2016 INFORMS Annual Meeting Program

SB21

INFORMS Nashville – 2016

SB20 106C-MCC Methods and Applications of Network Sampling Invited: Tutorial Invited Session Chair: Mohammad Al Hasan, Indiana University/Purdue University, Department of Computer Science, Indianapolis, IN, 46202, United States, alhasan@iupui.edu 1 - Methods And Applications Of Network Sampling Mohammad Al Hasan, Indiana University/Purdue University, Indianapolis, IN, 46202, United States, alhasan@iupui.edu Network data appears in various domains, including social, communication, and information sciences. Analysis of such data is crucial for making inferences and predictions about these networks, and moreover, for understanding the different processes that drive their evolution. However, a major bottleneck to perform such an analysis is the massive size of real-life networks, which makes modeling and analyzing these networks simply infeasible. Further, many networks, specifically, those that belong to social and communication domains are not visible to the public due to privacy concerns, and other networks, such as the Web, are only accessible via crawling. Therefore, to overcome the above challenges, researchers use network sampling overwhelmingly as a key statistical approach to select a sub-population of interest that can be studied thoroughly. In this tutorial, we aim to cover a diverse collection of methodologies and applications of network sampling. We will base the discussion of network sampling in terms of population of interest (vertices, edges, motifs), and sampling methodologies (such as Metropolis-Hastings, random walk, and importance sampling). We will also present a number of applications of these methods. Chair: Sarah Kadish, Dana-Farber Cancer Instititue, 450 Brookline Ave, Boston, MA, 02215, United States, sarah_kadish@dfci.harvard.edu 1 - Determining The Optimal Schedule For Premixing Chemotherapy Drugs Donald Richardson, University of Michigan, 2753 IOE Building, 1205 Beal, Ann Arbor, MI, 48109-2117, United States, donalric@umich.edu, Amy Cohn In collaboration with the University of Michigan Comprehensive Cancer Center, we have developed an optimization-based approach to improve make-ahead policies for chemotherapy drugs for infusion patients. We first present our optimization model. Then we present our proposed user interface to aid our collaborators with interpreting the solutions. 2 - Operations Research Applications At A Comprehesive Cancer Center Victoria Jordan, University of Texas MD, vsjordan@mdanderson.org Industrial and Systems Engineering is an emerging field in healthcare. The University of Texas MD Anderson Cancer Center has Healthcare Systems Engineers working in the Office of Performance Improvement. This session will provide a high level overview of OR applications to improve patient flow and the patient experience, reduce costs, and improve safety. Also presented will be more clinical applications to improve delivery of care and the care itself. 3 - Factors That Predict Discharge Disposition At Admission For Veterans Nicholas Ballester, Wright State University, 207 Russ Center, 3640 Col Glenn Hwy, Dayton, OH, 45435, United States, ballester.2@wright.edu, Pratik J Parikh Discharge delays reduce inpatient quality of care and reverberate back through the health care system, tying up valuable resources needed by incoming patients. Discharge disposition, in particular, has a significant effect as different dispositions require vastly different procedures for insurance and transportation coordination. We have developed model to predict discharge disposition upon admission for veterans admitted to a general internal medicine unit. This model considers patient factors known at admission such as demographics, medical history, and living status. Preliminary findings from a trial implementation will also be discussed. SB21 107A-MCC Applications in Healthcare Sponsored: Health Applications Sponsored Session

2 - Parallel Branch And Bound Revisited Lluis-Miquel Munguia, Georgia Institute of Technology, lluis.munguia@gatech.edu, Geoffrey Malcolm Oxberry, Deepak Rajan Branch and Bound (B&B) is a widely used algorithm for solving Mixed Integer Programs (MIPs). Despite its straightforward parallelization, current B&B implementations have shown to scale inconsistently. In this talk, we propose a new decentralized and lightweight implementation of parallel Branch & Bound for PIPS-SBB, a distributed-memory parallel stochastic MIP solver. In this work, we exploit parallelism at two levels of the optimization process with the objective of increasing parallel efficiency. We present computational results to evaluate the effectiveness of our approach. 3 - Scalable Strategies Exploited by Parallel Nonlinear Solver PIPS-NLP Nai-Yuan Chiang, United Technology Research Center, chiangn@utrc.utc.com We present PIPS-NLP, a software library for the solution of large-scale structured nonconvex optimization problems on high-performance computers. We focus on linear algebra parallelization strategies and discuss how such strategies influence the choice of algorithmic frameworks, while all the proposed approaches guarantee global convergence. Small examples about using parallel solver PIPS- NLP via AMPL or StructureJuMP are given, to illustrate how to exploit the problem structures. Numerical studies from large-scale security-constrained ACOPF and line-pack dispatch in natural gas networks also demonstrate the robustness and efficiency of PIPS-NLP. SB19 106B-MCC Bilevel Programming: Methodology and Applications Sponsored: Computing Sponsored Session Chair: Juan S. Borrero, University of Pittsburgh, 1012 Benedum Hall, Pittsburgh, PA, 15261, United States, jsb81@pitt.edu Co-Chair: M. Hosein Zare, University of Pittsburgh, 1012 Benedum Hall, Pittsburgh, PA, 15260, United States, moz3@pitt.edu 1 - A Sampling-based Exact Approach For The Bilevel Mixed Integer Programming Problem Leonardo Lozano, Clemson University, llozano@g.clemson.edu Cole Smith We examine bilevel mixed integer programs which are difficult to solve because the leader feasible region is defined in part by optimality conditions governing the follower variables, which are difficult to characterize because of the nonconvexity of the follower problem. We propose an exact finite algorithm for these problems based on an adaptive sampling scheme, and demonstrate how this algorithm can be tailored to accommodate either optimistic or pessimistic assumptions on the follower behavior. Computational experiments demonstrate that our approach outperforms existing algorithms that are tailored to problems in which all functions are assumed to be linear. 2 - On Bilevel Progams With Inexact Follower M. Hosein Zare, University of Pittsburgh, moz3@pitt.edu We consider classes of bilevel programs, where the upper-level decision-maker (i.e., the leader) needs to consider the uncertain behavior of the lower-level decision maker (i.e., the follower). We derive some theoretical properties of the proposed models, and illustrate our results with numerical illustrations. 3 - Reliable Vehicle Sharing Program Network Design Ran Zhang, University of South Florida, ranzhang@mail.usf.edu, Bo Zeng This talk develops a bi-level optimization model to achieve a reliable vehicle sharing program network design. A set of numerical study will be presented to demonstrate our design model. 4 - Sequential Max-min Bilevel Programming With Incomplete Information And Learning Juan S. Borrero, University of Pittsburgh, jsb81@pitt.edu Oleg A Prokopyev, Denis R. Saure We consider an adversarial bilevel problem where the leader and follower interact repeatedly. At each period the leader implements an upper-level solution after which the follower reacts by solving the lower-level problem. The leader has incomplete information about the variables, constraints, and data of the follower’s problem, and learns about them from observing his reaction to her actions. Given that the leader’s objective is to maximize the costs the follower incurs across all periods, we study a set of greedy and robust decision policies that are able to find an optimal solution to the full-information bilevel problem in finite time periods, and moreover, are worse-case optimal.

49