资源描述：
《The Hugin Tool for Learning Bayesian Networks》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、TheHuginToolforLearningBayesianNetworksAndersL.Madsen,MichaelLang,UffeB.Kjærulff,andFrankJensenHuginExpertA/SNielsJernesVej10DK-9220AalborgØDenmark{Anders.L.Madsen,Michael.Lang,Uffe.Kjaerulff,Frank.Jensen}@hugin.comAbstract.Inthispaper,wedescribetheHuginToolasanefficienttoolforknowledgedisco
2、verythroughconstructionofBayesiannetworksbyfusionofdataanddomainexpertknowledge.TheHuginToolsupportsstructurallearning,parameterestimation,andadaptationofparametersinBayesiannetworks.TheperformanceoftheHuginToolisillustratedusingreal-worldBayesiannetworks,commonlyusedexamplesfromtheliter
3、ature,andrandomlygeneratedBayesiannetworks.1IntroductionProbabilisticgraphicalmodelssuchasBayesiannetworks[9,3]areefficientmod-elsfor(automated)reasoningunderuncertainty.ABayesiannetworkcanbeusedasanefficienttoolforknowledgerepresentationandinference.Unfortu-nately,theconstructionofaBayesian
4、networkcanbeaquitelaborintensivetasktoperform.Forthisreason,automatedconstructionofBayesiannetworkshaveinrecentyearsreceivedalotofattention.Thisattentionhasfocusedontheautomatedconstructionofmodelsfromacombinationofdataanddomainex-pertknowledge.Inthispaper,weconsiderthemodelconstructiont
5、askasataskoffusingobservationaldataanddomainexpertknowledge.Throughautomatedconstruction,Bayesiannetworkscanbeusedasefficienttoolsforknowledgediscoveryanddatamining[5].TheHuginTool[1,6]isageneralpurposetoolforprobabilisticgraphicalmodelssuchasBayesiannetworksandinfluencediagrams.Inthispaper
6、,wede-scribetheknowledgediscoveryfunctionalityoftheHuginToolrelatedto(auto-mated)constructionofBayesiannetworksthroughlearning.Thatis,wedescribethecapabilitiesoftheHuginToolforlearningthestructureandparametersofaBayesiannetwork.In[6]arecentsurveyofthegeneralfunctionalityoftheHuginToolisg
7、iven.ThepresentpaperextendsanddetailsthedescriptionofthelearningfunctionalityoftheHuginToolgivenin[6].2PreliminariesandNotationABayesiannetworkN=(G=(V,E),P)consistsofanacyclic,directedgraph(DAG)GandasetofprobabilitydistributionsP.EachnodeX∈VrepresentsaT.D.NielsenandN.L.Zh
