ebook img

an anytime algorithm for hindsight knowledge-base optimization PDF

328 Pages·2006·1.31 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview an anytime algorithm for hindsight knowledge-base optimization

D -D R : EPENDENCY IRECTED ECONSIDERATION A A A N NYTIME LGORITHM FOR H K -B O INDSIGHT NOWLEDGE ASE PTIMIZATION by Frances Laird Johnson January 11, 2006 A dissertation submitted to the Faculty of the Graduate School of the State University of New York at Buffalo in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Computer Science and Engineering Copyrightby FrancesLairdJohnson 2006 ii Acknowledgements AsIthinkofthosewhohavesupportedandinspiredmeovertheyears,Iamoverwhelmedbytheirnumber. Space limitations and a somewhat legendary scatterbrain memory assure that I will be revising this list for yearstocome,soIapologizeinadvancetothosewhoarenotmentionedbelow. Withoutquestion,thisresearchisatestimonytotheguidance,patienceandperseveranceofmyadvisor, Stuart Shapiro. He is a very rare and unique researcher—a versatile and ever publishing scholar who also loves working with implementations and is an excellent programmer. This made Stu the perfect advisor for my dissertation. I am especially grateful to him for holding the reins tight enough to guide me to this destination yet loose enough to choose my own path. Stu’s ethics and professionalism meet the highest standards,andIonlyhopethatIcanfollowhisexample. WilliamRapaportandCarlAlphoncejoinedStutoformmycommittee. Billhasalwaysbeenareliable source of knowledge, grammar, vocabulary, support, and humor. There were many times when Stu and I wouldsettlesomepointofconfusionwith“askBill,he’llknow.”Carl’sinputwasanexcellentcomplement to Bill and Stu’s, and this combination improved the readability of my work. My thanks also go to Erwin Segal whose feedback from a background outside that of computer science was very helpful; relocation forcedhimtostepdownfrommycommittee. There is no way that I can sufficiently thank Sven Ove Hansson—who agreed to receive a 300-page dissertation duringthe winter holidays(having never metme or, as far asI know, read my work), andwho returnedafavorablereportwithrevisionsuggestionswithintwoweeks! Hisreportwillbeapermanentpart iii ofmycollectionoftreasuredcorrespondence. IamgratefulforthefriendshipandcodingskillsofMichaelKandefer,whoseassistancenotonlyhelped metorefinemyimplementation,butitfreedmetocompletethisdissertationwithinthetimeframerequired. Ken Regan, Helene Kershner and Jodi Reiner provided a measure of support that was essential to my successful survival and the completion of this work. Together, they gave me (and the graduate school) confidence that I would finish. I am especially grateful to Helene for her consistent support and advice throughoutmyentiretimeatUB. Ihadanespeciallybadtimein2005-2006,physically,andtraditionalmedicinefailedme. SteveGrande, Joe Cerceo, and John Zilliox kept me out of a wheelchair and, possibly, a hospice bed. This dissertation wouldhaveneverbeencompletedwithouttheirhealingknowledge. TheinterestandfeedbackofthemembersoftheSNePSResearchGroupwasinvaluableinthedevelop- mentofmythesis. ThosenotalreadymentionedincludeJosephineAnstey,DebraBurhans,BharatBushan, Alistair Campbell, Albert Goldfain, Haythem Ismail, Jean-Pierre Koenig, Joa˜o Martins, David Pierce, and JohnSantore. IespeciallycommendJoa˜oforhisinitialcontributiontothedevelopmentoftheSNePSbelief revisionsystem. And,IwillforeverbegratefulthatIhadHaythemandJohnasmylong-termoffice-mates. Theywerealwaysthereforme,andtheyaretrulyexemplarypeople. Other UB professors and staff and fellow students (past and present) who have helped, encouraged and advised me during this journey are: Eloise Benzel, Jan Chomicki, Chris Egert, Sally Elder, Margaret Evans,JoannGlinski,ElleHeffner,BharatJayaraman,KathyPohlman,BinaRamamurthy,KennethSmith, MatthewStock,JayneeStraw,MarianneSullivan,LenTalmy,PallaviTambay,LynneTerrana,andDeborah Walters. Thank you all. Special thanks go to James Llinas who jump-started my belief revision research through funding from a group headed by Joseph Karakowski, who later supported this work (both profes- sionallyandfinancially)forseveralyears. iv InadditiontothoseatUB,manyresearchersoutsidetheUBcommunityhaveprovidedvaluablefeedback and insight. I am especially grateful to Samir Chopra, Doug Lenat, Thomas Meyer, Leora Morgenstern, RenataWassermann,andMary-AnneWilliams. To the teachers from my past, without the foundation you laid, I would not be the person I am, today. From my earlier years, I especially thank Bill ”Doc” Chase, Debbie Coates, Mary Louise Kelley, Bob- bette Mason, and Betty Richardson. And to Linda Brinkerhoff, Jeannette Neal, and Alison Girod, of Erie CommunityCollege,yourexcellenceasteachersinspiredmetocontinueonatthePh.D.level. My full support structure of friends and family is too large to include here, but I must mention a select few. ToHoward,Bud,Rose,Letty,Margie,Ted,and,especially,Dad—Iloveyouandmissyouall. Mirjana, you have been such a help and inspiration to me. Wick and Donna, your advice and encouragement and constant support for both me and Tom has been invaluable. Carol, your unconditional faith in me is an ever-present source of strength. You are my bestest friend in the whole wide world. Clint and Tillie, you havelightenedmyloadinsomanyways;thankyouforyoursupport. Mom,youweremyfirstteacherandI stillhavesomuchtolearnfromyou;IhopeIwillcontinuetobecomemorelikeyoueveryday. Tomygirls, ElizabethandKatherine: youaremyfavoriteteachers,andIloveyoumorethantonguecantell. And,Tom, youaremyeverything;themagnitudeofyourfaithinmeandloveformeamazesmeeveryday. You’rethe best. v Formyfamily vi Contents Acknowledgements iii Abstract xix 1 Introduction 1 1.1 BeliefChange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 NotationandTerminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3.1 ImplementingKR&RRequiresaDeparturefromtheIdeal . . . . . . . . . . . . . . 8 1.3.2 SpecificObservationsMotivatingthisResearch . . . . . . . . . . . . . . . . . . . . 9 1.4 ResearchContributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.5 DissertationOutline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2 BackgroundandRelatedWork 15 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.1.1 PurposeofThisChapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.1.2 BeliefChangevs. BeliefRevision . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.1.3 MergingFactionsofBeliefChange . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.1.4 Agents: Idealvs. Resource-Bounded . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.2 AGM/BaseResearch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.2.1 FoundationsandCoherenceApproaches . . . . . . . . . . . . . . . . . . . . . . . . 19 vii 2.2.2 IntegrityConstraintsforBeliefChangeOperationsonaKnowledgeSpace . . . . . . 20 2.2.3 AGMOperations: Expansion,Contraction,andRevision . . . . . . . . . . . . . . . 21 2.2.4 Hansson’sPostulatesforExpansion,Contraction,andRevision,ofaBeliefBase . . 25 2.2.5 Hansson’sBeliefBasePostulatesforBeliefChangeOperationsUsingKernels . . . 28 2.2.6 Wassermann’sResource-BoundedTheories . . . . . . . . . . . . . . . . . . . . . . 32 2.2.7 BeliefLiberation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.3 Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.4 TMSResearch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.4.2 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.4.3 JTMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.4.4 ATMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 2.4.5 LTMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.4.6 ITMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.4.7 GeneralProsandCons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 2.4.8 WhatdocurrentTMSsystemsdo? . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.4.9 TMSConclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2.5 DifferentWaystoOrdertheBaseBeliefs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2.6 SomeKR&RImplementedSystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 2.6.1 SATEN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.6.2 Cyc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.6.3 SNePS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 2.7 TheNeedforaDOBSFormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 viii 2.7.1 SomeLimitationsoftheAboveSystems . . . . . . . . . . . . . . . . . . . . . . . . 59 2.7.2 NeedsofaDOBS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3 Reconsideration 63 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.1.1 MotivationsandAssumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.1.2 ABriefExampleShowingOperationOrderSide-Effects . . . . . . . . . . . . . . . 67 3.1.3 Jury-roomReasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.1.4 AnticipatingReconsideration—OtherBeliefChangeOperationsthatAppearSimilar 69 3.2 Reconsideration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.2.1 TheKnowledgeStateforReconsideration . . . . . . . . . . . . . . . . . . . . . . . 70 3.2.2 DefininganOptimalBaseandanOptimalKnowledgeState . . . . . . . . . . . . . 73 3.2.3 OperationsonaKnowledgeState . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.2.4 ComparingReconsiderationToBeliefLiberation . . . . . . . . . . . . . . . . . . . 86 3.3 ImprovingRecoveryforBeliefBases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 3.3.1 ReconsiderationProvidesanImprovedRecoveryAspecttoBaseBeliefChange . . . 88 3.3.2 ComparingRecovery-likeFormulations . . . . . . . . . . . . . . . . . . . . . . . . 91 3.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 3.4 BeliefChangeIssuesforNon-linearOrders . . . . . . . . . . . . . . . . . . . . . . . . . . 98 3.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 3.4.2 ASingleNogoodWithaNon-SingletonSetofMinimalBeliefs . . . . . . . . . . . 98 3.4.3 MultipleNogoodsHavingNon-SingletonSetsofMinimalBeliefs . . . . . . . . . . 100 3.4.4 Dealingwithnon-linearissues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.5 BeliefChangeinaTotalPre-order: SixStrategiesUsedbySATEN . . . . . . . . . . . . . . 101 ix 3.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.5.2 StandardAdjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 3.5.3 Maxi-Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 3.5.4 HybridAdjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 3.5.5 GlobalAdjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 3.5.6 LinearAdjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 3.5.7 QuickAdjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 3.5.8 PreparingtoTestReconsideration . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 3.6 ReconsiderationonaTotalPre-orderUsingSATENStrategies . . . . . . . . . . . . . . . . 111 3.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 3.6.2 PerformingaSecondRevision: (((B+¬a)!+a)!) . . . . . . . . . . . . . . . . . . 111 3.6.3 Reconsideration: (((B+¬a)+a)!) . . . . . . . . . . . . . . . . . . . . . . . . . . 112 3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4 Dependency-DirectedReconsideration 119 4.1 Introduction,MotivationandAssumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 4.2 Dependency-DirectedReconsideration(DDR) . . . . . . . . . . . . . . . . . . . . . . . . . 121 4.2.1 TheDDRProcess . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 4.2.2 DefiningaKnowledgeState . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 4.2.3 DefiningaNAND-set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.2.4 DefiningComplete-I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.2.5 DefiningCulprits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 4.2.6 DefinitionsRelevanttoKnowledgeStateConsistencyandOptimality . . . . . . . . 125 4.2.7 DDRBeliefCredibilityandBaseCredibility,Ordering,andOptimality . . . . . . . 127 x

Description:
There is no way that I can sufficiently thank Sven Ove Hansson—who agreed to . 2.2.5 Hansson's Belief Base Postulates for Belief Change Operations Using Kernels . The propositional language L is closed under the . In this dissertation, the term “belief” does not refer to a weakened assertion
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.