presentation - IIASA

NEXT GENERATION
MODEL MANAGEMENT
AND INTEGRATION
Prof. Daniel Dolk
CSM Workshop
August 2006

OVERVIEW
“It is possible to view every process that occurs in nature or elsewhere as a computation”
– Stephen Wolfram
“Information is static…computation computation is dynamic” – Rudy Rucker
• Retrospective of model management
• Elements of NGMMI
• Emergence of computational modeling and
computational experimentation in scientific
inquiry
• Virtual environments in the form of Society of
Simulations

SCIENTIFIC METHOD,
15 TH -20 TH CENTURIES
Theory
Design
Analysis and
Explanation
Experiment

SCIENTIFIC METHOD,
1950 - PRESENT
Analysis &
Explanation
Modeling &
Simulation
Analysis
Design
Design
Theory
Design
Validation
Experiment

MODEL MANAGEMENT REDUX
• “The process of representing, solving, analyzing and
integrating analytical models (primarily in the
information and management sciences (OR/MS)) in
computer executable form” -- Dolk
• Initially conceived as a modeling counterpart to data
management
– Models as a corporate resource
– Models, like data, need to be systematically managed
– Rich vein of OR/MS-based models and solvers
– Model management system
• Part of the troika of DSS
– Data, models, dialogue (later, knowledge)
– Models as the lynchpin for decision-making (Simon et al)

MODEL MANAGEMENT REDUX:
MODEL MANAGEMENT SYSTEM
“If DBMS, why not MMS?”
Desirable Requirements/Features of an MMS:
• Support entire modeling life cycle
• Uniform representation of models
• Modeling languages
• Portfolio of cross-paradigm OR/MS models
• Library of solvers
• Model integration
• Separation of models and data
• Separation of models and solvers
• Leverage RDBMS for data manipulation
Central issue: Model Representation

MODEL MANAGEMENT REDUX:
MODEL REPRESENTATION
Theoretical driver: Is there a way to represent
models that is comparable in power to the
relational theory representation of data?
• Relational (Blanning)
• Object-oriented (Dolk)
• Structured modeling (Geoffrion)
• Logic modeling (Bhargava, Kimbrough)
• Graph grammars (C. Jones)
• Metagraphs (Blanning, Basu)

FEEDMIX MODEL
SM GENUS GRAPH
NUTR
MATERIAL
NUTR_
MIN
MATERIAL
Q
UCOST
ANALYSIS
TOTCOST
T:NLEVEL
NLEVEL

SM ELEMENTAL DETAIL TABLES
FOR FEEDMIX MODEL

STRUCTURED MODELING
• Contributions
CONTRIBUTIONS
– Formal semantic ontology for models
– Math models as conceptual models
– Full language (SML) and implementation
(FW/SM)
– Model reuse and integration
– Multiple modes of model representation

LIMITATIONS OF
STRUCTURED MODELING
“Why isn’t t structured modeling (or equivalent) used as a matter of course in
OR/MS modeling endeavors?”
• Endogenous factors:
– No graph-driven implementation
– Static vs. dynamic representations
– Complexity of indexing semantics
• Exogenous factors:
– Math as legal tender: OR analysts are overwhelmingly mathematicians
– Overhead of conceptual modeling: Even database analysts resist conceptual c
modeling
– Models don’t t command the same respect as data
– UML has become the lingua franca of conceptual modeling
– Internet and distributed computing
• (“It can be done” & “It should be done”) ) ~=> (“It(
will be done”)

CONTRIBUTIONS OF 1 ST
GENERATION MODEL
MANAGEMENT
• Decoupling of models, solvers and data
– “run-time” binding of data and solver to model representation
• Model representation formalisms
– Structured modeling (Geoffrion(
Geoffrion); meta-modeling modeling (Blanning(
and
Basu); graph grammars (Jones); logic modeling (Bhargava and
Kimbrough); object-oriented oriented (Dolk(
Dolk)
– Modeling languages (AMPL: Fourer and Gay)
• Model integration
– Dimensional analysis (Bradley and Clemence)
– Semantic consistency (Bhargava and Kimbrough)
– Relational data systems for managing data
– Model composition (Dolk(
and Kottemann; Geoffrion)

LIMITATIONS OF 1 st
GENERATION MODEL
MANAGEMENT
• Decision-makers are, on average, “model averse”
• Never really a market for the MMS
– Cross-paradigm myopia in the OR/MS community
– “The spreadsheet is the MMS”
– Result: a fully functional MMS was never implemented
• Data more important than models
• No comprehensive, integrating theory (as in relational
data world)
• Internet shifted attention from static representations to
dynamic, distributed resources

MODEL MANAGEMENT AND
THE INTERNET
• Internet shifted the focus on many different
levels:
– from stand-alone alone machine centric (static) to
distributed network-centric (dynamic)
– from top down to bottom up
– from MMS as single monolithic system to MMS as
dynamic, configurable S/W components
– from S/W as commodity to S/W as service
– from individual problem solving to collaborative
problem solving
• AME

ELEMENTS OF
NEXT GENERATION
MODEL MANAGEMENT
There is still a need for model management, but this seems to go
largely unrecognized.
• Model management as an exemplar of knowledge management rather than t
an
extension of data management
– Models recast in the context of Knowledge and Knowledge Flow enablers, , or
– Models in the context of the Pentagram Creative Space (Involvement, nt, Imagination,
Intervention, Integration, Intelligence) (Nakamori(
Nakamori)
– “Model dynamics” (?) rather than “model management”
– Decision as a process ( (decision decision supply chain) rather than a point estimate
– Collaborative decision-making vs. individual decision-making
• Shift from analytical modeling to computational modeling and virtual environments
– Concept of complexity has changed radically
– Evolutionary biology has replaced physics as the scientific paradigm of interest in the social
sciences
– Ascendancy of network “science” and agent technology
– Model integration in one form as a Society of Simulations

KNOWLEDGE FLOWS IN
THE MODELING LIFECYCLE
Model Versioning
And Security
Problem
Identification
Model
Formulation
Model
Maintenance
Model
Implementation
Model
Validation
Model
Solution
Model
Interpretation

COMPUTATIONAL SCIENCE
• Computational science involves using computers to study scientific ic problems and
complements the areas of theory and experimentation in traditional al scientific
investigation.
• Computational science seeks to gain understanding of science principally through the
use and analysis of computational models, often on high performance computers.
• Computational modeling and simulation is being accepted as a third methodology in
engineering and scientific research that fills a gap between physical experiments and
analytical approaches.
• Experiments traditionally performed in a laboratory, wind tunnel, , or the field are
being augmented or replaced by computational experimentation (simulations).
• These simulations provide both qualitative and quantitative insights into many
phenomena that are too complex to be dealt with by analytical methods (e.g.,
organizational dynamics) or too expensive or dangerous to study by experiments
(e.g., bioterrorist attacks, nuclear repository integrity).

ASPECTS OF
COMPUTATIONAL MODELING
• Procedural as separate from equational or axiomatic
– E.g., cellular automata, Monte Carlo simulations for solving systems of
PDEs numerically
• Constructivist, or very nearly so, in nature
– “if you can’t t build it, you don’t t understand it” (Langton)
– Artifact-building vs. theory-building
• Emergent behavior vs. hierarchical decomposition & recomposition
• Types of models
– “what is”; ; descriptive (ex: discrete event simulation)
– “what should be”; ; prescriptive (ex: optimization)
– “what will be”; ; predictive (ex: econometric forecasting)
– “what could be”; ; constructive (ex: artificial life)

EXAMPLES OF COMPUTATIONAL
MODELING FOR SOME REFERENCE
DISCIPLINES
• Biology: DNA and the genome; artificial life
[Keller 2002]
• Physics: numerical analysis of systems of
PDEs
• Mathematics: Mathematica [Wolfram 2002]
• Finance: options pricing

COMPUTATIONAL MODELING in the
INFORMATION and SOCIAL
SCIENCES
• Computational models of human behavior
– How do we construct agents?
– Computational models of cognition [Edelman 1987]
– Experimental economics
– Economic decision-making under uncertainty (Tversky(
&
Kahneman)
• Organization science: Computational organizations
[Prietula
& Carley 1994; Levitt 2004]
• Economics: evolutionary economics [Nelson and Winter
2002]; synthetic economies [Epstein & Axtell 1996];
• Network “science”:: [Barabasi[
2002]; social network
analysis [Wassermann 1994]

COMPUTATIONAL
EXPERIMENTATION
• Computational experimentation as an alternative or augmentation to analytical /
laboratory and field experimentation
from [Nissen and Buettner 2004]

Computational
Modeling
Virtual Environments linked
via Network interfaces with
shared semantics
Design
Analysis
Hypothesis
Generation
Computational
Experimentation
Analysis
Design
Design
Theory
Design
Analysis, Confirmation/Refutation
Experiment
-Live: Laboratory
and Field
COMPUTATIONAL MODELING AND VIRTUAL
ENVIRONMENTS

MODEL DYNAMICS AND
VIRTUAL ENVIRONMENTS
• LEAD (Linked Environments for Atmospheric Discovery)
– Collaboration among meteorologists, computer scientists,
educational experts
– Objective:
• Respond to weather phenomena in real time
• Execute multi-model model simulations of weather forecasts distributed on
the Grid
• Adapt computing resources dynamically
– Services:
• Workflow system: dynamic control of experiments
• Metadata catalog for managing experimental results
• Notification system as a communications layer

SOCIETY of SIMULATIONS
APPROACH TO LINKING
VIRTUAL ENVIRONMENTS
• Problem: How do you link local virtual environments
(models) developed with local semantics into a global
virtual environment (integrated model) with a common
semantics? (This is the problem of the Semantic web;
also, to a large degree, the aggregation problem)
• A Society of Simulations is analogous to a society of
people, as both are loosely coupled constructs in which
independent individuals contribute toward a single
societal identity. A society is an organized group of
individuals who associate for common purposes.
• Likewise, autonomous simulations in a Society of
Simulations work together to achieve the common goal
of modeling the system.

SOCIETY of SIMULATIONS
COMPONENTS
• Members: Stand-alone alone simulations or models
(ABS, DES, SD, OR/MS, etc), built specifically for
a Society, other components such as
visualizations and user interfaces
• Shared Reality: stores the shared aspects of a
Member’s model(s)
• Liaisons: links Members with Shared Reality

MORE ELEMENTS OF
NEXT GENERATION
MODEL MANAGEMENT
• Solver environments
– Combining information systems and model development techniques
– Meta-heuristic environments
– Grid computing
– Network science
• Data (structured + semi- / unstructured)
– Search engine technology
– Advanced data/text/image mining
– Semantic Web
– Dynamically configurable and executable models a la Google type interfaces
• Application areas
– Supply chain management
– Services science (?), management and engineering: Web services, service-
oriented architecture as IME
– Computational economies, societies, organizations
– Network science (social network analysis)

APPLICATION AREA:
SEMANTIC WEB
• The Semantic Web is a web of data. There is lots of data we all use every
day, and its not part of the web. I can see my bank statements on the web,
and my photographs, and I can see my appointments in a calendar. But
can I see my photos in a calendar to see what I was doing when I took
them? Can I see bank statement lines in a calendar?
• Why not? Because we don't have a web of data. Because data is controlled
by applications, and each application keeps it to itself.
• The Semantic Web is about two things. It is about common formats for
interchange of data, where on the original Web we only had interchange of
documents. Also it is about language for recording how the data relates to
real world objects. That allows a person, or a machine, to start off in one
database, and then move through an unending set of databases which are
connected not by wires but by being about the same thing.
( http://www.w3.org/2001/sw/ )
( http://www.scientificamerican.com/article.cfm?articleID=00048144-
10D2-1C70
1C70-84A9809EC588EF21&pageNumber=1&catID=2
)
• Model Dynamics counterpart: Composing services to satisfy a user r request
is the same problem as composing models to solve a particular application.
plication.
• Research areas: Ontologies, , semantic resolution, dimensional consistency,
logical vs physical integration

APPLICATION AREA:
SERVICES MANAGEMENT
AND ENGINEERING
• “Services sciences, Management and Engineering
hopes to bring together ongoing work in computer
science, operations research, industrial
engineering, business strategy, management
sciences, social and cognitive sciences, and legal
sciences to develop the skills required in a services-
led economy.”
http://www.research.ibm.com/ssme/

APPLICATION AREA:
SERVICES MANAGEMENT
AND ENGINEERING
• “The science comes in through modeling. You model kernels of a
work practice to gain insight and for the purposes of automation”
Richard Newton, Dean of the College of Engineering at the
University of California, Berkeley.
• Modeling, simulation, abstraction, measurement and metrics, and
process design and analysis will emerge as core disciplines of
science-based services
• Equipped with the right tools (e.g. dynamically reconfigurable
architectures for “on demand” computing), nonprogrammers will
be able to design, model, and simulate business processes.

SOME RESEARCH AREAS
FOR NGMMI
• Computational models of human behavior
– Experimental economics, cognitive science, psychology, decision science
– Agent representations
• Ontologies
– Model assumptions, structural representations, dynamic representations, agent
behavior
• Model integration
– How to integrate inter-paradigm models such as ABS, DES, Optimization,
Forecasting, Soft vs. Crisp, Quantitative vs. Qualitative, etc., etc. models? How
do you represent these models and how do you merge them semantically? ally? (Ex:
artificial labor market)
– How to integrate intra-paradigm models? E.g., how do you integrate an ABS
whose agents are people with an ABS whose agents are strategies?
– Ontology integration (meta-ontology)
• Model validation, esp. for emergent (“what(
could be”) ) models
– What is (are) the role(s) ) of “what could be” models in scientific inquiry?
• Measurement of knowledge flows resulting from analytical/computational tional models
– How useful are models, really?
– “Good” vs. “Bad” models and their effects upon the Knowledge Base

Backup Slides

SOME REFERENCES
• COMPUTATIONAL EXPERIMENTATION
– Nissen, M. and Buettner, R. Computational experimentation with the t
Virtual Design
Team: Bridging the chasm between laboratory and field research in C2. " Proceedings
Command and Control Research and Technology Symposium, , San Diego, CA, 2004.
– Kevrekidis, , I. Equation-Free Modeling for Complex Systems. Frontiers of Engineering:
Reports on Leading-Edge Engineering from the 2004 NAE Symposium on Frontiers of
Engineering, 69-76.
• COMPUTATIONAL EXPLANATION
– Keller, E.F. Making Sense of Life: Explaining Biological Development with Models,
Metaphors, and Machines. Harvard University Press, Cambridge, MA, 2002.
– Kimbrough, S. Computational Modeling and Explanation: Opportunities ies for the
Information and Management Sciences. Computational Modeling and Problem Solving in
the Networked World, , Hemant K. Bhargava and Nong Ye, eds., Kluwer, , Boston, MA, 31-
57, 2003.
• COMPUTATIONAL ORGANIZATIONS
– Carley, , K. M. & Prietula, , M. J. (Eds.), 1994, Computational Organization Theory,
Hillsdale, NJ: Lawrence Erlbaum Associates.
– Levitt, , R. E., (2004). Computational Modeling of Organizations Comes of Age. Journal of
Computational & Mathematical Organization Theory, 10(2); 127-145, 145, July 2004.

REFERENCES (cont’d)
• EVOLUTIONARY ECONOMICS AND SYNTHETIC ECONOMIES
– Chaturvedi, A., Mehta, S., Dolk, D., Ayer, R. Agent-based simulation for
computational experimentation: developing an artificial labor market.
European
Journal of Operations Research 166:3, 694-716, 2005.
– Epstein, J. and Axtell, R. Growing Artificial Societies: Social Science from the
Bottom Up. . The Brookings Institution and the MIT Press, Washington D.C. and
Cambridge, MA, 1996.
– Nelson, R. and Winter, S. An Evolutionary Theory of Economic Change. The
Belknap Press of Harvard University Press, Cambridge MA, 1982.
• MODEL MANAGEMENT
– Basu, , A. and Blanning, , R. Model integration using metagraphs. Information
Systems Research, 5:3; 195-218, 1994.
– Bhargava, H. and Kimbrough, S. Model management: An embedded languages
approach. Decision Support Systems, 10; 277-299, 299, 1993.
– Dolk, D. Model integration in the data warehouse era. European Journal of
Operational Research, April 2000.
– Geoffrion, , A.M. An introduction to structured modeling. Management Science,
33: 5, 547-588, 588, May 1987.
– Jones, C. An introduction to graph based modeling systems, Part I: Overview.
ORSA Journal on Computing, , 136-151, 151, 1990.

REFERENCES (cont’d)
• NETWORK SCIENCE
– Barabasi, , A-L. A
Linked: How Everything is Connected to Everything Else and What
It Means for Business, Science, and Everyday Life. Plume Press, 2003.
– J.C. Doyle, D. Alderson, L. Li, S. Low, M. Roughan, , S. Shalunov, , R. Tanaka, and W.
Willinger. . The "robust yet fragile" nature of the Internet. Proc. Nat. Acad. Sci.
USA. October 4, 2005.
• SOCIAL NETWORK ANALYSIS
– Wasserman, S. and Faust, K. (1994) Social Network Analysis: Methods and
Applications. . Cambridge: Cambridge University Press.
– Krackhardt, , K. and Hanson, J. Informal networks: The company behind the
chart. Harvard Business Review, 103-111, 111, July-August, 1993.
• KNOWLEDGE MANAGEMENT AND DYNAMICS
– Wierzbicki, A. and Nakamori, , Y. (2006) Creative Space: Models of Creative
Processes for the Knowledge Civilization Age. Springer Press.
– Nissen, M. (2006) Harnessing Knowledge Dynamics: Principled Organizational
Knowing. IRM Press.

SOCIETY of SIMULATIONS:
MEMBER COMPONENT
• Each simulation, or model, in a Society is an
autonomously managed Member which
cooperates with other Members to reach its
personal goals.
• In the process of meeting its personal goals, a
Member contributes to societal goals.
• Satisfaction of societal goals emerges as all
Members progress towards their personal goals.

SOCIETY of SIMULATIONS:
MEMBER COMPONENT
• Members:
– Inputs/Outputs:
• Syntax: data structure and type
• Granularity: spatial and temporal (can differ widely across different
simulations
• Semantics: the meaning of an input/output (e.g., A door in a building layout
means a wooden obstacle to a FireSim and a removable blockage on an exit
route to a HumanSim)

SOCIETY of SIMULATIONS:
SHARED REALITY
• Shared Reality:
– Shared aspects of a Member’s s models
– Does not manage how the Members operate
– Persistent information space
– The intelligence for transforming information within Shared Reality into
a form a consumer can digest and for synchronizing a consumer with
produced data is pushed from the data exchange mechanism of Shared
Reality onto the linkages (Liaisons) that connect the Members to Shared
Reality.
– Shared Reality is lightweight, in the sense that overheads increase less
than linearly as the number of Members or the amount of data being
exchanged increases.
– Decouples the producers and consumers of data
• Member’s s design is separated from the data exchange mechanism.
• Extensions to a Member’s s design do not require changes to the design of
Shared Reality.

SOCIETY of SIMULATIONS
APPROACH
• Liaisons:
– Each Member in a Society accesses Shared Reality through a Member-
specific Liaison
– Liaison consists of the intelligence needed to interact with and control a
Member and to interact with the rest of the Society.
– Liaison is configured to use Member-specific mechanisms—
initializations, inputs, outputs, and control mechanisms.
– Same Member can be used in different Societies and be continuously
developed without being forced to address Society-specific
characteristics, enabling reuse and distributed development.
– Liaison Tasks:
• Synchronizes the Member with data the Member depends on
• Starts, stops, restarts, and checkpoints a Member.
• Gathers data from Shared Reality, transforms its syntax, converts s its
granularity, and translates its semantics.
• Places the Member’s s outputs into Shared Reality coupled with semantic
information describing the syntax, granularity, and semantics of the data.

EXAMPLE: EVACUATION
SOCIETY

BENEFITS of SoS APPROACH
• Enables distributed development
• Heterogeneity is supported by allowing independent development of o
Member designs
• Autonomous management is enabled by linking Members to
information instead of to other Members
• Avoids publisher-subscriber subscriber dependence
• Society of Simulations approach allows simulations to cooperate, yet
remain autonomous, an inherently modular and scalable approach
for linking heterogeneous simulations.
• Example: Urban Resolve 2015 (15 simulations, 6 of which use SoS;
2000 players; 2 weeks duration)
• SoS works primarily at the syntactic level; Next step: extend to the t
semantic level (Semantic Web)

STRUCTURED MODELING and
the 21 st CENTURY
• UML, ERD still do not support decision models
and OR/MS applications
• OLAP Extension: SM and OLAP
• Model Standardization: SM and XML
• SM and Ontology
• SM and KM: Wikipedia counterpart for models
• Dynamic SM
• SM and Computational Modeling: opportunities
in the life sciences?