We are trying something new at ICS this year - advanced tutorials! The idea behind these tutorials is to provide an in-depth discussion of "hot" topics of current and growing interest to the ICS community.

Topics for this inaugural effort include:

Clearing the Jungle of Stochastic Optimization

Warren Powell, Princeton University

Abstract:

Stochastic programming, stochastic search, dynamic programming, and stochastic control represent the major communities that work on the rich array of problems that fall under the umbrella of computational stochastic optimization.  These major disciplines have spawned additional subcommunities using names such as model predictive control, approximate dynamic programming, reinforcement learning, and simulation optimization.  Differences in notation, but most importantly differences in applications, have disguised the parallel development of similar ideas.  The challenge, as always, is computation.  There continues to be a significant gap between our ability to design algorithms (sometimes with convergence proofs) and making them work on practical problems.  In this talk, I will build bridges between stochastic search, dynamic programming, and stochastic programming, including a step-by-step translation from the notation of classical Markov decision processes to that of stochastic programming.  I will demonstrate that stochastic programming is a form of lookahead policy (also known as model predictive control), which can be optimized using the tools of stochastic search.  Throughout I will use examples to bring out the computational implications of each algorithmic class.  

Speaker Bio:

Warren B. Powell is a professor in the Department of Operations Research and Financial Engineering at Princeton University, where he has taught since 1981.  He is the director of CASTLE Laboratory (http://www.castlelab.princeton.edu), which specializes in the development of stochastic optimization models and algorithms with applications in transportation and logistics, energy, health and finance.  He recently established the Princeton Laboratory for Energy Systems Analysis (http://energysystems.princeton.edu) to take this work into the area of energy systems.  He is the author ofApproximate Dynamic Programming: Solving the curses of dimensionality and a co-author of Optimal Learning, both published by Wiley. 

Advanced Decomposition Methods in Mixed-Integer Programming

Ted Ralphs, Lehigh University

Abstract

Decomposition is a fundamental technique that underlies many solution methods for integer programming problems. The principle of decomposition is to identify and exploit tractable substructures, which generally arises either from the relaxation of constraints or the fixing of variables. Decomposition can be used both to improve bounds when the linear programming relaxation is weak and to improve tractability when the linear programming relaxation is difficult to solve. In this tutorial, we present a conceptual framework for understanding decomposition.

In the first part of the tutorial, we walk through a number of examples that illustrate the principles and relationships of various decomposition-based methods. In the second part of the tutorial, we describe the issues involved in implementation of these methods. We discuss recent efforts to develop decomposition-based generic solvers for mixed integer linear optimization problems, as well as recent work on the development of a modeling language and software framework that makes it possible to easily experiment with various advanced decomposition approaches, such as Dantzig-Wolfe decomposition, Lagrangian relaxation, and cutting plane methods.

Speaker Bio

Ted Ralphs received his Ph.D. in Operations Research from Cornell University in 1995. He is currently an associate professor in the Department of Industrial and Systems Engineering (ISE) at Lehigh University, where he is director of the Laboratory for Computational Optimization Research and chair of Lehigh's High-performance Computing Steering Committee. He is a co-founder of the COIN-OR Foundation, a non-profit foundation promoting the development of open source software for operations research and is currently chair of the Technical Leadership Council and a member of the Strategic Leadership Board, as well as project manager of a half dozen projects hosted in the COIN-OR open source software repository. He serves on the editorial boards of several leading journals. His research interests include development of methodology for solving discrete optimization problems, including those with multiple levels or multiple objectives;  development of parallel search algorithms; development of open source software; and applications of discrete optimization.

Cloud Computing for Optimization

Jeff Linderoth, University of Wisconsin Madison

I will describe (mostly) old experiences with using ubiquitous, convenient, on-demand, shared computing resources to solve large-scale optimization problems.  In olden days, these computing resources were known as a "metacomputer."  In the early 2000's, the name changed to the "computational grid," and now using computing resources in this way is known as "cloud computing."  I will also survey the landscape of cloud computing and its use and potential for Operations Research and Optimization.

Speaker Bio

Jeff Linderoth is a Professor in the departments of Industrial and Systems Engineering and Computer Sciences (by courtesy) at the University of Wisconsin-Madison, joining both departments in 2007. Dr. Linderoth received his Ph.D. degree from the Georgia Institute of Technology in 1998.  From 1998-2000, he was employed with the Mathematics and Computer Science Division at Argonne National Laboratory, and from 2000-2002, he was a Senior Consultant with the optimization-based financial products firm of Axioma Inc.  Prior to joining University of Wisconsin-Madison, from 2002-2007, he was a Assistant Professor at Lehigh University.  His research interests lie in the areas of discrete optimization, stochastic programming, and computing.  In 2002, he was awarded with his coauthors the SIAM Activity Group on Optimization Prize.  He currently is an area editor for Mathematical Programming Computation and serves on the editorial boards of four additional journals, including the INFORMS Journal on Computing and Operations Research.  Dr. Linderoth formerly served the INFORMS Computing Society as its Secretary Treasurer from 2006-2008 and as its Newsletter Editor from 2009-2010, where he managed to turn out a whopping one issue.  Dr. Linderoth's ability to be a slacker is further evidenced by the fact that he was the last person to submit his abstract for the 2013 ICS Conference.  Dr. Linderoth's personal interests include golf, chess, beer, and human pyramids.