Transcription of Decision T - Statistics
{{id}} {{{paragraph}}}
Example: bankruptcy
Decision T - Statistics
BayesianLearninginProbabilisticDecisionT reesMichaelI JordanMITC ollaboratorsRobertJacobs Rochester LeiXu HongKong Geo reyHinton Toronto StevenNowlan Synaptics MarinaMeila MIT LawrenceSaul MIT Outline decisiontrees probabilisticdecisiontrees EMalgorithmandextensions modelselection Bayesiancomputations empiricalresults systemidenti cation classi cation theoreticalresults trainingseterror testseterrorSomeproblemswithmulti layeredneuralnetworks thelearningalgorithmsareslow hardtounderstandthenetwork hardtobuildinpriorknowledge poorperformanceonnon stationarydata notnaturalforsomefunctionsSupervisedlear ning akaregression classi cation Weassumethatthelearnerisprovidedwithatra iningset X f x t y t gT wherexisaninputvectorandyisanoutputvecto r Wewillgaugeperformanceonatestset Xs f x t y t gTs Decisiontreesx < < < dropthedatasetdownthetree ateachnode tryto ndasplitoftheinputspace ahalf plane thatyieldsthelargestgainin purity onleftandright buildalargetreeandprunebackwardtocre ateanestedsequenceoftrees pickthebesttreefromthes
Sup ervised learning ak a regression classication W e assume that the learner is pro vided with a tr aining set X f x t y g T where x is an input ve ctor and y is an ...
Loading..
Tags:
Information
Domain:
Source:
Link to this page:
Please notify us if you found a problem with this document: