Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SB14

2 - Perishable Inventory Sharing in a Two-location System Can Zhang, Duke University, Durham, NC, United States, Turgay Ayer, Chelsea C. White Motivated by a platelet inventory management problem in a local hospital network, we study an inventory sharing problem for perishable products in a two-location system. We derive structural properties of the optimal policy and present managerial insights that are significantly different from those in existing studies for traditional non-perishable products. 3 - Operational Drivers of Stock-outs in Reproductive Health Supply Chains: An Empirical Study Amir Karimi, University of Minnesota, Minneapolis, MN, 55408, United States, Karthik Natarajan, Kingshuk K. Sinha Despite the widespread prevalence of contraceptive stock-outs in practice, very little is known about the drivers of stock-outs and this forms the motivating context for our study. Utilizing field data collected in developing countries, we investigate how different factors such as a health facility’s geographic location and product variety impact the likelihood of stock-outs. In addition, we explore the effect of two commonly used inventory management practices that have the potential to serve as mitigation mechanisms to reduce the risk of stock-outs. Our findings have important implications for resource allocation and supply management of health commodities in developing countries. 4 - Optimal Motorcycle Routing in Sample Transportation for Diagnostic Networks Jonas Oddur Jonasson, MIT Sloan School of Management, 30 Memorial Drive, E62-588, Cambridge, MA, 02142, United States, Sarang Deo, Emma L. Gibson, Mphatso Kachule, Kara Palamountain Due to limited resources, diagnostic and disease monitoring services in sub- Saharan Africa are delivered through a network of clinics and laboratories. An ongoing challenge is to develop cost-effective sample transportation systems to ensure short turnaround times of results. Using data from Riders for Health in Malawi we develop an algorithm for the daily route optimization of couriers in a diagnostic network and evaluate its impact on turnaround times and distance driven. Our method maintains current service levels while requiring approximately 90,000 km less distance travelled per year. n SB14 North Bldg 126C Stochastic Models in Service Operations Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: Baris Ata, University of Chicago,Chicago, IL, 60637, United States Co-Chair: Xiaoshan Peng, Indiana University, Bloomington, IN, 47401, United States 1 - Process Flexibility for Multi Period Production Systems Yuan Zhong, University of Chicago / Booth School of Business, 5807 S. Woodlawn Avenue, Chicago, IL, 60637, United States, Cong Shi, Yehua Wei We consider process flexibility in a multi-period make-to-order production (MTO) system. First, using a new chaining notion, termed the Generalized Chaining Gap (GCG), we prove that in a general system with high utilization, a sparse flexibility structure with m+n arcs is needed to achieve similar performance as full flexibility, where m and n are the number of plants and of products, respectively. We also provide a simple and efficient algorithm for finding such sparse structures. Moreover, we show that the requirement of m+n arcs is necessary, as for some systems, even the best flexibility structure with m+n-1 arcs cannot achieve the same asymptotic performance as full flexibility. 2 - On the Control of Polling System with Large Switchover Times Yue Hu, Columbia University, c/o Clara Magram, 4th Floor West, New York, NY, 10027, United States, Jing Dong, Ohad Perry We study the control problem of polling systems with large switchover times. Under the proper scaling and cyclic controls, we show that the exhaustive switching policy is asymptotically optimal in minimizing the long-run average holding cost among all switching policies that achieve certain long-run regularity. To establish the asymptotic optimality, we characterize the fluid scaling limit of the system as a hybrid dynamical system and prove the corresponding interchange-of-limit results. We also study properties of several classes of controls of interests.

2 - Recent Advances in Algorithm Configuration and Algorithm Tuning Meinolf Sellmann, General Electric, 82 Diamond Avenue, Cortlandt Manor, NY, 10567, United States We present new results on highly efficient parallel algorithm configuration as well as algorithm portfolios. Our deterministic parallelization of the GGA tuner is shown to achieve near-linear speed-ups up to 96 cores. Moreover, we show that a self-configured CSHC portfolio builder with a new recourse mechanism sets a new state of the art in solver selection. 3 - Learning to Search via Retrospective Imitation Jialin Song, California Institute of Technology, 1200 E. California Blvd, MC 305-16, Pasadena, CA, 91125, United States, Ravi Lanka, Albert Zhao, Yisong Yue, Masahiro Ono We study the general problem of learning a good search policy. To do so, we propose the retrospective imitation setting, which builds upon imitation learning in two ways. First, retrospective imitation uses feedback provided by retrospective analysis of search traces. Second, the policy can learn from its own decisions and mistakes without requiring repeated feedback from an external expert. Combined, these two properties allow our approach to iteratively scale up to larger problem sizes than the initial problem size for which expert demonstrations were provided. We showcase the effectiveness of our approach on learning node selection and pruning policies in branch-and-bound. 4 - Learning-Theoretic Foundations of Algorithm Configuration for Combinatorial Partitioning Problems Ellen Vitercik, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA, 15213, United States, Maria-Florina Balcan, Vaishnavh Nagarajan, Colin White Data-driven algorithm design, that is, choosing the best algorithm for a specific application, is a crucial problem in modern data science. Practitioners often optimize over parameterized algorithm families, tuning parameters based on problems from their domain. While effective in practice, these procedures generally have not come with provable guarantees. I will present work that helps put data-driven combinatorial algorithm configuration on firm foundations. We provide strong computational and statistical performance guarantees for several important problems, including clustering algorithm configuration and integer quadratic programming approximation algorithm configuration. 5 - Learning to Branch Tuomas W. Sandholm, Angel Jordan Professor of Computer Science, Carnegie Mellon University, Gates Center for Computer Science, Pittsburgh, PA, 15213, United States, Maria Florina Balcan, Travis Dick, Ellen Vitercik Tree search algorithms, e.g. branch-and-bound, are the most widely used tools for combinatorial and nonconvex problems (e.g., MIP and CSP). To keep the tree small, it is crucial to decide, when expanding a node, which question (e.g., variable) to branch on to partition the remaining space. Many techniques have been proposed, but theory was lacking. We show how to use machine learning to determine an optimal weighting of any set of partitioning procedures for the instance distribution at hand using samples from the distribution. We provide the first sample complexity guarantees for tree search algorithm configuration. The theory gives rise to a learning algorithm which dramatically reduces tree size. Nonprofit and For-profit Operations in Healthcare Sponsored: Manufacturing & Service Oper Mgmt/Healthcare Operations Sponsored Session Chair: Lauren Xiaoyuan Lu, University of North Carolina at Chapel Hill - Kenan Flagler, Chapel Hill, NC, 27599, United States 1 - Does Ownership Conversion from Nonprofit to For-profit Benefit the Public? Evidence from U.S. Nursing Homes Lauren Xiaoyuan Lu, University of North Carolina at Chapel Hill - Kenan Flagler, Kenan-Flagler Business School, CB #3490, Mccoll Building 4701, Chapel Hill, NC, 27599, United States, Susan F. Lu In the last few decades, many healthcare institutions converted their ownership from nonprofit to for-profit, contributing to an increased presence of for-profit ownership in the U.S. healthcare sector. There have been contradicting views on whether such ownership conversions benefit the public. Employing a large panel dataset of U.S. nursing homes dated from 2006 to 2015, we conduct a difference- in-differences analysis on converting facilities’ financial performance, operating policies, and service quality. n SB13 North Bldg 126B

35