An Introduction to Probabilistic Graphical Models for Relational Data 概率图模型
AnIntroductionto
ProbabilisticGraphicalModelsforRelationalData
LiseGetoor
ComputerScienceDepartment/UMIACS
UniversityofMaryland
CollegePark,MD20740
getoor@cs.umd.edu
Abstract
Wesurveysomeoftherecentworkonprobabilisticgraphicalmodelsforrelationaldata.Themodelsthatwedescribeareallbasedupon’graphicalmodels’[12].Themodelscancapturestatisticalcorrelationsamongattributeswithinasinglerelationaltable,betweenattributesindifferenttables,andcancapturecertainstructuralproperties,suchastheexpectedsizeofajoinbetweentables.Thesemodelscanthenbeusedforavarietyoftasksincluding llinginmissingvalues,datasummarizationandanomalydetection.Herewedescribetwocomplementarysemanticsforthemodels:oneapproachsuitedtomakingprobabilisticstatementsaboutindividualsandthesecondapproachsuitedtomakingstatementsaboutfrequenciesinrelationaldata.Afterdescribingthesemantics,webrie ydescribealgorithmsforautomaticallyconstructingthemodelsfromanexisting(non-probabilistic)database.
1Introduction
Therehasbeenagrowinginterestinprobabilisticdatabases.Thisgrowinginterestislargelyduetoexpandingdatacollectionopportunitiesandcapabilities.Inaddition,recentadvancesinprobabilisticmodeling,bothintermsofef cientlearningandestimationalgorithmsandpracticalinferencealgorithmshavemadetheuseofthesemodelsforlarge-scaledomainspractical.Inthissurveypaper,wereviewsomeoftheworkdevelopedintheprobabilisticreasoningandmachinelearningcommunitiescommonlyreferedtoasprobabilisticrelationalmodels(PRMs)[11].PRMsarebasedonexistingprobabilisticmodels(typicallyBayesiannetworks)andextendthemtorelationaland rst-orderdomains.
Halpern[9]identi edtwodistinctsemanticsfor rst-orderlogicsofprobability,referedtoasthepossible-worldsapproachandthedomain-frequencyapproach.Inthepossible-worldsapproach,weputadistributionoverpossibleworlds.Thedistributioncanbeusedtoanswerdegreeofbeliefquestionssuchas‘Whatistheprobabilitythataparticularperson,JoeSmith,haspurchasedaparticulartractorpart,WidgetSupreme?’.Inthiscase,thesemanticsoftheformulaisde nedbysumminguptheprobabilityoverallofthepossibleworldsinwhichJoeSmithhasboughtWidgetSupremes.However,asmanyhavenoted[2,1],therearedif cultieswithmakingstatisticalassertions,suchas‘75%ofthepurchaseswereforwidgets’.Thesecondapproach,referedto
1
Getting Started in Probabilistic Graphical Models_IT/计算机_专业资料。Message...Mixed membership analysis of highthroughput interaction studies: Relational data....
Probabilistic Graphical Models_IT/计算机_专业资料。贝叶斯网络的推理,主要介绍了变量消元法。Probabilistic Graphical Model 一.Representation(表示) 二.Inference(...
Survey of probabilistic graphical models_互联网_IT/计算机_专业资料。2013 10...network and Markov logic network in the area of statistical relational ...
Multi-relational data mi... 暂无评价 12页 免费 Probabilistic Graphical ..... Probabilistic Relational... 暂无评价 46页 免费 Probabilistic models for......
34 Classes of graphical models Probabilistic models Graphical models Directed ...? “AB: Automated Biologist” 62 Learning from relational data Can we ...
An important relevant class for relational data Discussed later QMR (Quick ...graphical models Koller and Friedman, Probabilistic Graphical Models: Principles...
An Introduction to Graphical Models Primary textbook_专业资料。An ...Structured data (text, web pages ...): probabilistic relational models 6 ...
Eric Xing @ CMU, 2006-2010 X8 Nucleus 5 Probabilistic Graphical Models, ...Model parameters yi i=1:n Data = {y1,…yn} Plate = rectangle in ...
An Empirical Study of Ef?ciency and Accuracy of Probabilistic Graphical Models Jens Dalgaard Nielsen and Manfred Jaeger Institut for Datalogi, Aalborg ...
models combine ?rst-order logical languages with probabilistic graphical models...Koller. Discriminative probabilistic models for relational data. In Proc. of ...
热门文档