalfitstoobservedprofiles.Hubble'smodel(1)isnotaverygoodfittoawiderangeofobservedellipticals,butisrelatedtopossibledensityprofilesofself-gravitatingisother-malspheresofgas.Itwouldbedesirabletohaveasinglemodelthatwoulddescribeallellipticalgalaxieswithavaria-tionofparameters,especiallyifthismodelcouldbelinkedwiththephysicsofcollapsinggascloudsandtheformationofself-gravitatingellipsoidalsystems.

Therehasbeenalotofeffortintheliteraturetofindap-propriatemathematicalfunctionstodescribeawiderangeofprofilesofvariouscomponentsofspiralandellipticalgalaxies,andtofindplausiblevaluesforparametersthatwouldfitobservedprofiles.Bothtasksarenon-trivial.Theusualap-proachistopostulateamathematicalmodelcomprisingofcommonmathematicalfunctions,thentoapplyfittingalgo-rithmstofindsuitableparametersforthesefunctions.Theparameterfittingalgorithmusuallyadoptedisthenon-linearreducedχ2minimisationgivenbyv2¼

1X½ImodelðiÞÀIobsðiÞ 2

m;

ð4Þ

idwhereIobs(i)istheindividualobservedprofilevalue,Imodel(i)isthevaluecalculatedfromthefittingfunction,νisthenumberofdegreesoffreedom,andδisthestandarddeviationofthedata.Onedisadvantageofnon-linearminimisationalgo-rithmsistheirsensitivitytotheinitialvaluesprovided.

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email