Courses
  /  
Descriptions
  /  
Keep
IEMS 490-1: Selected Topics in Industrial Engineering

Quarter Offered

Fall : TTH 3:30-4:50 ; Staum (Frontiers in Simulation) - see below for course details
Winter : T 1:00-3:50 ; Contractor (Computational Social Science) - see below for course details
Spring : MW 2:00-3:20 ; Nohadani (Robust Optimization) - see below for course details

Description

Frontiers of Simulation - Fall Quarter 2015

Prerequisite: Permission of the instructor: familiarity with stochastic simulation in industrial engineering and operations research (IE/OR), plus PhD-level research maturity, are necessary.

Objectives: Learn about current research directions in stochastic simulation in IE/OR and contribute to setting research agendas.  Build skills valuable in a research career, such as: summarizing knowledge in a particular area, presenting, asking intelligent questions, performing a critical review of a research paper, and technical writing.

Class Format: Each of our meetings will begin with a 40-50 minute presentation of one or more papers.  Registered students will be required to read the paper(s) in advance.  The presentation will be followed by group discussion.

Tentative Outline of Topics Covered:

  • Optimization via simulation
  • Simulation input uncertainty analysis
  • Simulation metamodeling
  • Parallel computing in simulation
  • Simulation analytics and green simulation

Evaluation:

  • Class participation will be evaluated.  Asking good questions and giving good answers to questions are valuable.  Attendance will be a factor in the class participation grade. 
  • Each student will make one or two presentations. Each presentation will also involve a written report.  Topics and presentation dates will be assigned by the instructor near the beginning of the quarter. 

Computational Social Science - Winter Quarter - 2016

Winter/Spring Quarters (.5 credit in each quarter or 1 credit in either quarter) 2015/2016

Noshir Contractor, Northwestern University
Joseph N. Cappella, University of Pennsylvania
Dhavan Shah, University of Wisconsin Madison
 
Time: 1:00-4:00 CST / 2:00-5:00 EST on Tuesdays between 1/12/16 to 5/10/16
Class meetings will be synchronous across locations

Schedule:   Sixteen instructional sessions are slotted on Tuesdays at the specified time during the specified dates. Each registering students is required to attend at least thirteen (13) of the sixteen (16) scheduled meetings. Given the three-hour meeting time of the seminar, that will generate 39.0 hours of instructional time at each participating institution. The formal meeting dates will be bookended with an optional kickoff meeting and an optional closing meeting. See a complete schedule below, including weeks by coordinating faculty member, topic areas, and tentative guest speaker.  

Structure:  After the introductory session when we discuss goals and explore points of collaboration, each participating faculty member will be responsible for programing five sessions, with a commitment to lead at least one session and secure four leading researchers in the domain of computational social science as speakers.  Northwestern will handle the technology support, with each participating unit securing a room with a large monitor (for simulcasting), a conference calling center, and a webcam (for interaction),  Each weekly session will have one hour dedicated to hearing a presentation from a distinguished researcher on the topic, followed by an hour of Q&A and joint discussion that will include the presenter, the participating faculty, and the students from each institution.  After that, instructors will go offline for a focused session of one hour with the students at their institution.

Class will culminate with a workshop hosted at Northwestern University:

June 21-22, 2016 at Evanston, IL

The workshop is scheduled to preceed the  2016 International Conference on Computational Social Science, June 23-26 at Northwestern University http://www.kellogg.northwestern.edu/news-events/conference/ic2s2/2016.aspx

Profile:      Students who take this class should have some familiarity with computational methods and tools, and an active research project or program that involves use of these approaches in social science inquiry.

Topics:        Data acquisition, management and preparation, Data ethics, retention, and privacy protections, Computational approaches to language processing, Computational approaches in network science, Computational approaches to social media data, Computational approaches to recommendation systems, Emerging computational models and modeling tools.

Robust Optimization - Spring Quarter 2016

Today, the modeling of uncertainty is often done assuming that the uncertain variables and/or parameters follow a probability distribution, allowing us to apply the concepts of probability theory. The advantage of this approach is that a large set of methods already exist to provide solutions to many problems using the concept of stochastic programming. The success of these approaches are limited two-fold: i) many uncertainties do not follow a probabilistic distribution and even if, this information is often not readily and reliably available; and ii) probabilistic methods have not been developed to be also computationally tractable. Therefore, their applicability is restricted, whenever the dimension of the problem increases, as is the case in some queuing networks, auction design in multi-items, multi-bidder auctions, network information, and optimization under uncertainty to mention a few. In this context, robust optimization offers a new paradigm that allows to leap beyond these limitations and warrant solutions that are both optimal and immune against uncertainties.

Course Objectives

  • A brief introduction to probability theory based on the Kolmogorov axioms. Mathematical programs are discussed, where some of the data incorporated either into the objective or at times in constraints are uncertain and is modeled by a probability distribution.
  • An alternative via robust optimization (RO) is proposed to model uncertain phenomena. The main idea is that instead of assuming a distribution on the uncertainty, the results of probability theory are exploited. This is done by replacing the Kolmogorov axioms and the concept of random variables with uncertainty sets that are derived from some of the asymptotic implications of probability theory. This results in highly structured optimization problems (linear, semidefinite, mixed integer) for which there exist efficient, practical algorithms.
  • RO is discussed as a tractable methodology for solving optimization problems in the presence of uncertainties. Specifically we review under what conditions RO problems remain tractable when modeling uncertainty via uncertainty sets.
  • Students are exposed to a large number of applications ranging from supply chains, revenue man- agement, energy, portfolio theory, options pricing, risk management, Kalman filtering, queueing the- ory, information theory, statistics and engineering design. Students learn how to model and optimize uncertain phenomena using RO.
  • Students are exposed to the tools for large scale computation for RO, as necessary for real-world problems.