2015 Informs Annual Meeting

SA21

INFORMS Philadelphia – 2015

2 - Event-driven Predictive Models for Socio-economic Indicators Lakshminarayana Subramanian, Associate Professor, NY, NY, United States of America, lakshmi@cs.nyu.edu, Sunandan Chakraborty I will describe how to extract real-world events using unstructured news streams to understand their impact on the volatility macro-economic indicators. The hypothesis is that the factors triggering sudden fluctuations in such indicators can be characterized by events. Given a news corpus, we describe how to build event- driven predictive models that can potentially predict fluctuations in specific indicators. We describe specific results about what triggers fluctuations in food prices in India. 3 - Efficient Coflow Scheduling in Data Center Networks Yuan Zhong, Columbia University, 500 W. 120th Street, In this talk, we consider the efficient scheduling of coflows - an abstraction introduced in [Chowdhury and Stoica 2012] to capture communication patterns of large-scale data center jobs. We introduce the problem of minimizing the total weighted coflow completion times, show that it is strongly NP-hard, and develop the first polynomial-time approximation algorithms for this problem. We also evaluate the practical performances of a variety of algorithms through numerical experiments. 4 - Optimizing for Tail Response Times of Cloud Clusters Lydia Chen, IBM Zurich, yic@zurich.ibm.com, Natarajan Gautam Motivated by the volatile system dynamics in cloud cluster, we develop an approximation scheme that can capture the high performance variability caused by neighboring VMs, especially in terms of tail response times. The approximation of tail response times is based on the large deviation analysis. We evaluate the proposed analysis on simulation as well a wiki prototype cluster in the cloud. SA21 21-Franklin 11, Marriott Stochastic Models for Medical Decision Making and Healthcare Delivery Sponsor: Health Applications Sponsored Session Chair: F. Safa Erenay, Assistant Professor, University of Waterloo, 200 University Ave. CPH 4323, Waterloo, Canada, ferenay@uwaterloo.ca 1 - The Impact of Optimization on the Allocation of Livers for Organ Transplantation Mustafa Akan, Associate Professor, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, akan@andrew.cmu.edu, James Markmann, Heidi Yeh, Zachary Leung, Sridhar Tayur Patients on the waitlist for liver transplantation are prioritized according to their MELD scores, which reflects the severity of liver disease. Recent studies have shown that hepatocellular carcinoma (HCC) patients have significantly higher liver transplant rates than non-HCC patients. We recommend a family of alternative MELD score policies based on a fluid model approximation of the queueing system and an optimization model that achieves an optimal balance between efficiency and equity. 2 - Physician Staffing in the Emergency Department: Opening the Blackbox Caglar Caglayan, Georgia Institute of Technology, Atlanta, GA, United States of America, ccaglayan6@gatech.edu, Kalyan Pasupathy, David Nestler, Mustafa Sir, Thomas Hellmich, Turgay Ayer, Gomathi Marisamy, Thomas Roh We propose an “intuitive”, “realistic” and “tractable” model of the emergency department (ED) by a multi-class multi-stage queuing network with multiple targeted service levels. Based on infinite-server approximation and offered load analysis, we employ square-root safety principle to determine the right number of physicians in the ED. Our model is detailed enough to capture the key dynamics of the ED but simple enough to understand, infer results and implement in a clinical setting. 3 - Deriving Better Strategies for Influenza Vaccines Allocation F. Safa Erenay, Assistant Professor, University of Waterloo, 200 University Ave. CPH 4323, Waterloo, Canada, ferenay@uwaterloo.ca, Osman Ozaltin, Onur Ozden Dalgic In influenza pandemics, available vaccines are allocated considering individual risk profiles of the patients. Using a network based stochastic simulation model and mesh adaptive direct search, we derived effective age-specific vaccine allocation strategies for cost, health outcomes, and equity metrics. In most scenarios, the proposed method outperforms the current guidelines and policies developed based on deterministic compartmental models. New York, NY, 10027, United States of America, yz2561@columbia.edu, Cliff Stein, Zhen Qiu

4 - Reliable Facility Location Model for Disaster Response Abdelhalim Hiassat, PhD Student, University of Waterloo, ahhiassa@uwaterloo.ca, Osman Ozaltin, F. Safa Erenay We formulate a reliable facility location model for disaster response, and consider the problem of minimizing expected service cost. Candidate facility locations might become unavailable after the disaster, and victims patronize relief facilities based on their preferences. We propose a Lagrangian-decomposition-based branch-and-bound method for this problem. Our computational results show the efficiency of the solution approach and the significance of incorporating preferences into the model. SA22 22-Franklin 12, Marriott Matching Markets Sponsor: Applied Probability Sponsored Session Chair: Itai Ashlagi, MIT, 100 Main St, Cambridge, MA, 02139, United States of America, iashlagi@mit.edu 1 - Welfare-sensitive Assortment Optimization: An Application to School Choice Peng Shi, MIT Operations Research Center, 1 Amherst Street, E40-149, Cambridge, MA, 02139, United States of America, pengshi@mit.edu In many settings, a planner gives a set of options to agents who choose among them to maximize their own value, but agents’ choices have externalities on system revenue/cost. Examples include school choice, public housing, and health insurance. Welfare-Sensitive Assortment Optimization is to find a set of options that maximize the sum of agents’ values and system revenue. We give efficient algorithms under MNL utilities and various constraints, and apply this to improve school choice in Boston. 2 - Near Feasible Stable Matchings with Couples The National Resident Matching program strives for a stable matching of medical students to teaching hospitals. With the presence of couples, stable matchings need not exist. For any student preferences, we show that each instance of a matching problem has a `nearby’ instance with a stable matching. The nearby instance is obtained by perturbing the capacities of the hospitals. 3 - Matching with Externalities Jacob Leshno, Columbia University, 3022 Broadway, Uris Hall, 406, New York, NY, 10027, United States of America, jleshno@columbia.edu We show existence of stable matching in markets with a continuum of students. Stable matchings are characterized as rational expectations market clearing cutoffs. 4 - What Matters in Tie-breaking Rules? How Competition Guides Design Afshin Nikzad, Stanford University, 37 Angell Court, APT 116, Stanford, Ca, 94305, United States of America, afshin.nikzad@gmail.com, Assaf Romm, Itai Ashlagi School districts that adopt the Deferred Acceptance (DA) mechanism to assign students to schools face the tradeoff between fairness and efficiency when selecting how to break ties among equivalent students. We analyze a model with with random generated preferences for students and compare two mechanisms differing by their tie-breaking rules: DA with one single lottery (DA-STB) and DA with a separate lottery for each school (DA-MTB). We identify that the balance between supply and demand in the market is a prominent factor when selecting a tie-breaking rule. When there is a surplus of seats, we show that neither random assignments under these mechanisms stochastically dominates each other, and, the variance of student’s assignments is larger under DA-STB. However, we show that there is essentially no tradeoff between fairness and efficiency when there is a shortage of seats: not only that DA-STB (almost) stochastically dominates DA- MTB, it also results in a smaller variance in student’s rankings. We further find that under DA-MTB many pairs of students would benefit from directly exchanging assignments ex post when there is a shortage of seats, while only few such pairs exist when there is a surplus of seats. Our findings suggest that it is more desirable that ``popular” schools use a single lottery over a separate lottery in order to break ties, while in other schools there is a real tradeoff. Thanh Nguyen, Krannert School of Management, Purdue University, West Lafayette, IN, United States of America, nguye161@purdue.edu, Rakesh Vohra

44