Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WA38

optimality equation in continuous space and time. In cases where the ``Taylor-ed’’ equation is tractable, it provides a useful modeling tool coupled with explicit optimality-gap bounds. Computationally, our framework leads to an ``aggregation’’ approach with performance guarantees. While the guarantees are grounded in PDE theory, the practical use of this approach requires no knowledge of such theory. 2 - Interpretable Optimal Stopping Optimal stopping is the problem of deciding when to stop a stochastic system to obtain the greatest reward. State of the art methods for this problem rely on obtaining an approximate value function for use within a greedy policy. However, such policies are generally not interpretable, in that it is often difficult to see how the policy maps each system state to a decision.We propose a new data-driven method for optimal stopping that directly learns interpretable policies in the form of binary trees. We apply our approach to a canonical option pricing problem and show that it quickly obtains interpretable policies that outperform existing non- Based on a dynamic model of the stochastic repayment behavior exhibited by delinquent credit-card accounts in the form of a self-exciting point process, a bank can control the arrival intensity of repayments using costly account- treatment actions. A semi-analytic solution to the corresponding stochastic optimal control problem is obtained. For a linear cost of treatment effort, the optimal policy in the two-dimensional (intensity, balance)-space is described by the frontier of a convex action region. The unique optimal policy significantly reduces a bank’s loss given default and concentrates the collection effort onto the best possible actions at the best possible times. 4 - Dynamic Pricing of Scarce and Relocating Resources in Large Networks Chen Chen, Duke University, Santiago Balseiro, David Brown We study dynamic pricing of resources distributed over a network of locations (e.g., shared vehicle systems). Customers with private willingness-to-pay sequentially request to relocate one resource. We focus on networks with a hub- and-spokes structure. We develop a dynamic pricing policy based on a Lagrangian decomposition and show that the policy is asymptotically optimal as the number of spokes grows large but the number of resources per location remains fixed. n WA38 North Bldg 225B Dynamic Queueing Control Sponsored: Applied Probability Sponsored Session Chair: Mark E. Lewis, Cornell University, Ithaca, NY, 14853, United States 1 - Dynamic Control of Running Servers Douglas Down, McMaster University, Department of Computing and Software, 1280 Main Street West, Hamilton, ON, L8S 4K1, Canada, Esa Hyytia, Pasi Lassila, Samuli Aalto Motivated by a data center setting, we study the problem of joint dispatching and server sleep state control in a system consisting of two queues in parallel. Using the theory of Markov decision processes and a novel lookahead approach, we explicitly determine near-optimal control policies that minimize a combination of QoE costs, energy costs, and wear and tear costs due to switching. Guidelines are provided as to when these combined policies are most effective. 2 - Server Collaboration in Queueing Systems: When and How? Junqi Hu, Georgia Institute of Technology, Atlanta, GA, United States, Sigrun Andradottir, Hayriye Ayhan Consider a form of server collaboration where each job is decomposed into subtasks, and the job is finished when all its subtasks are completed. We identify the task assignment policy that maximizes the long-run average throughput for one station when servers are either static, flexible, or collaborative and internal buffers are either finite or infinite. Then we compare this model with other forms of server collaboration to determine when and how servers should collaborate. Finally, we investigate server collaboration for longer lines, and provide numerical results for two tandem stations. Velibor Misic, UCLA Anderson School of Management, 110 Westwood Plaza, Suite B406, Los Angeles, CA, 90095, United States, Florin Ciocan interpretable policies based on simulation-regression. 3 - Dynamic Credit-collections Optimization Naveed Chehrazi, McCombs School of Business, 2110 Speedway, Stop B6000, Austin, TX, 78712, United States, Peter W. Glynn, Thomas A. Weber

n WA36 North Bldg 224B Joint Session Drones/Practice Curated: Drone Applications in Healthcare Emerging Topic: Robotics, Drones and Autonomous Vehicles in Logistics Emerging Topic Session Chair: Justin James Boutilier, University of Toronto, Toronto, ON, M6G 2P5, Canada 1 - Response Time Optimization with Queuing Constraints for Drone Delivered Defibrillators Justin James Boutilier, University of Toronto, Toronto, ON, M6G 2P5, Canada, Timothy Chan Out-of-hospital cardiac arrest (OHCA) is a time-sensitive medical emergency claiming over 400,000 lives each year and an automated external defibrillator (AED) is one of the only effective methods for treating OHCA. In this paper, we propose an integrated location-queuing model to minimize the total number of drones required to meet an average response time goal while guaranteeing a sufficient number of drones are located at each base. We develop a novel reformulation technique that allows us to solve real-world instances and determine the optimal deployment of AED-enabled drone resources using eight years of data covering 26,000 square kilometers around Toronto, Canada. 2 - Drone-aided Healthcare Services for Patients with Chronic Diseases in Rural Areas Gino Lim, University of Houston, Professor and Chairman, Dept of Industrial Engineering, Houston, TX, 77204, United States, Seon Jin Kim We present the drone-aided delivery and pickup planning of medication and test kits for patients with chronic diseases in rural areas. A two-stage decision model is proposed to find the optimal number of drone center locations and schedule pickup and delivery in which drones deliver medicine to patients and pick up exam kits on the way back such as blood and urine samples. A preprocessing algorithm, a Partition method, and a Lagrangian Relaxation (LR) method are developed to solve the model. A cost-benefit analysis method is developed as a tool to analyze the benefits of drone-aided healthcare service. 3 - Cure for Healthcare Using Drone Technology in Hospital Processes an Explorative Analysis Mike Krey, Zurich University of Applied Sciences, Winterthur, Switzerland The paper at hand aims at outlining the potential and possible fields of application using drone technology inside the Swiss hospital environment. As part of a survey, face-to-face interviews with IT executives of four Swiss hospitals were conducted on the status quo of drone technology providing valuable insights about the current situation and the possible fields of application. The study revealed that different priorities, lack of motivation, and budget constraints were mentioned as the three major obstacles to improving the situation. 4 - A Framework for Using Drones after a Tornado Sean Grogan, PhD Candidate, École Polytechnique de Montréal, 2500 Edouard Monpetit, 2900 Chemin Polytechnique Office C314.6, Montreal, QC, H3T 1J4, Canada, Robert Pellerin, Michel Gamache Recent innovations and advancements in unmanned aerial vehicles (UAVs) have increased their utility while reducing the cost of the devices. UAVs in the aftermath of a disaster have an obvious utility where they can fly over a search area rather than be subject to potential road blocks and other impediments. To allocate the UAVs in the aftermath of a tornado has the potential to incorporate weather information and predictions from the National Weather Service, sighting information from spotters and the media, and geographic information system information to allocate UAV resources and, in turn, rescuer resources. n WA37 North Bldg 225A Advances in Approximate Dynamic Programming Sponsored: Applied Probability Sponsored Session Chair: David Brown, Duke University, Durham, NC, 27708, United States 1 - On the Taylor Expansion of Value Functions Itai Gurvich, Cornell University, 2 W. Loop Rd, New York, NY, 10044, United States, Anton Braverman, Junfei Huang We introduce a framework for approximate dynamic programming that we apply to discrete time and space chains. Simply put, our approach stipulates applying Taylor expansion to the value function in the Bellman equation to obtain an

423