claude tadonki and jean-philippe vialnew

claude tadonki and jean-philippe vialnew

ID:34621783

大小:285.39 KB

页数:17页

时间:2019-03-08

claude tadonki and jean-philippe vialnew_第1页
claude tadonki and jean-philippe vialnew_第2页
claude tadonki and jean-philippe vialnew_第3页
claude tadonki and jean-philippe vialnew_第4页
claude tadonki and jean-philippe vialnew_第5页
资源描述:

《claude tadonki and jean-philippe vialnew》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库

1、TheLinearSeparationProblemRevisitedwithACCPMClaudeTADONKIandJean-PhilippeVIALHEC/LOGILAB,40BdduPontd'Arve,CH-1211Geneva,SwitzerlandEmail:{claude.tadonki,jean-philippe.vial}@hec.unige.chMay21,2002AbstractWeinvestigatetheuseofaccpm(analyticcentercuttingplanemetho

2、d)forsolvingthelinearseparationproblem,whichisanimportantinstanceofthegeneraldataminingconcept.Giventwodisjointsubsetsofpoints,theproblemistondahyperplanewhichseparatesthesetwosubsetsaswellaspossible.Ourmethodconsistsofminimizingthetotalsumofmisclassicationgap

3、s.Foralargeamountofdata,thenumberofglobaliterationsintheaccpmprocesscanbereducedbysplittingtheobjectiveintoseveralparts.Hence,wereducethetotaltimecostinvolvedbytheevaluationsoftheobjectiveanditssubgradients.However,asthenumberofcutsincreaseswiththenumberofsubfun

4、ctions,thetimecostassociatedwithbecomesconsiderable.Therefore,wehavetoachieveacertainbalancebetweenthegainintheobjectiveevaluationsandthepricewepayfortheadditionalcutsgenerated.Afterprovidinganalgorithmtocomputeagoodstartingpointwhichyieldsafasterconvergence,wep

5、erformsomeexperimentswithfoursampledatabasesovervariousdisaggregatedforms.Wehaveuseda500MhzSUNUltrasparcwith256MBofRAM.Thereportofmeasuredexecutiontimesshowsthatthemethodisquiteecient,andthecompromisepreviouslyexplainediswellillustrated.Anasymptotictimeanalysis

6、isprovidedandfromthis,wederiveanexpressionapproximatingthevalueoftheoptimalpartitionsize.Keywords:Optimization,convexprogramming,linearseparation,heuristic,complexity.1IntroductionTheseparationproblemisaninstanceofthemoregeneralcaseofthedataminingconcept[14],awi

7、delystudiedtopic.Peoplewithvariedinterestsandbackgrounds(statisticians,economists,physicians,chemists,biologists,mathematicians,computerscientists,...)havedevotedasubstantialpartoftheiractivitiesonthisdataclassicationapproach.Successfulapplicationsarereportedin

8、manydierentareas,e.g.cancerdiagnosis[21],humangenome[15],gamestrategies[19],patternrecognition[22],decision/selectionmaking[30],andmore.FNRSPosdoctoralposition12Cla

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。