Leo Lopes' Major Research Initiatives

The Stochastic Programming IDE An Integrated Development Environment for modeling, solving, and analysing Stochastic Programming Problems.


The idea of this initiative is to simplify the entire life cycle of developing optimization problems. It builds on previous progress in the fields of Operations Research and Computer Science.


Problem-specific structure is often invaluable in Operations Research. A very significant portion of the OR literature is devoted to solution methodologies based on detecting and exploiting special structure in problem descriptions. Exploiting problem structure in modeling systems provides two major benefits: increasing the productivity of the modeler, often the most expensive resource in an OR-based solution approach; and providing significant new opportunities for detecting and manipulating problem structure.


A similar approach is rampant in all broadly applicable areas of Computer Science (data base systems, operating systems, systems analysis and design, programming languages, etc...). Once sufficient progress in the performance engines is accomplished, modeling tools specialize. These specialized tools then place on the machine the burden of converting a modeler-centric representation of a problem into one that is computationally useful.


Another benefit from the approach of specializing modeling environments to specific applications is that it would allow less sophisticated users access to the technology. This is an important motivation for my research in general. I believe that tremendous benefits can be obtained by improving communication between the problem owner (typically a decision maker) and the problem solver (typically an engineer or consultant). Where the problem owner does not have sufficient technical sophistication, the onus should fall on the technical side to bridge the gap


The first concrete outcome of this initiative is some work on manipulating instances of Stochastic Optimization problems so that their structure can be made amenable to different solution approaches. Preliminary work on that problem was presented at the IX International Stochastic Programming Conference, in Berlin, Germany, in September 2001, and a research paper has subsequently been submitted to the journal Annals of Operations Research. 

A second outcome, on an AMPL-based modeling tool that emphasizes Stochastic Programming-specific constructs, is in advanced stages. A research paper has been completed and will be submitted soon.


The third concrete outcome is a UML-based modeling language for aiding the development of Stochastic Optimization Models. Results from this research have been presented at the International Symposium on Mathematical Programming in Copenhagen, Denmark and at the INFORMS Annual Meeting in Atlanta, GA in 2003. A paper describing this work is in its final stages of completion and will be submitted soon.

A Comprehensive Computational Infrastructure for Stochastic Programming: The idea of this work is to create the strategic components needed to improve the availablity of large-scale solutions incorporating stochastic opimization models. This includes, but is not limited to: solution algorithms designed for a grid; mechanisms for data communication; integration of stochastic optimization modeling tools with related techniques, such as optimization via simulation. Initial work on this initiative is involving adding a Stochastic Modeling Interface to COIN, a project lead by Alan King of IBM Research. Preliminary progress on this work will be presented at the INFORMS Annual Meeting in Atlanta, GA in 2003.

HUGS and the Job Shop Problem: Human Guided Search is an approach for combinatorial optimization designed by Neal Lesh, Joe Marks, Brian Mirtich, et. al. at Mitsubishi Electric Research Laboratories. It consists of a collection of search procedures that are guided out of local minima by the use of human knowledge. It has shown to be competitive with current metaheuristic search procedures in important cases, and has many advantages with regard to flexibility and interactivity when compared to more traditional techniques. In my internship at MERL in the summer of 2000, I created a HUGS-based application for the Job Shop Problem. Work on that approach was presented by Neal Lesh at the INFORMS 2001 Meeting in Miami Beach, Fl., and subsequently a research paper on the topic was published in the 3rd International NASA Workshop on Planning and Scheduling for Space.

HUGS is another example of how keeping the decision maker involved in the solution process can be beneficial. In addition to the obvious benefit of taking advantage of knowledge that originates in the decision-makers experience with the problem and can not be conveyed easily in mathematical form, this approach has a component of ownership which might increase the success probability in practical implementations, where a human factor can not be ignored.

XML and communication of Optimization Problems: This is an initiative for developing a modern exchange mechanism for instances of optimization problems. This effort is undertaken in collaboration with many other researchers, commercial interests in our industry, and clients of optimization technology. It is inspired by progress made by the Computer Algebra community and is an attempt to address shortcomings in our current representation mechanisms. It is work in progress, but has attracted much attention both in the Academic and the Business community.


The first concrete outcome of this initiative is a paper with Kipp Martin at the University of Chicago GSB and my advisor, Bob Fourer at Northwestern University on an XML-based standard for Linear programs. This paper is in its final completion stages and will be submitted soon.