Slide 1Take-Away MessageOutlineSlide 4MotivationReal-World Data (Dramatically Simplified)Slide 7Slide 8A (very) Brief HistorySeveral SRL formalisms => Endless Possibilities(Propositional) Logic Program – 1-slide IntroLogic Programming (LP)Model Theoretic ViewProbabilities on Possible worldsProof TheoreticProbabilities on ProofsSlide 17Slide 18First-Order/Relational Logic + Probability = PLMModel-Theoretic ApproachesProbabilistic Relational Models – Getoor et al.Relational SchemaProbabilistic Relational ModelsSlide 24Bayesian Logic Programs (BLPs)Bayesian Logic Programs (BLPs) – Kersting & De RaedtProof theoretic Probabilistic Logic MethodsProbabilistic Proofs -PRISMProbabilistic Proofs -PRISMPRISMProbabilistic Proofs – Stochastic Logic Programs (SLPs)Slide 32Slide 33Slide 34Undirected Probabilistic Logic ModelsMarkov Logic Networks (Richardson & Domingos)Example: Friends & SmokersPlethora of ApproachesMultiple Parents ProblemMultiple Parents for “population”Solution 1: Aggregators – PRM, RDN, PRL etcSolution 2: Combining Rules – BLP, RBN,LBN etcSlide 43LearningParameter EstimationParameter Estimation – Model TheoreticParameter Estimation – Proof TheoreticSlide 48Slide 49Slide 50Sriraam NatarajanIntroduction to Probabilistic Logical ModelsIntroduction to Probabilistic Logical ModelsSlides based on tutorials by Kristian Kersting, James Cussens, Lise Getoor & Pedro DomingosTake-Away Message Learn from rich, highly structured dataProgress to date•Burgeoning research area•“Close enough” to goal•Easy-to-use open-source software available•Lots of Challenges/Problems in the futureIntroductionProbabilistic Logic ModelsDirected vs Undirected ModelsLearningConclusionOutlineIntroductionProbabilistic Logic ModelsDirected vs Undirected ModelsLearningConclusionMotivationMost learners assume i.i.d. data(independent and identically distributed)–One type of object–Objects have no relation to each otherTo predict if the image is “eclipse”Real-World Data (Dramatically Simplified)PatientID Gender Birthdate P1 M 3/22/63 PatientID Date Physician Symptoms Diagnosis P1 1/1/01 Smith palpitations hypoglycemic P1 2/1/03 Jones fever, aches influenzaPatientID Date Lab Test Result P1 1/1/01 blood glucose 42 P1 1/9/01 blood glucose 45PatientID SNP1 SNP2 … SNP500K P1 AA AB BB P2 AB BB AAPatientID Date Prescribed Date Filled Physician Medication Dose Duration P1 5/17/98 5/18/98 Jones prilosec 10mg 3 monthsNon- i.i.dMulti-RelationalSolution: First-Order Logic / Relational DatabasesShared ParametersThe World is inherently UncertainGraphical Models (here e.g. a Bayesian network) - Model uncertainty explicitly by representing the joint distributionFever AcheInfluenzaRandom VariablesDirect InfluencesPropositional Model!Logic + Probability = Probabilistic Logic aka Statistical Relational Learning ModelsLogicProbabilitiesAdd ProbabilitiesAdd RelationsStatistical Relational Learning (SRL)Uncertainty in SRL Models is captured by probabilities, weights or potential functionsA (very) Brief HistoryProbabilistic Logic term coined by Nilsson in 1986Considered the “probabilistic entailment” i.e., the probabilities of all sentences between 0 and 1Earlier work by (Halpern, Bacchus and others) focused on the representation and not learningNiem and Haddawy (1995) – one of the earlier approachesLate 90’s: OOBN, PRM, PRISM, SLP etc‘00- ‘05 : Plethora of approaches (representation)Learning methods (since ‘01)Recent thrust – Inference (Lifted Inference techniques)Several SRL formalisms => Endless PossibilitiesWeb data (web)Biological data (bio) Social Network Analysis (soc)Bibliographic data (cite)Epidimiological data (epi)Communication data (comm)Customer networks (cust)Collaborative filtering problems (cf)Trust networks (trust)Reinforcement LearningNatural Language ProcessingSAT…(Propositional) Logic Program – 1-slide IntroClauses: IF burglary and earthquake are true THEN alarm is trueburglary.earthquake.alarm :- burglary, earthquake.marycalls :- alarm.johncalls :- alarm.Herbrand Base (HB) = all atoms in the program burglary, earthquake, alarm, marycalls, johncallsProgramatombodyheadLogic Programming (LP)2 views:1) Model-Theoretic2) Proof-TheoreticModel Theoretic ViewLogic Program restricts the set of possible worldsFive propositions – Herbrand baseSpecifies the set of possible worlds An interpretation is a model of a clause C If the body of C holds then the head holds, too.burglary.earthquake.alarm :- burglary, earthquake.marycalls :- alarm.johncalls :- alarm.burglaryearthquakealarmmarycalls johncalls truefalsetruefalsetruefalsetruefalsetruefalseProbabilities on Possible worldsSpecifies a joint distribution P(X1,…,Xn) over a fixed, finite set {X1,…,Xn}Each random variable takes a value from respective domainDefines a probability distribution over all possible interpretationsburglaryearthquakealarmmarycalls johncalls truefalsetruefalsetruefalsetruefalsetruefalseProof Theoreticburglary.earthquake.alarm :- burglary, earthquake.marycalls :- alarm.johncalls :- alarm.:- alarm.:- burglary, earthquake.:- earthquake.{}A logic program can be used to prove some goals that are entailed by programGoal :- johncallsProbabilities on ProofsStochastic grammarsEach time a rule is applied in a proof, the probability of the rule is multiplied with the overall probabilityUseful in NLP – most likely parse tree or the total probability that a particular sentence is derivedUse SLD trees for resolution1.0 : S NP, VP 1/3 : NP i 1/3 : NP Det, N 1/3 : NP NP, PP....Full Clausal LogicFull Clausal LogicRelational Clausal LogicRelational Clausal Logic Propositional Clausal LogicIntroductionProbabilistic Logic ModelsDirected vs Undirected ModelsLearningConclusionFirst-Order/Relational Logic + Probability = PLMModel-Theoretic vs. Proof-TheoreticDirected vs. UndirectedAggregators vs. Combining RulesModel-Theoretic ApproachesProbabilistic Relational Models – Getoor et al.Combine advantages of relational logic & Bayesian networks: –natural domain modeling: objects, properties,
View Full Document