机器学习介绍(英文版：备注里有中文翻译)

MachineLearning2020年11月23日星期一Machinelearning,asabranchofartificialintelligence,isgeneraltermsofakindofanalyticalmethod.Itmainlyutilizescomputersimulateorrealizethelearnedbehaviorofhuman.2020年11月23日星期一2020年11月23日星期一1）Machinelearningjustlikeatruechampionwhichgohaughtily;2）Patternrecognitioninprocessofdeclineanddieout;3）Deeplearningisabrand-newandrapidlyrisingfield.theGooglesearchindexofthreeconceptsince20042020年11月23日星期一Theconstructedmachinelearningsystembasedoncomputermainlycontainstwocoreparts：representationandgeneralization.Thefirststepfordatalearningistorepresentthedata,i.e.detectthepatternofdata.Establishageneralizedmodelofdataspaceaccordingtoagroupofknowndatatopredictthenewdata.Thecoretargetofmachinelearningistogeneralizefromknownexperience.Generalizationmeansapowerofwhichthemachinelearningsystemtobelearnedforknowndatathatcouldpredictthenewdata.SupervisedlearningInputdatahaslabels.Thecommonkindoflearningalgorithmisclassification.Themodelhasbeentrainedviathecorrespondencebetweenfeatureandlabelofinputdata.Therefore,whensomeunknowndatawhichhasfeaturesbutnolabelinput,wecanpredictthelabelofunknowndataaccordingtotheexistingmodel.2020年11月23日星期一UnsupervisedlearningInputdatahasnolabels.Itrelatestoanotherlearningalgorithm,i.e.clustering.Thebasicdefinitionisacoursethatdividethegatherofphysicalorabstractobjectintomultipleclasswhichconsistofsimilarobjects.2020年11月23日星期一Iftheoutputeigenvectormarkscomefromalimitedsetthatconsistofclassornamevariable,thenthekindofmachinelearningbelongstoclassificationproblem.Ifoutputmarkisacontinuousvariable,thenthekindofmachinelearningbelongstoregressionproblem.2020年11月23日星期一ClassificationstepFeatureextractionFeatureselectionModeltrainingClassificationandpredictionRawdataNewdata2020年11月23日星期一Featureselection（featurereduction）CurseofDimensionality：Usuallyrefertotheproblemthatconcernedaboutcomputationofvector.Withtheincreaseofdimension,calculatedamountwilljumpexponentially.Corticalfeaturesofdifferentbrainregionsexhibitvarianteffectduringtheclassificationprocessandmayexistsomeredundantfeature.Inparticularafterthemultimodalfusion,theincreaseoffeaturedimensionwillcause“curseofDimensionality”.2020年11月23日星期一PrincipalComponentAnalysis,PCAPCAisthemostcommonlineardimensionreductionmethod.Itstargetismappingthedataofhighdimensiontolow-dimensionspaceviacertainlinearprojection,andexpectthevarianceofdatathatprojectthecorrespondingdimensionismaximum.Itcanusefewerdatadimensionmeanwhileretainthemajorcharacteristicofrawdata.2020年11月23日星期一Lineardiscriminantanalysis,LDAThebasicideaofLDAisprojection,mappingtheNdimensiondatatolow-dimensionspaceandseparatethebetween-groupsassoonaspossible.i.e.theoptimalseparabilityinthespace.Thebenchmarkisthenewsubspacehasmaximumbetweenclassdistanceandminimalinter-objectdistance.2020年11月23日星期一Independentcomponentanalysis,ICAThebasicideaofICAistoextracttheindependencesignalfromagroupofmixedobservedsignaloruseindependencesignaltorepresentothersignal.2020年11月23日星期一Recursivefeatureeliminationalgorithm,RFERFEisagreedyalgorithmthatwipeoffinsignificancefeaturestepbysteptoselectthefeature.Firstly,cyclicorderingthefeatureaccordingtotheweightofsub-featureinclassificationandremovethefeaturewhichrankatterminalonebyone.Then,accordingtothefinalfeatureorderinglist,selectdifferentdimensionofseveralfeaturesubsetfronttoback.Assesstheclassificationeffectofdifferentfeaturesubsetandthengettheoptimalfeaturesubset.2020年11月23日星期一ClassificationalgorithmDecisiontreeDecisiontreeisatreestructure.Eachnonleafnodeexpressesthetestofafeaturepropertyandeachbranchexpressestheoutputoffeaturepropertyincertainrangeandeachleafnodestoresaclass.Thedecision-makingcourseofdecisiontreeisstartingfromrootnode,testingthecorrespondingfeaturepropertyofwaitingobjects,selectingtheoutputbranchaccordingtotheirvalues,untilreachingtheleafnodeandtaketheclassthatleafnodestoreasthedecisionresult.2020年11月23日星期一NaiveBayes,NBNBclassificationalgorithmisaclassificationmethodinstatistics.Ituseprobabilitystatisticsknowledgeforclassification.Thisalgorithmcouldapplytolargedatabaseandithashighclassificationaccuracyandhighspeed.2020年11月23日星期一Artificialneuralnetwork,ANNANNisamathematicalmodelthatapplyakindofstructurewhichsimilarwithsynapseconnectionforinformationprocessing.Inthismodel,amassofnodeformanetwork,i.e.neuralnetwork,toreachthegoalofinformationprocessing.Neuralnetworkusuallyneedtotrain.Thecourseoftrainingisnetworklearning.Thetrainingchangethelinkweightofnetworknodeandmakeitpossessthefunctionofclassification.Thenetworkaftertrainingapplytorecognizeobject.2020年11月23日星期一k-NearestNeighbors,kNNkNNalgorithmisakindofclassificationmethodbaseonlivingexample.Thismethodistofindthenearestktrainingsampleswithunknownsamplexandexaminethemostofksamplesbelongtowhichclass,thenxbelongstothatclass.kNNisalazylearningmethod.Itstoressamplesbutproceedclassificationuntilneedtoclassify.Ifsamplesetarerelativelycomplex,itmaybeleadtolargecomputationoverhead.Soitcannotapplytostronglyreal-timeoccasion.2020年11月23日星期一supportvectormachine,SVMMappingthelinearlyinseparabl