2015 Informs Annual Meeting

TA13

INFORMS Philadelphia – 2015

4 - Higher Rank-Order Semidefinite Cutting Planes for Nonconvex QCQPs Xiaofei Qi, PhD Student, Nanyang Technological University, 50 Nanyang Ave, Singapore, Singapore, xqi001@e.ntu.edu.sg, Jitamitra Desai, Rupaj Nayak We introduce a polynomial-time scheme to generate higher rank-order semidefinite cutting planes that serve to tighten convex relaxations of nonconvex quadratically constrained quadratic programs (QCQPs) and significantly improve lower bounds. Suitably defined row-and-column based operations are used to speed up the process of generating these cuts, and computational comparisons across different types of relaxations shows the efficacy of these new cutting plane strategies. TA13 13-Franklin 3, Marriott Optimizing Sharing Service/Economy Under Uncertainty Sponsor: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Siqian Shen, Assistant Professor, University of Michigan, 1205 Beal Avenue, Ann Arbor, MI, 48105, United States of America, siqian@umich.edu 1 - Optimal Location Design of Carsharing Fleet under Uncertain One-way and Round-trip Demands Zhihao Chen, czhihao@umich.edu, Siqian Shen We allocate vehicles in a homogeneous carshare fleet to contracted locations, to maximize the expected revenue from random demand for one-way and round trip rentals. We use a spatial-temporal network and optimize both risk-neutral and CVaR-based risk-averse two-stage stochastic programs with high demand satisfaction rates. The two-stage problems are solved via branch-and-cut with mixed integer rounding and we give insights on carsharing location design from data reported by Zipcar in Boston. 2 - Online Resource Allocation with Limited Flexibility Xuan Wang, New York University, 44 West 4th Street, Suite 8-154, New York, NY, 10012, United States of America, xwang3@stern.nyu.edu, Jiawei Zhang, Arash Asadpour We consider a general class of online resource allocation problems with limited flexibility, where a type j request can be fulfilled by resource j or resource j+1, and we call this limited flexibility the long chain pattern. The long chain has been studied in process flexibility and has been shown to be very effective in coping with demand uncertainty under offline arrivals. We provide preliminary results that show the effectiveness of the long chain when the arrivals are online. 3 - On-demand Staffing: Incentive Wage Contracts with Guaranteed Fill Rates Zhichao Zheng, Singapore Management University, Lee Kong China School of Business, 50 Stamford Road, Singapore, 178899, Singapore, danielzheng@smu.edu.sg, Tao Lu, Yuanguang Zhong We study the on-demand economy and its impact on labor market efficiency. We consider n employers with uncertain and time-varying demands, and a platform operator providing on-demand staffing services. We propose a novel fill rate- based allocation policy enabling the on-demand workforce to be shared efficiently among employers. We propose a form of incentive contracts based on fill rate guarantees, and show that our contracts can induce the system-wise optimality in decentralized systems.

2 - Near Optimal Ambiguity Sets in Distributionally Robust Optimzation

Vishal Gupta, Assistant Professor, USC Marshall School of Business, 3670 Trousdale Parkway, Bridge Hall 401 G, Los Angeles, CA, 90089-0809, United States of America, guptavis@usc.edu

We assess the strengths of data-driven ambiguity sets in distributionally robust optimization (DRO) by bounding the relative size of a candidate set to a specific, asymptotically optimal set. We find popular ambiguity sets are much larger than this asymptotically optimal set, suggesting current DRO models are overly conservative. We propose new “near-optimal” sets that are only a constant factor larger than the optimal set and satisfy the usual robustness properties. 3 - A Time Based Choice Model Tauhid Zaman, MIT Sloan School of Management, 50 Memorial Drive, Cambridge, MA, 02139, United States of America, zlisto@mit.edu We present a choice model which incorporates the time it takes the user to make a decision. Our model assumes that the further apart two items are in terms of user preference, the faster a decision is made. We conduct a set of online polls and find that this model captures actual human behavior. We also show that using this time based choice model can learn user preferences with high accuracy than standard choice models for a fixed sample size. TA15 15-Franklin 5, Marriott Patient Scheduling in Health Care Sponsor: Optimization in Healthcare Sponsored Session Chair: Joseph Milner, Associate Professor Of Operations Management, Rotman School of Management, University of Toronto, 105 St.George Street, Toronto, ON, M5S3E6, Canada, Joseph.Milner@Rotman.Utoronto.Ca 1 - Dynamic Patient Scheduling for Multi-Appointment Health Care Programs Adam Diamant, Assistant Professor Of Operations Management, Schulich School of Business, York University, 111 Ian Macdonald Boulevard, Toronto, ON, M3J1P3, Canada, adiamant@schulich.yorku.ca, Fayez Quereshy, Joseph Milner We investigate the scheduling practices of a multidisciplinary, multistage, outpatient health care program with no-shows. We formulate the problem as a Markov Decision Process and use approximate dynamic programming to find policies to schedule patients to appointments. We examine the quality of our solutions via structural results and compare them to a simulation of the clinic. Our results applied to the operation of a bariatric surgery program at a large tertiary hospital in Toronto, Canada. 2 - Flexible Hospital-wide Patient Scheduling Daniel Gartner, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, dgartner@andrew.cmu.edu, Rema Padman We study a patient scheduling problem with admission decisions, clinical pathways, day and overnight hospital resources, ward and surgical team assignment flexibility, and overtime considerations. We model the problem using Mixed-Integer Programming and embed it in a rolling horizon planning to take into account uncertain recovery times of and remaining resource capacity for patients. We analyze the impact of flexibility and uncertainty on several metrics. 3 - Coordinated Scheduling for a Multi-station Healthcare Network Ester Dongyang Wang, PhD Candidate, University of Texas, IROM Dept., Austin, TX, United States of America, wdy@utexas.edu, Douglas Morrice, Kumar Muthuraman As the population ages, our healthcare industry must face the challenge of increasing demand for care under constrained budget and resources. Our research focuses on one of the central factors to the success of healthcare reform–outpatient appointment scheduling. We develop a mechanism that coordinates appointment scheduling among multiple services in a healthcare network to improve access of care and reduce patient no-show rate. Our Mohamad Soltani, University of Alberta, PhD Office, Business Building,, University of Alberta, Edmonton, AB, T6G 2R3, Canada, soltani@ualberta.ca, Michele Samorani It is commonly believed that clinics that schedule appointments have lower patients’ waiting time and providers’ overtime than clinics that only allow walk- ins. However, if we consider patient unpunctuality, walk-in-only clinics may achieve a higher performance. In this research, we investigate the conditions under which each strategy is preferable. approach has the potential to yield a global optimal solution. 4 - Appointment Scheduling and Walk-in Strategies with Unpunctual Patients

TA14 14-Franklin 4, Marriott Data-driven Optimization Sponsor: Optimization/Optimization Under Uncertainty Sponsored Session

Chair: Gah-Yi Vahn, Assistant Professor, London Business School, Sussex Place, Regent’s Park, London, NW1 4SA, United Kingdom, gvahn@london.edu 1 - Data-driven Estimation of (s, S) Policy Gah-Yi Vahn, Assistant Professor, London Business School, Sussex Place, Regent’s Park, London, NW1 4SA, United Kingdom, gvahn@london.edu I derive a tractable algorithm for computing the optimal (s,S) policy when the decision maker has access to historical demand data. I show that this scheme yields asymptotically optimal (s, S) policy and derive analytical characterisations of confidence intervals, which is useful for operational decision-making.

262