Research
Research
I love optimization and high-performance and grid computing. Below is a little overview of the research projects in which I am involved.
MW is a software toolkit for making master-worker style applications running in distributed, heterogeneous, opportunistic environments known as computational grids. MW applications often use Condor as a resource management tool.
This is joint work with Steve Wright and the very talented Condor team, specifically Greg Thain.
Support for MW from the National Science Foundation, Division of Shared Cyberinfrastructure under Award #0330607 is gratefully acknowledged
With Jerry Shen, we are working on a nested-decomposition based algorithm for multi-stage stochastic linear programming that is engineered to run on a computational grid, using the MW software. Our goal is to have the software be able to solve instances of unprecedented size, and to use the solver to empirically test how sampled approximations of multistage problems compare to the true problem.
We are working on a set-partitioning-based model of the stochastic vehicle routing problem. We are interested in comparing how this approach to routing decisions compares to alternative approaches such as stochastic dynamic programming. This is joint work with Rosemary Berger, Clara Novoa, and Bob Storer
With Banu Gemici and David Wu, we built and implemented a three-phase decision support system for a large semiconductor manufacturer. This company was interested in constructing a "good" portfolio of products to develop while controlling the enormous amount of risk inherent in their business. Two keys to success of the approach were a smapling-based scenario model and multistage stochastic programming.
The Network Interdiction Problem involves interrupting an adversary's ability to maximize flow through a capacitated network by destroying certain portions of the network. A budget constraint limits the amount of the network that can be destroyed. In this paper, we study a stochastic version of the network interdiction problem in which the destruction of an arc of the network is a Bernoulli random variable, and the objective is to minimize the maximum expected flow of the adversary. Using duality and linearization techniques, an equivalent deterministic equivalent mixed integer program is formulated. The integer variables in the reformulation appear only in the first stage problem, so efficient decomposition techniques exist for its solution. Using a parallel algorithm designed to run on a computational grid, we give computational results showing the efficacy of a sampling based-approach to solving the true problem.
This is joint work with my Ph.D. student Udom Janjarassuk