欢迎来到天天文库
浏览记录
ID:31998364
大小:785.50 KB
页数:31页
时间:2019-01-30
《人工智能_贝叶斯网络》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、1ArtificialIntelligence:BayesianNetworks2GraphicalModelsIfnoassumptionofindependenceismade,thenanexponentialnumberofparametersmustbeestimatedforsoundprobabilisticinference.Norealisticamountoftrainingdataissufficienttoestimatesomanyparameters.Ifablanketassumptionofconditionalindependenceismade,effi
2、cienttrainingandinferenceispossible,butsuchastrongassumptionisrarelywarranted.Graphicalmodelsusedirectedorundirectedgraphsoverasetofrandomvariablestoexplicitlyspecifyvariabledependenciesandallowforlessrestrictiveindependenceassumptionswhilelimitingthenumberofparametersthatmustbeestimated.BayesianNe
3、tworks:Directedacyclicgraphsthatindicatecausalstructure.MarkovNetworks:Undirectedgraphsthatcapturegeneraldependencies.3BayesianNetworksDirectedAcyclicGraph(DAG)NodesarerandomvariablesEdgesindicatecausalinfluencesBurglaryEarthquakeAlarmJohnCallsMaryCalls4ConditionalProbabilityTablesEachnodehasacondi
4、tionalprobabilitytable(CPT)thatgivestheprobabilityofeachofitsvaluesgiveneverypossiblecombinationofvaluesforitsparents(conditioningcase).Roots(sources)oftheDAGthathavenoparentsaregivenpriorprobabilities.BurglaryEarthquakeAlarmJohnCallsMaryCallsP(B).001P(E).002BEP(A)TT.95TF.94FT.29FF.001AP(M)T.70F.01
5、AP(J)T.90F.055CPTCommentsProbabilityoffalsenotgivensincerowsmustaddto1.Examplerequires10parametersratherthan25–1=31forspecifyingthefulljointdistribution.NumberofparametersintheCPTforanodeisexponentialinthenumberofparents(fan-in).6JointDistributionsforBayesNetsABayesianNetworkimplicitlydefinesajoint
6、distribution.ExampleThereforeaninefficientapproachtoinferenceis:1)Computethejointdistributionusingthisequation.2)Computeanydesiredconditionalprobabilityusingthejointdistribution.7NaïveBayesasaBayesNetNaïveBayesisasimpleBayesNetYX1X2…XnPriorsP(Y)andconditionalsP(Xi
7、Y)forNaïveBayesprovideCPTsforthene
8、twork.8IndependenciesinBayesNetsIfremovingasubsetofnodesSfromthenetworkrendersnodesXiandXjdisconnected,thenXiandXjareindependentgivenS,i.e.P(Xi
9、Xj,S)=P(Xi
10、S)However,thisistoostrictacriteriaforconditionalind
此文档下载收益归作者所有