SAS/OR slides from talks at the 2014 INFORMS Annual Conference


The 2014 INFORMS Annual Conference in San Francisco was quite a success. Record attendance, diverse program, great city, lovely weather: who can ask for more? SAS and, in particular, SAS/OR was well-represented with a number of talks in all areas of operations research. Here is a somewhat arbitrary selection, please click the titles to get the PDF version of the slides.

Mixed integer optimization

Two notable talks here:

Imre Pólik (with Menal Güzelsoy, Philipp Christophel, Amar Narisetty, Yan Xu and Matthew Galati), "New Features and Improvements in SAS/OR 13.2"

This was our overview talk about what we had done since INFORMS 2013, especially in the MILP solver. This session, with the same participants (XPRESS/FICO, IBM/CPLEX, Gurobi, and SAS/OR) has been a permanent one for the last few years. It is always fun to see what each vendor is working on. Make sure to come to one of these sessions in an upcoming INFORMS annual meeting and learn what it means to be on the cutting edge of MILP!

Menal Güzelsoy, Imre Pólik, Philipp Christophel, "Using and Reusing Dual Information in Branch and Bound"

This talk generated quite a lot of discussion both online and offline. The idea is very simple: store the dual solutions and dual rays that are obtained during branch and bound, and reuse them later in the tree. What makes this practically possible is that most MILP problems have all boxed variables, so the duals are feasible universally in the tree. The gains are quite significant.

Data mining and machine learning

As you can imagine SAS has an extensive portfolio of machine learning and data mining procedures, including large-scale and big data variants. In this particular session of the conference we talked about some of the algorithms and implementations behind these procedures.

Yan Xu, Joshua Griffin, "High Performance Second-Order Procedures for Dense SVM and Quantile Regression"


Wenwen Zhou, "Quasi-Newton Extensions of an Active-Set Approach for Mixed Linear Models"


Joshua Griffin, Ben-Hao Wang, "Distributed Hessian-Free Optimization for Data Mining Applications"



But SAS/OR is not only about optimization, we also have discrete-event simulation. Let me close with two notable talks from this conference, one on theory, one on applications:

James Wilson, Kai-Wen Tien (NCSU), Christos Alexopoulos, David Goldsman (Georgia Tech), Anup Mokashi, "A Sequential Procedure for Estimating Steady-State Quantiles"

Our simulation group is especially active in working with academics to develop new methods. They regularly publish their results in journals and conferences. Talks like this one are the result of this collaboration.

James Wilson (NCSU), Emily Lada, Anup Mokashi, "Effective Simulation Warm-up for a Neonatal Intensive Care Unit"

The second talk is from a case study. Using operations research in healthcare has been a growing trend for the last decade at all levels: diagnostics, treatment, scheduling, optimization. Developing reliable models for the operation of a hospital ward is essential to be able to make informed decisions. In this talk they focused on developing the model for a neonatal intensive care unit and validating it with historical data. Emily Lada is going to blog about this project in more detail in the coming weeks, so check back to learn more.


Have a question on any of the talks? Have you done some related research? Leave us a comment and we'll get back to you.


About Author

Imre Pólik

Principal Operations Research Specialist

Imre Pólik is an Optimization Solver Developer for SAS/OR, working on the linear and mixed-integer optimization solvers. He earned his M.S. in Mathematics from Eötvös University in Budapest, Hungary, and his Ph.D. in Mathematics from McMaster University, Hamilton, Ontario, Canada. He has worked for SAS since 2010, following a couple of years as a visiting assistant professor at Lehigh University. Currently he is managing the Optimization and Network Algorithms team within the Analytics R&D Division.

Leave A Reply

Back to Top