MR POTATOHEAD goes onto OWL

wafflebazaarInternet and Web Development

Oct 21, 2013 (3 years and 11 months ago)

199 views

MR POTATOHEAD goes onto OWL
A Web Ontology Language representation of the
MR POTATOHEAD meta-model for agent-based models of
land-use and cover change
Dawn C.Parker
y
,J.Gary Polhill
z
,Seyed M.M.Rizi
y
y
Department of Computational Social Science,George Mason University
z
The Macaulay Institute
April 17,2008
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 1/17
Communicating ABM-LUCC
The problem
Why it is difficult to communicate ABM-LUCC
Technical
Model development and implementation ABM-LUCCs are implemented in
various coding languages and approaches.
Complex model structures ABM-LUCCs model complex systems that defy
mathematical representation due to processes of interaction across spatial,
temporal and institutional scales among heterogeneous,adaptive agents.
Heterogeneous rule structures ABM-LUCCs include both mathematical and
non-mathematical rules,so mathematics is insufficient as a language for model
communication.
Societal
Disciplinary variation Variety in ABM-LUCC modelers results in different end
uses,and coding platforms,styles and standards for ABM-LUCCs across
disciplines.
Misaligned incentives Problemsolving,publication,and promotion standards
do not require adherence to a common protocol for model communication.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 2/17
Communicating ABM-LUCC
The solutions
Symbols,graphics and conceptual design patterns
Symbolic
Symbolic solutions that rely on formal systems of representation such as mathematics
and predicate logic can classify run-time processes of the model,but fail to address
the substance of an ABM-LUCC.
Graphical
Graphical tools like UML are useful for model development,but their static nature
severely limits the quality of knowledge representation required for ABM-LUCC
communication.
Conceptual
A conceptual design pattern (CDP) is an abstract framework that describes the
conceptual structure of a model.
Procedural CDPs emphasize the model building process rather than model
components and the relations among them.
Ontological CDPs represent classes of concepts used in a model,their properties
and the relations among themconsistently.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 3/17
Ontological solutions
What is an ontology?
Definition
An ontology is a “formal,explicit specification of a shared conceptualization.”
(Gruber 1993).
Characteristics of an ontology
Conceptualization An ontology is an abstraction of reality.
Formal An ontology is based on logic;hence machine-readable.
Explicit All required concepts of an ontology are described.
Shared Descriptions in an ontology represent some consensus among a
community that matches their own understanding.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 4/17
Ontological solutions
Ontology basics
Classes,objects and relations
Components
Class A collection or type of objects Farmer
Instance A member of a class Jim
Property Attribute of a class hasCapital
Relations Relations among classes Farmer growsCrop
Subclass Is-a Jim Is-a Farmer
Composition Has-a Farmer Has-a Land
Restrictions Truth condition for an assertion
Rules If-then statements for inference
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 5/17
Ontological solutions
Ontology languages
Web Ontology Language (OWL)
OWL in brief
Description OWL is a family of languages based on description logics for
development of and inference on ontologies.
Advantages
Inference OWL is founded on description logics,so it achieves some inference
capabilities.In principle,some species of OWL are both complete and decidable;
however,the worst case scenario for inference on ontologies is NP.
Web integration OWL applications can be integrated into the web because OWL is
encodable in RDF and XML.
Code development OWL is developed through Prot´eg´e,an intuitive Java-based
development environment.
Active development Active development and applications support is making OWL
the de facto standard language for ontologies.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 6/17
Ontological solutions
Ontology languages
Prot´eg´e application development environment
OWL development in Prot´eg´e
Acknowledgment
This work was conducted using the Prot´eg´e resource,which is supported by grant LM007885 fromthe United States National Library of Medicine.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 7/17
MR POTATOHEAD
Structure of MR POTATOHEAD
Classes and object properties
Description of MR POTATOHEAD
MR POTATOHEAD consists of classes,class properties and class relations.Class
properties are either other classes or primitives.
Top tier view of MR POTATOHEAD
Second tier view of MR POTATOHEAD
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 8/17
MR POTATOHEAD
Structure of MR POTATOHEAD
A more in-depth look at MR POTATOHEAD
We can view ontologies in two dimensions of depth and breadth.Depth refers to the
maximumlevel of hierarchy in an ontology.Breadth refers to the lateral relations
among classes.MR POTATOHEAD is not overly deep;however,it is quite dense in
lateral relations.
Land exchange rules
Spatial data structures
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 9/17
MR POTATOHEAD
Uses of MR POTATOHEAD
Model comparison I
Comparison by graphics
LUCITA and IMSHED differ in both the mechanismand process of land exchange.
Rules for land exchange in LUCITA and IMSHED
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 10/17
MR POTATOHEAD
Uses of MR POTATOHEAD
Model comparison II
Comparison by graphics
Spatial data structures of SYPRIA and SLUDGE share the same kind of decision
making units and spatial data types and differ on all other attributes.
Spatial data structures in SYPRIA and SLUDGE
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 11/17
MR POTATOHEAD
Uses of MR POTATOHEAD
Model inference
Logical consistency
OWL tests the consistency of an ontology:assume A !B means A is the parent of
B.Imagine an ontology asserts A !B !C !A.OWL enables reasoning
software to catch this logical inconsistency.Inferences based on MR POTATOHEAD
are guaranteed to be logically sound,because it is shown to be consistent.
Logical validity
A valid model satisfies non-axiomatic,necessary,sufficient or necessary and
sufficient conditions derived fromdomain knowledge and asserted by the modeler.
MR POTATOHEAD asserts that in an ABM-LUCC,18 classes such as spatial data
structures,potential land uses and agent decision model are necessary properties of
other classes.This means that an ABM-LUCC has all these classes in the right places
in the ontology;however,satisfying this necessary condition does not guarantee an
ABM-LUCC.For example,IMSHED is not an ABM-LUCC according to MR
POTATOHEAD,because it lacks two essential classes:Payoffs and
DecisionMakingUnits.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 12/17
Challenges
Ontology representation,inference and evolution
Representing processes of an ABM-LUCC
Ontology is a static representation of knowledge that cannot capture the dynamics of a
model.Can an ontological structure be expanded to represent dynamics?
Probabilistic inference
Reasoning on ontologies is based on crisp inference that provides a solid basis for
testing model consistency;but fails to recognize stochasticity in model validity:the
OWL representation of MR POTATOHEAD rejects a model as ABM-LUCC
regardless of how many necessary conditions it lacks.Intuitively,we rank models as
to how close they are to an ideal type.Crisp reasoning defies intuition.
Consortiumand community standards
Ontologies are developed in an interplay of inductive observation and deductive
principles.The authority of a consistent ontology is derived fromits validity as
established by the user community,so to a large degree,scientific practices of the
community for which ontologies are built determine their use and evolution.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 13/17
Roles for MR POTATOHEAD
Proposed roles for MR POTATOHEAD
Design template
Documenting model elements and structure
1
What are possible building blocks for an ABM-LUCC?
2
What are possible heterogeneous implementations of these elements?
Defining minimumrequired elements for an ABM-LUCC
Assisting modelers in model design
Model communication and comparison
Documenting specific choices or instantiations of a given ABM-LUCC,
contributing to replicability.
Comparing different implementations of an ABM-LUCC.
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 14/17
Roles for MR POTATOHEAD
Contributing to a theory of land-change science
Potential contributions
Consolidating and organizing acknowledged drivers of land-use change.
Demonstrating the scope of and relations among disciplinary elements of
ABM-LUCC.
Creating a meta-model that nests a variety of existing models (now 8) as special
cases.
Identifying areas in ABM-LUCC that are under-explored by current modeling
efforts such as model drivers and processes.
First steps
Comparison of 4 ABM-LUCCs of frontier regions in “Case studies,cross-site
comparisons,and the challenge of generalization:Comparing agent-based models of land-use
change in frontier regions,” Parker et al.,forthcoming,JLUS.)
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 15/17
Roles for MR POTATOHEAD
Future work
Graphical modeling front-end
Model instantiation in MR POTATOHEAD Code instantiation of the
meta-model and special cases where models would be designed and run through
a graphical modeling front-end using MR POTATOHEAD hierarchy to reduce
barriers to entry to ABM-LUCC modeling..
Animating MR POTATOHEAD!
Community modeling If resources are identified,MR POTATOHEAD could be
revisited or revised through a community modeling effort.
Toward a utopia for modelers:::
ABM-LUCC modelers would move toward the same future envisioned by pioneers in
computational biology:“a future in which not just:::models but all the pieces of
models should be sharable.In this utopia,models should be able to swap computer
code:::as easily as Mr.Potato Head swaps noses.” (Krieger 2006,p.189)
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 16/17
Reflections on OWL
What more can we do with OWL?
Representing process and causality
Ultimate goal is a replacement for analytical mathematics
OWL does not naturally represent causal relationships
These relationships are critical for understanding links between exogenous
drivers and emergent outcomes in ABM
Is there a way to represent these in OWL,or in any ontology?
First step:Creating model views
We plan as a first test to create multiple model views using OWL’s logical inference
capabilities:
Identify exogenous and endogenous elements
Identify parameters varied for experiments
Identify pattern-based outcomes/validation targets
Identify the scale of operation of different drivers
Other?
D.Parker (CSS,GMU)
MR POTATOHEAD goes onto OWL
April 17,2008 17/17