2015 Informs Annual Meeting

SC22

INFORMS Philadelphia – 2015

SC21 21-Franklin 11, Marriott Applied Operations in Health Services: Research by Bonder Scholars Sponsor: Health Applications Sponsored Session Chair: Jonas Jonasson, Student, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, jjonasson@london.edu 1 - Optimal Liver Cancer Surveillance in Hepatitis C-infected Populations Qiushi Chen, chenqiushi0812@gatech.edu, Turgay Ayer, Jagpreet Chhatwal Every 6-month surveillance for liver cancer is currently recommended in cirrhotic hepatitis C patients, but the optimal surveillance policy remains unknown. We develop a mixed-integer programming-based framework to analyze the most cost- effective surveillance policies, and find that (1) the optimal surveillance interval should depend on patients’ stage of hepatitis C and age, and (2) expanding surveillance to earlier stage of hepatitis C improves the cost-effectiveness of HCC surveillance. 2 - Ambulance Dispatching, Redeployment and Reallocation Amir Ali Nasrollah Zadeh, Clemson University, Freeman Hall 129, Clemson University, Clemson, SC, 29634, United States of America, snasrol@g.clemson.edu, Amin Khademi, Cem Saydam, Hari Rajagopalan, Maria Mayorga Larger cities, expensive medical cares and heavy traffics have led to an increasing number of medical emergency calls and associated costs. In this work we develop an optimization model to find near-optimal solutions to ambulance dispatching, redeployment and reallocation problem to minimize the total expected waiting time of patients. We use approximate dynamic programming to find high-quality solutions and compare our policies with current practices via a simulation model. 3 - Robust Post-donation Blood Screening under Uncertainty Hadi El-amine, PhD Student, Virginia Tech, 250 Durham Hall Perry St., Blacksburg, VA, 24061-0118, United States of America, hadi@vt.edu, Douglas Bish, Ebru Bish Blood product safety, in terms of being free of transfusion-transmittable infections, is crucial. Under uncertainties in testing parameters and prevalence rates, various objective functions were considered in order to determine a “robust” post-donation blood screening strategy that minimizes the risk of releasing an infected unit of blood into the blood supply. SC22 22-Franklin 12, Marriott Adaptive Sampling and Selection in Simulation, Medicine and Machine Learning Sponsor: Applied Probability Sponsored Session Chair: Peter Frazier, Assistant Professor, Cornell University, 232 Rhodes Hall, Cornell University, Ithaca, NY, 14850, United States of America, pf98@cornell.edu 1 - Large Scale Parallel Ranking and Selection Susan Hunter, Purdue University, Grissom Hall, 315 N. Grant St., West Lafayette, IN, United States of America, susanhunter@purdue.edu, Florin Ciocan, Eric Ni, Shane Henderson We discuss a new ranking and selection procedure that provides a good-selection guarantee and is suitable for parallel implementation on large scale problems. Two implementations, one using message-passing interface (MPI) and the other using MapReduce, both perform well in a high-performance computing environment. MPI is more efficient than MapReduce in the sense that cores are more heavily utilized, but less robust to issues such as core failure that may arise in cloud computing environments. 2 - Active Learning for Conjoint Analysis Stephen Pallone, Cornell University, 290 Rhodes Hall, Cornell Conjoint analysis is a method of learning a user’s preferences, where the user is offered a set of alternatives and chooses the preferred option. We model the user’s preferences as being characterized by a linear classifier, but allow for noise to contaminate responses. The rate at which we learn depends on the alternatives offered. We explore different policies for offering these alternatives, such as knowledge gradient and greedy posterior entropy reduction, and analyze their performance. University, Ithaca, NY, 14853, United States of America, snp32@cornell.edu, Peter Frazier, Shane Henderson

4 - An Exact Algorithm for the Pickup and Delivery Problem with Time Windows and Scheduled Lines Veaceslav Ghilas, PhD Student, Eindhoven University of Technology, Den Dolech 2, Eindhoven, 5612 AZ, Netherlands, v.ghilas@tue.nl, Jean - Francois Cordeau, Tom Van Woensel, Emrah Demir The Pickup and Delivery Problem with Time Windows and Scheduled Lines (PDPTW-SL) concerns routing and scheduling a set of vehicles to serve a set of freight requests such that a part of the journey can be carried out on a scheduled public transportation line. We propose a branch-and-price algorithm for the PDPTW-SL. A path-based set partitioning formulation is used as master problem, and a variant of elementary shortest path problem with resource constraints is studied as pricing problem. 5 - Signal Timing Detection Based on “Pseudo Vehicle Trajectory” on the Spatial-temporal Map Seyedamirali Mostafavizadeh, Graduate Research Assistant, Rutgers University, Core 736, 96 Frelinghuysen Road, This study introduces a new CCTV video based traffic signal timing detection method. A spatial-temporal (ST) map is generated by stacking the raw pixels along scan lines defined at the middle of each lane. Vehicles leave time-accurate but space-distorted “pseudo trajectories” on ST map without the need of intensive camera calibration. A moving Hough Line Transforms (MHLT) based method is then introduced to detect straight lines (queuing) to generate the starting and ending time of red lights. SC20 20-Franklin 10, Marriott Resource Allocation and Pricing in Cloud Computing Cluster: Cloud Computing Invited Session Chair: Julie Ward, Distinguished Technologist, HP Labs, 1501 Page Mill Rd, Palo Alto, CA, 94301, United States of America, jward@hp.com 1 - Cost-efficient Cloud Computing Cloud users today must navigate through a variety of pricing mechanisms, and must contend with issues such as supply/demand forecasting and the high- dimensional control problems that naturally arise when attempting to minimize cost. We study the problem of computation under a deadline. We provide a simple, scalable algorithm that requires no tuning and enjoys robust performance guarantees. Along the way, we address a generalization of the classical Secretary Problem and prophet inequalities. 2 - Results-based Pricing and Resource Allocation in Cloud Computing Filippo Balestrieri, HP Labs, 1501 Page Mill Rd, Palo Alto, CA, 94301, United States of America, filippo.balestrieri@hp.com, Julie Ward, Bernardo Huberman Cloud services are sold today via a resource-based model, in which customers pay per instance-period. New technologies for predicting job requirements and completion times allow Cloud providers to consider new mechanisms. We compare a resource-based model to a results-based mechanism, in which the provider offers a menu of completion times and prices to each customer for his specific job. We identify conditions under which one mechanism produces higher revenue for the provider than the other. 3 - Selling Guaranteed Completion Times on the Cloud Andrew Li, MIT Operations Research Center, 77 Massachusetts Avenue, Bldg. E40-149, Cambridge, MA, 02139, United States of America, aali@mit.edu, Filippo Balestrieri, Julie Ward In today’s cloud market, users execute their own jobs without a guarantee that deadlines will be met. Instead, providers can take control of job execution and charge for guaranteed completion times, but they face the joint challenges of dynamically pricing such contracts and scheduling jobs to fulfill these contracts. We address these challenges with a revenue management formulation, and apply a fluid approximation that is computationally efficient and optimal for large systems. Piscataway, NJ, 08854, United States of America, amirali.mostafavizadeh@rutgers.edu, Peter J. Jin Vivek Farias, Associate Professor, MIT, 100 Main Street, Cambridge, United States of America, vivekf@mit.edu, Muhammad Amjad, Andrew Li, Devavrat Shah

99