marko逻辑网的推理算法研究.pdfVIP

UlY2346885IlllIIIlIIIlIllllll MarkovMarkovMarkovMarkovmeetsmachinebeenbebasisbasedareaaanaaareasasasareonhashashascorenetworksnetworksnetworksoffirst-orderofastotOextracttoforandcombinationestimationthetheforstructuresettime．MarkovsuchtwOisininternationalislineardataMarkovobtainedthesemethodthelikelihoodforthemodelintherelational AbStracmethodLresultsresultcarloffeatureforthisextractionIIIshow’thatline 22．22．333．33．3．13．3．233．3．23．3．43．3．571．11．2 21 ProbabilisticRelationalof1．1 NMNModelsBHLOHHHRMN．RMMTheClassificationRelationalStatisticalof2iii1．2 4 22 6 MonteBaseTab．2-1RandomChain2for2．2 2．3 AMarkovNetworkTab．2-2921．51．11．1 ANetworkMarkovof10 3 X∈NF2一XeNFInNF23．3．13．312 MC—SATWithMTheallfhrofcktofordock∈clausesendendforfor3．3．2satisfied MaxWalkSATunsatisfiedThevarsrandomofcosofofnormaltotoforforsodoclausesdotomax—triestruththenend3．3．214solnIn+--ainini1ifif withwithVvC--aCunsatisfiedThevariablevariableofccosactivefromforforforforfoundsosochosenclausechosendoelseeachendendendend3．3．4Inlowestifinif withVVfDBCunsatisfiledunsatisfiedvfnotvfrandomvariablevariableactive—atomsactive——atomsccosactive——atoms-．--atomsactivatedactive—atomsaddaddactivatedactive——atomsofofoftomaxtotototoforforfromforforfoundsosodoclausesclausesdochosenclausechosendoclausesforfortriestruththenendelseeachendendthenendendendcost÷一sumsatisfiedsolnInlowestininifififinifififif 13．3．5 MarkovMLNsTheTab．3—1Small—scaleofinference MC．SATMaxWmkSATMC．SATMedium．sizedMLNsTab．3．2ErrorO．14105830．009950．009530．009530．009950．5890．214600．174000．015400．004200．007000．010600．195000．143330．010000．007660．010000