Informs Annual Meeting 2017

WA03A

INFORMS Houston – 2017

WA03A Grand Ballroom A Machine Learning Sponsored: Applied Probability Sponsored Session Chair: David Goldberg, Georgia Institute of Technology, Atlanta, GA, 30332-0205, United States, dgoldberg9@isye.gatech.edu 1 - Weak Convergence Approach to the Multi Arm Bandit Problem Yilun Chen, Georgia Institute of Technology, 755 Ferst Drive NW, Atlanta, GA, 30332-0205, United States, ylchen@gatech.edu For the Bayesian multi-arm bandit problem, the optimal policy is known to be the so-called Gittins index policy. However, this policy is essentially a ‘’black box”, and it is apriori unclear how such a policy behaves, i.e. what an observer would actually see if a decision-maker implemented this policy. Using the framework of weak convergence, we shed light onto this question and the behavior of several associated stochastic processes. Time permitting, we will also discuss some related results for Thompson sampling. 2 - Optimal Solution to the Multinomial Selection Problem for Two Alternatives Timur Tankayev, Georgia Institute of Technology, Atlanta, GA, United States, timur.tankayev@gatech.edu, Craig Aaron Tovey The multinomial selection problem is to find a stopping policy for repeated independent trials, each of which reports a winner among competing alternatives, that has low expected cost and high probability of correct selection (PCS) of the best alternative. In 1959, Bechhofer, Elmaghraby and Morse formulated the problem as minimizing the worst case expected number of trials, subject to a lower bound on PCS and upper bound on the maximum number of trials, over all probability vectors outside an indifference zone. For the case of two alternatives, we prove that if one employs a particular probability vector known as the slippage configuration, then a linear program always finds an optimal stopping policy. 3 - Adaptive Learning with Robust Generalization Guarantees Rachel Cummings, California Institute of Technology, 1200 E California Boulevard, MC 305-16, Pasadena, CA, 91125, United States, racheladcummings@gmail.com Rachel Cummings, Georgia Institute of Technology, Atlanta, GA, United States, racheladcummings@gmail.com, Katrina A. Ligett, Kobbi Nissim, Aaron Roth, Steven Zhiwei Wu The traditional notion of generalization – i.e., learning a hypothesis whose empirical error is close to its true erro – is surprisingly brittle. As has recently been noted, even if several algorithms have this guarantee in isolation, the guarantee need not hold if the algorithms are composed adaptively. In this talk, we study three notions of generalization – increasing in strength – that are robust to postprocessing and amenable to adaptive composition, and examine the relationships between them. 4 - Learning via Distributionally Robust Optimization with Optimal Transport Costs Karthyek Murthy, Columbia University, New York, NY, United States, karthyek@gmail.com, Jose Blanchet, Yang Kang, Fan Zhang The idea of distributionally robust optimization (DRO) has been of active research recently in Operations Research to help beat “optimizer’s curse” when performing optimization under uncertainty. In this talk, we shall discuss how various machine learning estimators such as generalized Lasso estimator for linear regression, regularized logistic regression, support vector machines, etc. can be viewed under the unifying lens of DRO using optimal transport costs. In addition, for a larger class of useful models, we shall see that these DRO problems can be solved almost as fast, and in some cases even faster than, the respective traditional empirical risk minimization schemes.

3 - Integrated Optimization of Proactive and Reactive Project Scheduling to Minimize Uncertain Cost Yanting Wang, Xi’an Jiaotong University, Xi’an, China, wangyt.66@163.com In this paper, we present an integrated proactive-reactive model to investigate the optimal fashion of integrating the two scheduling methods under the objective of minimizing the total cost. In proactive phase, resource occupation cost is incurred through inserting buffers to protect potential disruptions and adjustment cost is generated to re-optimize the schedule when disruption occurs. An improved hybrid algorithm that combines a variable neighborhood search and a probabilistic tabu search is developed. Detailed computational experiments are presented to evaluate the improvement strategies and to prove the efficiency of the hybrid algorithm.

Wednesday, 7:30 - 9:00AM

WA01

310A Healthcare Decision Analysis II Sponsored: Decision Analysis Sponsored Session Chair: Mehmet Ayvaci, University of Texas at Dallas, Richardson, TX, mehmet.ayvaci@utdallas.edu Co-Chair: Yeongin Kim, Yeongin.Kim@utdallas.edu 1 - Determinants of Meaningful Usage of Health Information Technology Jingyun Li, California State University Stanislaus, Turlock, CA, United States, jli9@csustan.edu, Indranil Bardhan As Meaningful Use Stage 2 has been rolling out since 2014, a large portion of U.S. hospitals have finished or are undergoing the transition from paper-based medical records to electronic ones and use certified health information technology to improve healthcare quality, safety, and efficiency. Drawing on data from AHA Annual Survey Information Technology Supplement Survey and other databases, we study the characteristics of hospitals that are likely to achieve MU. 2 - Socio-emotional Support in Online Healthcare Q&A Community Zhe Shan, University of Cincinnati, 2925 Campus Green Dr (Linder Hall) Rm 509, Cincinnati, OH, 45221, United States, zhe.shan@uc.edu, Xin Xu In online health Q&A platforms, to be selected as best answer is not only about the quality of the response itself, but also affected by other factors such as social- emotional support. In this study, we collected data from “Yahoo! Answers” specifically in the allergies, autism, cancer and diabetes areas, and then used content analysis to encode the questions and their answers respectively. Based on the correlation analysis between different evaluation criteria for information quality, we explored whether socio-emotional support is an important factor in best answer selection. 3 - Accreditation Regime Change: Does “Announcing the Visit” Matter? Sehwon Kang, University of Minnesota, 321 Nineteenth Avenue South, Suite 3-150, Minneapolis, MN, 55455, United States, kangx584@umn.edu While accrediting agencies have changed their on-site accreditation survey visits from preannounced to unannounced, little extant literature has empirically examined the impact of this regime change on quality. We examine this impact on the relationship between accreditation and quality performance in US hospitals. 4 - Risk Assessment of Cyber Threats in Healthcare Devices Arik Sadeh, HIT Holon Institute of Technology, 52 Golomb St., Holon, 58102, Israel, sadeh@hit.ac.il A treat in a medical device may lead to various damages that can be categorized into two groups: damages to information systems and physical damages to patients. Damages may occur in different level of severity. This research is aimed to explore the importance of 60 cyber threats in healthcare devices. Data about 60 threats was collected from 30 expert respondents. Each threat was analyzed with respect to three information technology dimensions and two healthcare dimensions. The chance for occurrence of that threat was also considered. The concept of winning combinations using the algorithm of Quine-McClusky is used to determine the most important and serious threats.

441