A Unified Framework for Distributed Stochastic Optimization
Abstract
Stochastic optimization is a widely adopted paradigm for decision making under uncertainty. Many different stochastic optimization models have been proposed depending on the performance criteria to be optimized. These modelstypically give rise to very large scale" optimization problems, and different solution approaches have been developed for the different problem classes. Most of the existin""g methods are not well suited for modern distributed computingarchitectures, especially in the nonconvex setting, e.g. in the prese""nce of integer variables. In this project, we propose to develop a unified framework for distributed decomposition of the different"" classes of large scale stochastic optimization problems. In particular, we will investigate decomposition approaches based on split""ting the problem across the uncertainty dimension, and their implementations on heterogenous distributed computing architectures. So"me key theoretical challenges to be studied are regarding the nonseparability of general stochastic optimization objective functions", nonconvexity and duality gap issues in the presence of integer variables, and handling asynchronous iterations in distributed impl"ementations. We will demonstrate and validate our algorithmic innovations in several important classes of stochastic optimization pr"oblems, including, day-ahead generation scheduling arising in power systems operations, stochastic network interdiction problems ari""sing in defense applications, and stochastic routing problems arising in logistics applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 23, 2018
- Source ID
- N000141812075
Entities
People
- Shabbir Ahmed
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy