Agents Learning in a Three-Valued Logical Setting

AgentsLearninginaThree-ValuedLogicalSettingEvelinaLammaFabrizioRiguzziDEIS,UniversitadiBologna,VialeRisorgimento240136Bologna,Italy,felamma,friguzzig@deis.unibo .itLusMonizPereiraCentrodeIntelig^enciaArti cial(CENTRIA),DepartamendeInformatica,UniversidadeNovadeLisb oa,2825MontedaCaparica,Portugallmp@di.fct.unl.ptAbstractWeshowthattheadoptionofathree-valuedsettingforinductiveconceptlearningisparticularlyusefullearninginsingleandmultipleagentsystems.Distinguishingb etweenwhatistrue,falseandwhatisunknowncanb eusefulinsituationswheredecisionshavetob etakenonthebasisofscarceinfor-mation.Suchsituationo ccurs,forexample,whenanagentincrementallygathersinformationfromthesur-roundingworldandhastoselectitsownactionsonthebasisofsuchacquiredknowledge.Inathree-valuedsetting,welearnde nitionforb oththetargetconceptanditsopp osite,consider-ingp ositiveandnegativexamplesasinstancesoftwodisjointclasses.Tothispurp ose,weadoptExtendedLogicPrograms(ELP)underaWell-FoundedSeman-ticswithexplicitnegation(WFSX)astherepresenta-tionformalismforlearning.StandardInductiveLogicProgrammingtechniquesarethenemployedtolearntheconceptanditsopp osite.Thelearntde nitionsofthep ositiveandnegativconceptsmayoverlap,b othwhenlearningconictingrulesforapredicateanditsexplicitnegationbysin-gleagentorwhencombiningtheknowledgelearnedbymultipleagents.Inthepap er,wehandleissueofstrategiccombinationofp ossiblycontradictorylearntde nitions.1Intro ductionMostworkoninductiveconceptlearningconsidersatwo-valuedsetting.Insuchasetting,whatisnoten-tailedbythelearnedtheoryisconsideredfalse,onbasisoftheClosedWorldAssumption(CWA)[25].However,inpractice,itismoreoftenthecasethatearecon dentab outthetruthorfalsity ofonlyalimitednumb eroffacts,andarenotabletodrawanyconclu-sionab outtheremainingones,b ecauseavailableinformationisto oscarce.Likeithasb eenp ointedoutin[9,8],thisistypicallythecaseofanautonomousagentthat,inanincrementalway,gathersinforma-tionfromitssurroundingworld.Suchanagentneedstodistinguishb etweenwhatistrue,falseandwhatisunknown,andthereforeneedstolearnwithinaricherthree-valuedsetting.Forthispurp ose,weadopttheclassofExtendedLogicPrgrams(ELP,forshort,inthesequel)asrepresentationlanguageforlearninginathree-valuedsetting.ELPcontaintwokindsofnegation:defaultnegationplusasecondformofnegation,calledex-plicit,whosecombinationhasb eenrecognizedasaveryexpressivemeansofknowledgerepresentation.TheadoptionofELPallowsonetodealdirectlyinthelanguagewithincompleteknowledge,exceptionsthroughdefaultnegation,aswellwithtrulynega-tiveinformationthroughexplicitnegation[22,34].Forinstance,in[3,5,10718]itisshownhowELPareapplicabletosuchdiversedomainsofknowledgerepresentationasconcepthierarchies,reasoningab outactions,b eliefrevision,counterfactuals,diagnosis,up-datesanddebugging.Inthiswork,e rstdiscussvariousapproachesandstrategiesthatcanb eadoptedinInductiveLogicPro-gramming(ILP,henceforth)forlearningwithathree-valuedsettingsbyasingleagent.Then,wediscusshowtheknowledgelearnedbyseparateagentscanb ecombinedtoobtainacommonknowledgebase.Asin[14,13],thelearningpro cessasingleagentstartsfromasetofp ositiveandnegativexamplesplussomebackgroundknowledgeintheformofanex-1

[1]  José Júlio Alferes,et al.  Well Founded Semantics for Logic Programs with Explicit Negation , 1992, ECAI.

[2]  Michael Gelfond,et al.  Logic Programs with Classical Negation , 1990, ICLP.

[3]  Luís Moniz Pereira,et al.  Generalizing Updates: From Models to Programs , 1997, LPKR.

[4]  L. D. Raedt.  Interactive theory revision: an inductive logic programming approach , 1992 .

[5]  Stan Matwin,et al.  Sub-unification: A Tool for Efficient Induction of Recursive Programs , 1992, ML.

[6]  Ofer Arieli,et al.  Paraconsistent Semantics for Extended Logic Programs , 2002, IC-AI.

[7]  Stephen Muggleton,et al.  Inverse entailment and progol , 1995, New Generation Computing.

[8]  Evelina Lamma,et al.  Integrating Induction and Abduction in Logic Programming , 1999, Inf. Sci..

[9]  Stephen Muggleton,et al.  Efficient Induction of Logic Programs , 1990, ALT.

[10]  José Júlio Alferes,et al.  SLX - A Top-down Derivation Procedure for Programs with Explicit Negation , 1994, ILPS.

[11]  Luís Moniz Pereira,et al.  Strategies in Combined Learning via Logic Programs , 2004, Machine Learning.

[12]  Aiko M. Hormann,et al.  Programs for Machine Learning. Part I , 1962, Inf. Control..

[13]  Saso Dzeroski,et al.  Inductive Logic Programming: Techniques and Applications , 1993 .

[14]  Kenneth A. Ross,et al.  The well-founded semantics for general logic programs , 1991, JACM.

[15]  Luc De Raedt,et al.  On Negation and Three-Valued Logic in Interactive Concept-Learning , 1990, ECAI.

[16]  Luís Moniz Pereira,et al.  Prolegomena to Logic Programming for Non-monotonic Reasoning , 1996, NMELP.

[17]  Gordon Plotkin,et al.  A Note on Inductive Generalization , 2008 .

[18]  Luís Moniz Pereira,et al.  Abduction over 3-Valued Extended Logic Programs , 1995, International Conference on Logic Programming and Non-Monotonic Reasoning.

[19]  Stephen Muggleton,et al.  Machine Invention of First Order Predicates by Inverting Resolution , 1988, ML.

[20]  José Júlio Alferes,et al.  Reasoning with Logic Programming , 1996, Lecture Notes in Computer Science.

[21]  José Júlio Alferes,et al.  Dynamic Logic Programming , 1998, APPIA-GULP-PRODE.

[22]  Chitta Baral,et al.  Logic Programming and Knowledge Representation , 1994, J. Log. Program..

[23]  Jürgen Dix,et al.  A Classification Theory of Semantics of Normal Logic Programs: I. Strong Properties , 1995, Fundam. Informaticae.

[24]  Katsumi Inoue,et al.  Learning Abductive and Nonmonotonic Logic Programs , 2000 .

[25]  José Júlio Alferes,et al.  'Classical' Negation in Nonmonotonic Reasoning and Logic Programming , 1998, Journal of Automated Reasoning.

[26]  Ryszard S. Michalski,et al.  Discovering Classification Rules Using variable-Valued Logic System VL1 , 1973, IJCAI.

[27]  Stephen Muggleton,et al.  Non-monotonic learning , 1991 .