uncertainty representation and reasoning with mebn/pr-owl

18
Slide 1 of 18 Representation and Reasoning with MEBN/PR-OWL Kathryn Blackmond Laskey Paulo C. G. da Costa The Volgenau School of Information Technology and Engineering George Mason University - Fairfax, VA [klaskey, pcosta]@gmu.edu

Upload: jasia

Post on 19-Mar-2016

32 views

Category:

Documents


0 download

DESCRIPTION

Uncertainty Representation and Reasoning with MEBN/PR-OWL. Kathryn Blackmond Laskey Paulo C. G. da Costa The Volgenau School of Information Technology and Engineering George Mason University - Fairfax, VA [klaskey, pcosta]@gmu.edu. Uncertainty and Ambiguity are Ubiquitous. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 1 of 18

Uncertainty Representation and Reasoning with MEBN/PR-OWL

Kathryn Blackmond LaskeyPaulo C. G. da Costa

The Volgenau School of Information Technology and EngineeringGeorge Mason University - Fairfax, VA

[klaskey, pcosta]@gmu.edu

Page 2: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 2 of 18

Uncertainty and Ambiguity are Ubiquitous

Page 3: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 3 of 18

Semantic Awareness in an Uncertain World

Ontologies formalize our knowledge about entities and relationships in the world

Many relationships are intrinsically uncertain Traditional ontology formalisms lack built-in means

for handling uncertainty Without a means of expressing uncertainty we are

unable to say much of what we know

Methodologies and tools are needed for principled handling of uncertainty

in semantically aware systems

Page 4: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 4 of 18

Is Probability Ontological or Epistemic?

Intrinsically probabilistic phenomena may exist in Nature

There is an urgent practical need for sound and principled representation of uncertainties associated with our knowledge

Today’s existential phenomenon is tomorrow’s superseded theory

Page 5: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 5 of 18

Why Bayes? Requirement: reason in the presence of uncertainty about…

• Input data• Existence of relationships among entities• Strength of relationships• Constraints governing relationships

Solution: Bayesian inference• Combine expert knowledge with statistical data• Represent cause and effect relationships• Learn from observations• Prevent over-fitting• Clear and understandable semantics• Logically coherent

Page 6: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 6 of 18

Bayesian Network Parsimonious specification for joint

probability distribution over many random variables

• Graph encodes dependence relationships

• Local distributions encode numerical probability information

• Implicitly specifies full joint distribution

Computational architecture for evidential reasoning

• Condition on evidence• Compute updated beliefs on

unobserved variables• Efficient local computations• Bi-directional reasoning

Are BNs a suitable formal basis for probabilistic ontology?

Page 7: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 7 of 18

The Trouble with BNs

Traditional BNs are insufficiently Traditional BNs are insufficiently expressive for complex problemsexpressive for complex problemsHow many entities? What are their types?What are their features? How are they related to each other?How do they change over time?

Page 8: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 8 of 18

MEBN to the Rescue!

MEBN can express:Attribute value uncertaintyNumber uncertaintyType uncertaintyReference uncertainty Structure uncertaintyRepeated structure

RecursionExistence uncertaintyParameter uncertaintyStructure uncertainty Quantifiers

Page 9: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 9 of 18

MEBN: A First-Order Bayesian Logic

Represents knowledge as parameterized fragments of Bayesian networks

Expresses repeated structure Represents probability distribution on interpretations

of associated first-order theory Expressive enough to express anything that can be

said in FOL Suitable logical basis for probabilistic ontology

Page 10: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 10 of 18

MEBN Theory

Page 11: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 11 of 18

Situation Specific Bayesian Network

Own ship, 4 other starships, 1 zone, 4 reports, 2 time steps

Ordinary Bayesian network constructed to process probabilistic query on a MEBN Theory

Page 12: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 12 of 18

PR-OWL: A Language for Probabilistic Ontologies

Upper OWL Ontology Represents MEBN Theories

Page 13: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 13 of 18

MEBN / PR-OWL

Page 14: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 14 of 18

Logical Reasoning

Page 15: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 15 of 18

Logical *and* Plausible Reasoning

Page 16: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 16 of 18

MEBN/PR-OWL Probabilistic Ontologies

Allow both probabilistic and deterministic reasoning The “probabilistic part” is a complete or partial

MEBN theory Different people will build different MEBN theories

of their domains. MEBN logic is expressive enough to provide logical

basis for semantic integration.

Page 17: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 17 of 18

Probabilistic Semantic Mapping

Costa, P., Laskey, K.B. and Laskey, K.J., Probabilistic Ontologies for Efficient Resource Sharing in Semantic Web Services, Workshop on Uncertainty in the Semantic Web, International Semantic Web Conference, November 2006.

• A probabilistic ontology augments a standard ontology with a representation of uncertainty

• A mapping ontology represents mapping of terms between domain ontologies

Page 18: Uncertainty Representation and Reasoning with MEBN/PR-OWL

Slide 18 of 18

THANKS!!!