Trend or No Trend

《Trend or No Trend》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、TrendorNoTrend:ANovelNonparametricMethodforClassifyingTimeSeriesbyStanislavNikolovS.B.,MassachusettsInstituteofTechnology(2011)SubmittedtotheDepartmentofElectricalEngineeringandComputerScienceinpartialfulfillmentoftherequirementsforthedegreeofMasterofEngineeringinElectricalEngine

2、eringandComputerScienceattheMASSACHUSETTSINSTITUTEOFTECHNOLOGYSeptember2012cMassachusettsInstituteofTechnology2012.Allrightsreserved.Author..............................................................DepartmentofElectricalEngineeringandComputerScienceAugust15,2012Certifiedby....

3、......................................................Prof.DevavratShahJamiesonCareerDevelopmentAssociateProfessorofElectricalEngineeringandComputerScienceThesisSupervisorAugust15,2012Certifiedby..........................................................Dr.SatanjeevBanerjeeEnginee

4、r,TwitterInc.ThesisCo-SupervisorAugust15,2012Acceptedby.........................................................Prof.DennisM.FreemanChairman,MastersofEngineeringThesisCommittee2TrendorNoTrend:ANovelNonparametricMethodforClassifyingTimeSeriesbyStanislavNikolovSubmittedtotheDepart

5、mentofElectricalEngineeringandComputerScienceonAugust15,2012,inpartialfulfillmentoftherequirementsforthedegreeofMasterofEngineeringinElectricalEngineeringandComputerScienceAbstractInsupervisedclassification,oneattemptstolearnamodelofhowobjectsmaptolabelsbyselectingthebestmodelfrom

6、somemodelspace.Thechoiceofmodelspaceencodesassumptionsabouttheproblem.Weproposeasettingformodelspecificationandselectioninsupervisedlearningbasedonalatentsourcemodel.Inthissetting,wespecifythemodelbyasmallcollectionofunknownlatentsourcesandpositthatthereisastochasticmodelrelating

7、latentsourcesandobservations.Withthissettinginmind,weproposeanonparametricclassificationmethodthatisentirelyunawareofthestructureoftheselatentsources.Instead,ourmethodreliesonthedataasaproxyfortheunknownlatentsources.Weperformclassificationbycomputingtheconditionalclassprobabiliti

8、esforanobservationbasedonourstochasticmodel.Thi