Date post: | 18-Dec-2015 |
Category: |
Documents |
Upload: | brianna-brown |
View: | 216 times |
Download: | 0 times |
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Lecture 18 of 42
Knowledge Representation Continued: KE,Inheritance, & Representing Events over Time
Discussion: Structure Elicitation, Event Calculus
William H. Hsu
Department of Computing and Information Sciences, KSU
KSOL course page: http://snipurl.com/v9v3
Course web site: http://www.kddresearch.org/Courses/CIS730
Instructor home page: http://www.cis.ksu.edu/~bhsu
Reading for Next Class:
Section 10.4 – 10.9, p. 341 – 362, Russell & Norvig 2nd edition
IM: http://en.wikipedia.org/wiki/Information_management
Event calculus: http://en.wikipedia.org/wiki/Event_calculus
Protégé-OWL tutorials: http://bit.ly/3rM1pB, http://bit.ly/18pMgR
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Lecture Outline
Reading for Next Class: Sections 10.4 – 10.9 (p. 341 – 362), R&N 2e
Last Class: Knowledge Engineering (KE), Protocol Analysis, Fluents Ontology engineering: defining classes/concepts, slots Concept elicitation techniques
Unstructured
Structured
Protocol analysis (“thinking aloud”)
Today: Frames, Semantic Nets, Inheritance; Event & Fluent Calculi Structure elicitation Computational information and knowledge management (CIKM) Representing time, events
Situation calculus
Event calculus
Fluent calculus Brief tutorial: OWL ontologies in Protégé (http://bit.ly/18pMgR)
Coming Week: CIKM, Logical KR Concluded; Classical Planning
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Acknowledgements
© 2004 H. KnublauchTopQuadrant, Inc.(formerly University of Manchester)http://www.knublauch.com
© 2005 M. HauskrechtUniversity of PittsburghCS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos
Milos HauskrechtAssociate Professor of Computer ScienceUniversity of Pittsburgh
Holger KnublauchVice President, TopQuadrantpreviously Research Fellow, Stanford Medical Informatics & Univ. of Manchester
© 2005 N. Noy & S. TuStanford Center for Biomedical Informatics Researchhttp://bit.ly/jwOf3http://bit.ly/2NBeCIhttp://bmir.stanford.edu
Samson TuSenior Research Scientist
BMIR
Natasha NoySenior Research ScientistBMIR
© 2001 G. TecuciGeorge Mason Universityhttp://bit.ly/3tUACW http://lalab.gmu.edu/cs785/http://lac.gmu.edu
Georghe TecuciProfessor of Computer ScienceDirector, Learning Agents CenterGeorge Mason University
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Universe of Decision Problems
Recursive EnumerableLanguages
(RE)
RecursiveLanguages
(REC)
HL
VALIDLVALIDL
LSAT
SATLL L complem.
underclosure
DL
Decision Problems:Review
Co-RE (REC)
LH: Halting problem
LD: Diagonal problem
Semi-decidableduals:α LVALID iff¬α LSAT
C
Undecidabledualsα LVALID
C iff
¬α LSAT
α ⊢RES ?
α
Y
✓ ✗N
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
“Concept” and “Class” are used synonymously
Class: concept in the domain
wines wineries red wines
Collection of elements with similar properties
Instances of classes
Particular glass of California wine
Adapted from slides © 2005 N. Noy & S. TuStanford Center for Biomedical Informatics Researchhttp://bmir.stanford.edu
Concepts/Classes:Review
Middle
level
Top
level
Bottom
level
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Slots in class definition C Describe attributes of instances of C Describe relationships to other instances e.g., each wine will have color, sugar content, producer, etc.
Property constraints (facets): describe/limit possible values for slot
Adapted from slides © 2005 N. Noy & S. TuStanford Center for Biomedical Informatics Researchhttp://bmir.stanford.edu
Slots/Attributes/Relations:Review
Slots & facets for Concept/Class Wine
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Tabs partition different work areas
Buttons and widgetsfor manipulating slots
Area for manipulating the class hierarchy
Protégé – Default Interface:Review
Adapted from slides © 2005 N. Noy & S. TuStanford Center for Biomedical Informatics Researchhttp://bmir.stanford.edu
Downloads, primer, documentation:http://protege.stanford.edu
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Advanced approaches to KB and agent development
Elicitation based on the personal construct theory
A scenario for manual knowledge acquisition
Elicitation of expert’s conception of a domain
Knowledge acquisition for role-limiting methods
Knowledge Engineering:Review
© 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
A knowledge engineer attempts to understand how a subject matter expert reasons and solves problems and then encodes the acquired expertise into the agent's knowledge base.
The expert analyzes the solutions generated by the agent (and often the knowledge base itself) to identify errors, and the knowledge engineer corrects the knowledge base.
KnowledgeEngineer
DomainExpert
Knowledge Base
Inference Engine
Intelligent Agent
ProgrammingDialog
Results
© 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
How Agents Are Built:Review
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Defining problem to solve and system to be built:requirements specification
Choosing or building an agent building tool:Inference engine and representation formalism
Development of the object ontology
Development of problem solving rules or methods
Refinement of the knowledge base
Feedback loops
among all phases
Understanding the expertise domain
Adapted from slide © 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Agent Development Process:Review
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
(based primarily on Gammack, 1987)
1. Concept elicitation: methods(elicit concepts of domain, i.e. agreed-upon vocabulary)
2. Structure elicitation: card-sort method (elicit some structure for concepts)
3. Structure representation (formally represent structure in semantic network)
4. Transformation of representation (transform representation to be used for some desired purpose)
© 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Elicitation Methodology:Review
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
© 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Structure Elicitation:Card-Sort Method
The Card-Sort Method
(elicit the hierarchical organization of the concepts)
• Type the concepts on small individual index cards.
• Ask the expert to group together the related concepts into as many small groups as possible.
• Ask the expert to label each of the groups.
• Ask the expert to combine the groups into slightly larger groups, and to label them.
The result will be a hierarchical organization of the concepts
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Adapted from slide © 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Card-Sort Method:Illustration
Satchwell
Time Switch
Programmer
Thermostat
Set Point
Rotary Control Knob
Gas Control Valve
Solenoid
Electrical System
Electrical Supply
Electrical Contact
Fuse
Pump
Motorized Valve
Electric Time Controls
Thermostat
Gas Control
Electrical Supply
Electrical Components
Mechanical Components
Control Electricity
Part of the hierarchy of concepts from the card-sort method
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Strengths• gives clusters of concepts and hierarchical
organization• splits large domains into manageable sub-areas• easy to do and widely applicable
Weaknesses • incomplete and unguided • strict hierarchy is usually too restrictive
Card-Sort Method:Properties
Adapted from slide © 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Structure Representation [1]:Definition
© 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Represents the acquired concepts into a semantic network and acquires additional structural knowledge:
• Ask the expert to sort the concepts by considering each concept C as a reference, and identifying those related to it.
• Ask the expert to order the concepts related to C along a scale from 0 to 100, marked at the side of a table. The values are read off the scale and entered in a data matrix.
• Generate a network from the matrix, where the nodes are the concepts and the weighted links represent proximities.
• For each pair of concepts identified as related, ask the expert what that relationship is.
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Adapted from slide © 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Domestic Plumbing System
Time Switch
Electrical Supply
Pipe Water Supply
Feedback Loop
Electrical Contact
Flow Header Tank
Water Expansion
Thermostat
Thermal Circuit Heat
Radiator Control Valve
Gravity
Pilot Light Boiler Radiator Air
Gas Control Valve
Primary Circuit
Hot Water Cylinder
Immersion Heater
Main Gas Supply
Motorized Valve
Structure Representation [2]:Illustration
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Adapted from slide © 2001 G. Tecuci, George Mason UniversityCS 785 Knowledge Acquisition and Problem-Solving http://lalab.gmu.edu/cs785/
Structure Representation [3]:Properties
Strengths• gives information on the domain structure in the
form of a network• shows which links are likely to be meaningful• organizes the elicitation of semantic relationships
Weaknesses • results depend on various parameter settings • requires more time from the expert • combinatorial explosion limits its applicability
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Hierarchy and Taxonomy
© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Graphical Representation ofInheritance
© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Inheritance Networks [1]:Trees with Strict Inheritance
Based on slide© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Inheritance Networks [2]:Lattices with Strict Inheritance
Based on slide© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Inheritance Networks [3]:Defeasible Inheritance
Based on slide© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Problems withShortest Path
Based on slide© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Formal:Inheritance Hierarchy
© 2005 M. Hauskrecht, Univ. of Pittsburgh CS 2740 Knowledge Representationhttp://www.cs.pitt.edu/~milos/courses/cs2710/
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Protégé API(Classes, properties,
individuals, etc.)
Protégé GUI(Tabs, Widgets, Menus)
DBStorage
Pro
tég
é C
ore
Sys
tem
Protégé OWL API(Logical class defn’s,
restrictions, etc.)
Protégé OWL GUI(Expression Editor,
Conditions Widget, etc.)
OWL FileStorage
Jena API(Parsing, Reasoning)
OW
L P
lug
in
OWLExtension APIs(SWRL, OWL-S, etc.)
OWL GUI Plugins(SWRL Editors, ezOWL,OWLViz, Wizards, etc.)
OWL Plug-in Architecture
Adapted from slide © 2004 H. Knublauch (formerly University of Manchester)http://www.knublauch.com
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
OWLMetadata
(Individuals)
OWLMetadata
(Individuals)
OWLMetadata
(Individuals)
OWLMetadata
(Individuals)
Tourism Ontology
Web Services
Destination
AccomodationActivity
© 2004 H. Knublauch (formerly University of Manchester)http://www.knublauch.com
Tourism Semantic Web
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Adapted from material © 2003 – 2004 S. Russell & P. Norvig.
SituationcalculusFigure 10.2p. 329 R&N 2e
Actions, Situations, Time & Events [1]: Situation Calculus Revisited
Axioms: Truth of Predicate P Fully specify situations where P
true biconditional (, iff)
Original Predicates Describe state of world Each augmented with situation
argument s
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Actions, Situations, Time & Events [2]: Event Calculus
Domain-Independent Axioms
Domain-Dependent Axioms
Still Need to Solve Frame Problem (by Circumscription)
Figure © 2003 S. Russell & P. Norvig.
Event calculusFigure 10.3p. 336 R&N 2e
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Actions, Situations, Time & Events [3]: Fluent Calculus
Fluent: Condition (Predicate) That Can Change Over Time (e.g., On)
Fluent Calculus: Variant of Situation Calculus Defaults ∘ (concatenation) of fluents with state
Figure © 2003 S. Russell & P. Norvig.
State fluentsFigure 10.6p. 340 R&N 2e
Washington
Adams Jefferson
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
CIKM:Review
Information Management Data acquisition: instrumentation, collection, polling, elicitation Data and information integration: combining multiple sources
May be heterogeneous (different in quality, format, rate, etc.)
Underlying formats, properties may correspond to different ontologies
Ontology mappings (functions to convert between ontologies) needed Data transformation: preparation for reasoning, learning
Preprocessing
Cleaning Includes knowledge capture: assimilation from various sources
Knowledge Management Term used most often in business administration, management science Related to IM, but capability and process-centered Focus on learning and KA, organization theory, decision theory
Discussion, apprenticeship, forums, libraries, training/mentoring
Modern theory: KBs, Expert Systems, Decision Support Systems
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Terminology
Knowledge Engineering (KE): Process of KR Design, Acquisition Knowledge
What agents possess (epistemology) that lets them reason
Basis for rational cognition, action
Knowledge gain (acquisition, learning): improvement in problem solving Knowledge level (vs. symbol level): level at which agents reason Semantic network: inheritance and membership/containment relationships Knowledge elicitation: KA/KE process from human domain experts
Protocol analysis: preparing, conducting, interpreting interview
Less formal methods: subjective estimation & probabilities
Fluents: Conditions (Predicates) That Can Change over Time Classes, nominals (objects / class instances): spatial, temporal extent Fluent calculus: situation calculus with defaults, ∘ (concatenation)
Computational Information and Knowledge Management (CIKM) Data/info integration & transformation: collecting, preparing data Includes knowledge capture: assimilation from various sources
Computing & Information SciencesKansas State University
Lecture 18 of 42CIS 530 / 730Artificial Intelligence
Summary Points
Last Class: Knowledge Engineering, Elicitation, Knowledge Rep. Elicitation
Techniques: unstructured, structured, “think aloud” (protocol analysis)
Stages: concept (last time), structure (today) Knowledge acquisition (KA) Information management, knowledge management defined KR: situation calculus and successor state axioms; fluents, intervals
Today: KE, Ontologies Concluded; CIKM; Event and Fluent Calculi Structure elicitation From semantic networks to ontologies Information management Knowledge management Event calculus Fluent calculus
Next Class: Defaults, Defeasible Reasoning; Planning Preview Coming Week: Planning (Section IV)