您好,欢迎访问三七文档
当前位置:首页 > 商业/管理/HR > 管理学资料 > Local and global relational consistency
LocalandGlobalRelationalConsistencyRinaDechter1andPetervanBeek21DepartmentofInformationandComputerScienceUniversityofCalifornia,IrvineIrvine,California,USA92717dechter@ics.uci.edu2DepartmentofComputingScienceUniversityofAlbertaEdmonton,Alberta,CanadaT6G2H1vanbeek@cs.ualberta.caAbstract.Localconsistencyhasproventobeanimportantconceptinthetheoryandpracticeofconstraintnetworks.Inthispaper,wepresentanewde nitionoflocalconsistency,calledrelationalconsistency.Thenewde nitionisrelation-based,incontrastwiththepreviousde nitionoflocalconsistency,whichwecharacterizeasvariable-based.Weshowtheconceptualpowerofthenewde nitionbyshowinghowituni esknowneliminationoperatorssuchasresolutionintheoremproving,joinsinrela-tionaldatabases,andvariableeliminationforsolvinglinearinequalities.Algorithmsforenforcingvariouslevelsofrelationalconsistencyarein-troducedandanalyzed.Wealsoshowtheusefulnessofthenewde nitionincharacterizingrelationshipsbetweenpropertiesofconstraintnetwork-sandtheleveloflocalconsistencyneededtoensureglobalconsistency.1IntroductionConstraintnetworksareasimplerepresentationandreasoningframework.Aproblemisrepresentedasasetofvariables,adomainofvaluesforeachvariable,andasetofconstraintsbetweenthevariables.Acentralreasoningtaskisthento ndaninstantiationofthevariablesthatsatis estheconstraints.Ingeneral,whatmakesconstraintnetworkshardtosolveisthattheycancontainmanylocalinconsistencies.Alocalinconsistencyisaconsistentinstan-tiationofk 1ofthevariablesthatcannotbeextendedtoakthvariableandsocannotbepartofanyglobalsolution.Ifweareusingabacktrackingsearchto ndasolution,suchaninconsistencycanleadtoadeadendinthesearch.Thisinsighthasledtothede nitionofconditionsthatcharacterizetheleveloflocalconsistencyofanetwork[29,33]andtothedevelopmentofalgorithmsforenforcinglocalconsistencyconditionsbyremovinglocalinconsistencies(e.g.,[33,29,19,12,6]).Inthispaper,wepresentanewde nitionoflocalconsistencycalledrelationalconsistency3.Thevirtueofthenewde nitionoflocalconsistencyisthat, rstly,itremovestheneedforreferencingthearityoftheconstraintswhendiscussing3Apreliminaryversionofthisde nitionappearsin[37].relationshipsbetweenthepropertiesoftheconstraintsandlocalconsistency.Secondly,itisoperational,thusgeneralizingtheconceptofthecompositionoperationde nedforbinaryconstraints,andcanbeincorporatednaturallyinalgorithmsforenforcingdesiredlevelsofrelationalconsistency.Thirdly,ituni- esknownoperatorssuchasresolutionintheoremproving,joinsinrelationaldatabases,andvariableeliminationforsolvingequationsandinequalities,thusallowingtheformulationofaneliminationalgorithmthatgeneralizesalgorithmsappearingineachoftheseareas.Finally,itallowsidentifyingthoseformalismsforwhichconsistencycanbedecidedbyenforcingaboundedlevelofrelationalconsistency,likepropositionaldatabases,linearequalitiesandinequalities,andcrosswordpuzzlesfromgeneraldatabasesrequiringhigherlevelsofrelationalconsistency.Wealsodemonstratetheusefulnessofthenewde nitionincharac-terizingrelationshipsbetweenvariouspropertiesofconstraintnetworks|domainsizeandacyclicity|andtheleveloflocalconsistencyneededtoensureglobalconsistency.Followingde nitionandpreliminaries(section2),relationallocalconsistencyisde nedandalgorithmsforenforcingsuchconditionsareintroduced(section3).Section4showsthatthealgorithmsunifyalgorithmsappearinginpropositionaldatabasesandlinearinequalities.Finally,section5describesnewassociationsbetweenconstraintpropertiesandrelationallocalconsistencyneededforglobalconsistency.Discussionandconclusionsaregiveninsections6and7respectively.2De nitionsandPreliminariesDe nition1(constraintnetwork).4AconstraintnetworkRisasetofnvariablesX=fx1;...;xng,adomainDiofpossiblevaluesforeachvariablexi,1 i n,andasetoftrelationsRS1,...,RSt,whereSi X,1 i n.AconstraintorrelationRSoverasetofvariablesS=fx1;...;xrgisasubsetoftheproductoftheirdomains(i.e.,RS D1 Dr).ThesetofsubsetsfS1;...;Stgonwhichconstraintsarespeci ediscalledtheschemeofR.Abinaryconstraintnetworkisthespecialcasewhereallconstraintsareoverpairsofvariables.Aconstraintgraphassociateseachvariablewithanodeandconnectsanytwonodeswhosevariablesappearinthesameconstraint.De nition2(solutiontoaconstraintnetwork).Aninstantiationofthevari-ablesinX,denotedXI,isann-tuple(a1;...;an),representinganassignmentofai2Ditoxi,1 i n.Aconsistentinstantiationofanetworkisaninstan-tiationofthevariablessuchthattheconstraintsbetweenvariablesaresatis ed.Aconsistentinstantiationisalsocalledasolution.Theorderofthevariablesconstrainedbyarelationisnotimportant;thatis,wefollowtheset-of-mappingsformulationofrelations(see[34]).Thenotionofaconsistentinstantiationofasubsetofthevariablescanbede nedinseveral4Notethatallthede nitionsandalgorithmsareapplicabletorelationswithoutthe nitenessassumption;apointwemakeexplicitinsection4.22ways.Weusethefollowingde nition:aninstantiationisconsistentifitsatis esalloftheconstraintsthathavenouninstantiatedvariables.De nition3(consistentinstantiationofsubsetsofvariables).LetYandSbesetsofvariables,andletYIbeaninstantiationofthevariablesinY.Wede-notebyYI
本文标题:Local and global relational consistency
链接地址:https://www.777doc.com/doc-3380627 .html