Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC57

consumers, and provide strategic implications for startups such as Mishipay. 2 - Multi-period New Product Development and Risk Aversion Paradox Janne Kettunen, The George Washington University, Department of Decision Sciences, 2201 G. Street NW, Washington, DC, 20052, United States, Shivraj Kanungo We investigate how the conventional periodic risk-averse project selection approaches perform in multi-period new product development problems, in terms of satisfying the decision makers’ risk preferences. Our results show that the conventional project selection approaches can result in that the risk in attainable profit is systematically higher under the risk-averse selection approach than even under the risk-neutral selection approach. We call this phenomenon as the risk aversion paradox. We show how projects’ profits can be revised to overcome this paradox. 3 - Incentives for Managing Emissions in the Maritime Industry Leonardo P. Santiago, Copenhagen Business School, Department of Operations Management, Solbjerg Plads 3, Blok B. 5. sal, Frederiksberg, Denmark, Arundhati Srinivasan, Franz Buchmann This paper explores incentives for managing emissions in the maritime industry. We develop a market-based mechanism for smart carbon reporting. One of the key features of our model is to relax the assumption of information asymmetry by taking into account blockchain technology. We explore our model to discuss current challenges for policy makers and for potential opportunities for principals (ship owners). 4 - Familiar or Fresh. The Effect of Typicality on the Value of Product Design We examine how typicalityùthe degree to which a design is similar to other designs of the same categoryù affect the market value of designs. We compiled a dataset combining US design patent data and stock market reactions in the days subsequent to the design patent grant. We show that a focal design’s value rises when it is typical with respect to category-members released in the near past, and falls when it is typical with respect to category-members that are concurrently on the market. Finally, we show that designs in high clockspeed industries exhibit no positive effect of typicality on value. Our results show how firms can rethink the `familiar or fresh’ conundrum to develop valuable designs. Tian Chan, Emory University’s Goizueta Business School, 1300 Clifton Road, Atlanta, GA, 30322, United States Health Care and Bandits Sponsored: Health Applications Sponsored Session Chair: Lawrence M. Wein, Stanford University, Stanford, CA, 94305- 5015, United States 1 - Mostly Exploration-free Algorithms for Contextual Bandits Khashayar Khosravi, Stanford University, Stanford, CA, 94305, United States, Hamsa Sridhar Bastani, Mohsen Bayati The contextual bandit literature has focused on algorithms that address the exploration-exploitation tradeoff. Exploration-free greedy algorithms are desirable in settings where exploration may be costly or unethical. We prove that, under some assumptions on the distribution of the contexts, the greedy algorithm is rate-optimal for the two-armed bandit. Also, even absent these assumptions, we show that a greedy algorithm is optimal with nonzero probability. Thus, we introduce Greedy-First, an algorithm that uses only observed contexts and rewards to decide whether to follow a greedy algorithm or to explore. Greedy- First is rate-optimal without any additional assumption. 2 - Precision Healthcare Using Non-stationary Bandits Yonatan Mintz, UC Berkeley, Berkeley, CA, United States, Anil Aswani, Philip Kaminsky, Elena Flowers, Yoshimi Fukuoka Recommendations from fitness tracking devices are often ineffective in motivating users to exercise because the recommendations are not tailored to specific individuals. In this talk, we address this by using the data and infrastructure of fitness tracking devices to personalize exercise programs for users. We develop and analyze a new multi-armed bandit model, which we call the ROGUE multi- armed bandit, to adaptively learn each participant’s exercise preferences and personalize their exercise programs to increase adherence. We present both computational and theoretical results that show the efficacy of this modeling approach when compared to existing precision fitness approaches. 3 - The Impact of Using an Online Health Platform on Weight Management Yingfei Wang, University of Washington, Seattle, WA, United States, Lu Yan Overweight and obesity have raised severe social issue in the world. Many weight-loss communities are designed, aiming to provide individuals with community support and thus help with their weight management process. n TC56 West Bldg 101A

However, it has been noticed that such communities suffer from high drop-out rate. Therefore, we are interested in optimizing the online healthcare platforms strategy in suggesting a more manageable and encouragingweight loss goal, in order to facilitate user’s continuous participation. In order to take into account heterogeneity in need for treatment across individuals, we formulate our problem in a Bayesian contextual bandit setting. 4 - Best Arm Identification in Generalized Linear Bandits Abbas Kazerouni, 2071 El Camino Real, Palo Alto, CA, 94306, United States Many real-world problems in personalized medicine and advertisement can be formulated as best arm identification in a structured bandit. In such problems, the goal is to identify the best among a possibly infinite set of actions with the fewest number of trials. While there is a vast literature on best arm identification in Multi-Armed Bandit with independent arms, little is known about efficient exploration schemes for structured bandits with dependent arms. In this talk, we introduce an algorithm for best arm identification in generalized linear bandits and provide a bound on its sample complexity. We further illustrate the applicability of the proposed algorithm by providing simulation results. n TC57 West Bldg 101B Prevailing Issues in Public Sector OR Sponsored: Health Applications Sponsored Session Chair: Hrayer Y. Aprahamian, Virginia Tech, Blacksburg, VA, 24060, United States 1 - The Two-stage Group Testing Problem: The Exact Analytical Group testing, i.e., testing multiple subjects simultaneously with a single test, is essential for classifying a large population of subjects as positive or negative for a binary characteristic. We develop exact closed-form expressions for the two stage group testing problem, and use these results to gain novel insights on the dynamics of an optimal solution. These results enable us to exactly solve a robust formulation of the group testing problem, which, prior to our findings, was intractable to solve. We demonstrate the value of robust testing schemes with a case study on public health screening. 2 - An Inventory Policy Model for Intravenous Fluids Sasan Khorasani, Texas Tech University, Texas, Lubbock, TX, 79415, United States, Milton Louis Smith, Jennifer Cross Intravenous fluids (IV) waste is a major source of inpatient pharmacy cost related to medication waste. Moreover, few works have been conducted on inventory policy specifically for IV medications. The optimal inventory policy for IV medications requires considering detailed parameters and variables that significantly impact waste. Thus, this study aims to observe the IV delivery process in an inpatient pharmacy, discover the main roots of IV waste for different types of medications, and establish the optimal replenishment policy for IV delivery systems. 3 - Online Health Communities and Health Literacy Sejun Park, University of Texas at Dallas, 800 W. Campbell Rd, Richardson, TX 75080, Richardson, TX, 75080, United States, Taewoo Roh, Giyoon Kwag Knowledge sharing via online health communities has been considered as a major method of communication to fortify user health literacy. These connections improve patient access to extensive healthcare information. This research investigates the impact of knowledge sharing on patient health literacy based on the activities of users in the online health community. This study outlines whether users who are not literate in medical terminology benefit from these knowledge communities and expands on the dynamics of these virtual interactions. 4 - Optimal Resource Allocation for Surveillance of Emerging Infections Ngoc Nguyen, Virginia Tech, 1145 Perry Street, 214 Durham Hall, Blacksburg, VA, 24060, United States, Ebru Korular Bish, Douglas R. Bish Accurate prevalence estimation is essential for planning of healthcare services, especially for emerging infections such as Zika. Therefore, policy-makers need to allocate the scarce resources among the surveillance activities of different infections or regions in the most efficient way. Towards this end, we develop stochastic optimization models and determine structural properties of their optimal solutions. Our case study on surveillance efforts for Zika highlights the benefits of optimization-based approaches to this decision. We also establish guidelines on effective testing strategies for surveillance of emerging infections. Solution with an Application to Robust Group Testing Hrayer Y. Aprahamian, Virginia Tech, 1145 Perry Street, Blacksburg, VA, 24060, United States, Ebru Korular Bish, Douglas R. Bish

327