2015 Informs Annual Meeting

WD19

INFORMS Philadelphia – 2015

WD19 19-Franklin 9, Marriott Core Algorithms and Techniques for Computational Optimization Sponsor: Computing Society Sponsored Session Chair: Atharv Bhosekar, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, Pe, 15213, United States of America, abhoseka@andrew.cmu.edu 1 - Update Algorithms for the Roundoff-error-free LU and Cholesky Factorizations Adolfo Escobedo, PhD Candidate, Texas A&M, 3131, TAMU, We introduce efficient update algorithms for the Roundoff-error-free (REF) LU and Cholesky factorizations. The updates are addition, deletion, and replacement of rows and columns of a basis. Combined with REF substitution, the featured algorithms provide a complete framework for solving LPs exactly and efficiently. A significant advantage of the REF LP framework is that the length of any coefficient calculated via its algorithms is bounded polynomially without having to use gcd operations. 2 - Status Update on GCG, a Generic Branch-price-and-cut Solver GCG is a source-open extension to SCIP to solve mixed-integer programs (MIPs) via branch-price-and-cut without any user interaction. It is meant as a research vehicle and “another trick in the bag” to solve MIPs. For specially structured MIPs, GCG (not surprisingly) outperforms standard MIP solvers by far while it (also not surprisingly) colossally fails on other instances. We report on the project’s status, recent novelties, future plans, and potential benefits for a general audience. 3 - Integer Programming as Projection John Hooker, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, jh38@andrew.cmu.edu, H. P. Williams We develop an alternative theory of integer programming (IP) based on projection. We define valid inequalities that are analogous to Chvatal-Gomory cuts but based on congruences rather than rounding and have bounded rank. We show how to solve a general IP by branching solely on finite-domain auxiliary variables. We define an IP dual as a value function that is obtained by nested rounding of bounded depth and becomes shift periodic for large perturbations. 4 - Computational Experience with the Bam Global Optimization Algorithm for Derivative-Free Optimization College Station, TX, 77843, United States of America, adolfoescobedo@tamu.edu, Erick Moreno-centeno Marco Luebbecke, Aachen University, Germany, marco.luebbecke@rwth-aachen.de, Jonas Witt, Michael Bastubbe, Christian Puchert abhoseka@andrew.cmu.edu, Luis Miguel Rios, Nikolaos Sahinidis Branch-and-model (BAM) is a derivative-free optimization (DFO) algorithm that builds a model around each evaluated point using the surrounding evaluated points. The algorithm performs a dense search and employs a local search algorithm to identify local optima. We present extensive computational experience with the algorithm and comparisons with other DFO algorithms. WD21 21-Franklin 11, Marriott Operations Research Methodologies to Improve Chair: Nazanin Esmaili, PhD Candidate, University of Pittsburgh, 1048 Benedum Hall, Pittsburgh, PA, United States of America, nae22@pitt.edu 1 - Impact of Care Discontinuity on Patients’ Length of Stay Kimia Ghobadi, Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, MA, United States of America, kimiag@mit.edu, Retsef Levi, Ana Cecilia Zenteno Langle, Andrew Johnston The Department of Medicine in Massachusetts General Hospital serves a wide variety of patients with various care levels. This functional heterogeneity has led to distributed care models which combined with consistent growth in demand has resulted in a congested system. In this talk, we explore contributors to clinically unnecessary delays in patient progression through the hospital, and more Atharv Bhosekar, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, Healthcare Operations Sponsor: Health Applications Sponsored Session

specifically, the impact of care discontinuity during team handovers on patients’ total length of stay. 2 - Appointment Scheduling Based Upon Continuously and Periodically Reported Data Zlatana Nenova, University of Pittsburgh, 241 Mervis Hall, Roberto Clemente Drive, Pittsburgh, PA, 15260, United States of America, zdn3@pitt.edu, Jerrold H. May We discuss an approach to determining the optimal timing of appointments for diabetic patients, based upon their blood glucose, blood pressure, and cholesterol readings. Blood pressure and cholesterol are reported periodically; blood glucose is reported multiple times per day. The approach is illustrated using examples from the VA Health System. 3 - Simulation of a Phlebotomy Station in an Outpatient Chemotherapy Infusion Clinic Matthew Rouhana, University of Michigan, 1205 Beal Avenue, Ann Arbor, MI, United States of America, mrouhana@umich.edu, Amy Cohn, Pamela Martinez Villarreal, Marian Grace Boxer, Carolina Typaldos We study the organization and operation of a phlebotomy station at the University of Michigan Comprehensive Cancer Center. By developing a simulation of the patient and work flow through the station, we evaluate alternative methods to reduce patient wait time and improve full-day patient experiences. For educational purposes, we create a simplified table-top version to demonstrate the underlying probability theory for hospital leadership. 4 - Hospital Nurse Staffing Improvements through Better Prediction of Surgical Case Volume Nazanin Zinouri, Clemson University, 130 Freeman Hall, Clemson, SC, United States of America, nzinour@g.clemson.edu, Kevin Taaffe Accurate approximations of final surgical case volume are difficult due to the high surgical demand variations. Staffing adjustments a few days prior to the day of surgery are difficult and inefficient. Having accurate demand prediction weeks in advance would help improve staff scheduling and decrease nurse workload pressure by allowing more flexibility in the schedule. We have studied current staff scheduling and workload prediction methods at a hospital to identify areas for improvement. 5 - Hybrid Inventory Policies for Hospitals with Shelf Space Restriction Nazanin Esmaili, PhD Candidate, University of Pittsburgh, 1048 Benedum Hall, Pittsburgh, PA, United States of America, nae22@pitt.edu, Bryan Norman, Jayant Rajgopal We address the management of inventory for multiple non-perishable routine use items in a healthcare setting. We present a mixed-integer programming model for selecting the best inventory control system for each item and the best shelf configuration considering space, shelf, and inventory control policy constraints. The objective is to minimize the total average effort to monitor, manage and replenish items. We illustrate the model with data from a hospital. Chair: Toshikazu Aiyama, Professor, Tokyo Metropolitan University, 1-1 MinamiOhsawa, HachiOhji, Japan, tyaiyama@yahoo.com 1 - A Multi-Resolution Stochastic Graphene Growth Kinetics Model Sobambo Sosina, Harvard University, 1 Oxford Street, 7th Floor Science Center, Cambridge, MA, 02138, United States of America, sosina@fas.harvard.edu, Tirthankar Dasgupta, Qiang Huang Graphene has many important properties and has been identified as key component in many industrial applications. It is thus crucial that its growth process and the kinetics behind that process be thoroughly understood. Extending work done by collaborators, we derive a stochastic model for multi-resolution observations, which correctly quantifies uncertainty in the kinetics. 2 - Queueing Network Approximations with MAP(3)s Sunkyo Kim, Professor, Ajou University, 206 World Cup Road, Youngtong Gu, Suwon, 16499, Korea, Republic of, sunkyo@ajou.ac.kr We propose a minimal representation of MAP(3)s and present an exact moment matching method. The marginal and joint LST of stationary intervals of the MAP(3) is given in terms of nine parameters. As for queueing network applications, arrival and departure processes are approximated as a MAP(3) based on nine moments. We show that the MAP(3) approximation performs better than MAP(2) especially under moderate traffic intensities. WD22 22-Franklin 12, Marriott Stochastic Processes I Contributed Session

460