2 Takes on
CyberInfrastructure for OR
2004 NSF Design, Service and
Manufacturing Grantees and
Research Conference
Workshop on CyberInfrastructure
January 6, 2004
Craig Tovey
ISyE & COC, Georgia Tech
Cyberinfrastructure - Current Activities
Operations Research: take 1
• Research
1.
Repositories for implemented algorithms. Informs
resource collection: > 100 packages and online programs
1. COIN - OR
2. NEOS Server
3. Specific algorithms e.g. Ye’s semidefinite opt code
2. Repositories for test problems. Informs resource: 25
instance sets & generators.
1. DIMACS combinatorial challenge sets
2. MIPLIB, LP test sets
3. Problem specific sets e.g. TSP,QAP,VRPTW.
3. Simulation tools: relatively few, often commercial-linked
4. Stat tools: several libraries and web-available routines
Cyberinfrastructure - Current
Activities
Operations Research: take 1
•
Education
1. Lecture notes for many courses
2. Tutorials on specialized topics
•
Implementation
1. Commercial optimization and modeling software
2. Commercial simulation software
3. Spreadsheet add-ins
Cyberinfrastructure – Vision for DMII
Operations Research- Take 1
•
Research
1. Unified, robust algorithms. Make it the standard
that researchers add to the repository. We would
build more on each other’s work. Quality and
development speed would increase.
2. Simulation building blocks
1. Modeling
2. Output analysis
3. Parameter search
3. Problem testbeds and instance generators.
•
Education and Implementation
1. More general modeling language, unifying
CP, IP/LP, NLP.
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
What are the big
opportunities
in OR?
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
Claim: OR is unusual among
science/engineering disciplines in that the
applicability of present technology, as
perceived internally, is 10-100 times the
present level.
What’s wrong? OR is about decision making,
not just solving models. Old news…
but CyberInfrastructure could increase the
applicability of OR by an order of
magnitude.
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
How can CI help?
1. CASE STUDIES.
• Gap exists between business problem and model
• We have ~50 times as many algorithm sites as
case study sites.
• Computing and web resources now capable of
storing and transmitting the large, unstructured
variegated info accompanying a real problem
• Use testbeds for both research and education
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
2. TOOLS FOR ORGANIZING, VISUALIZING,
CLEANING, EXTRACTING DATA FOR USE
IN MODELING.
•
•
•
Organizing and cleaning data blocks many
applications.
Expose students to messy problems and data.
Greatly increase the number of graduates competent
to use serious OR tools on real problems. B.S.-Ph.D.
Collaborative work between statistics and OR might
produce useful tools for linking data, forecasting,
uncertainty in models.
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
3. AUTOMATED OR INTELLIGENTAGENT-GUIDED WEB-BASED
MODELING.
•
•
•
Enable end-user modeling (perhaps crude,
perhaps if-then)
It succeeds if it helps the decision-maker
realize that the problem calls for OR methods
Seamless links between modeling languages
& simulations, and spreadsheets & databases.
Cyberinfrastructure – Vision for DMII
Operations Research- Take 2
Virtual domains: e.g. jobshop
Descriptive modeling tools: e.g. warehouse,
supply chain