Informs Annual Meeting 2017

SC73

INFORMS Houston – 2017

SC73

2 - Mitigating Ambiguity in Markov Decision Processes using Multiple Models of Parameters Lauren N.Steimle, University of Michigan, 3261 Bolgos Circle, Ann Arbor, MI, 48105, United States, steimle@umich.edu, David L. Kaufman, Brian T. Denton Markov decision processes (MDPs) are a powerful tool for optimizing sequential decision-making under uncertainty. However, the usefulness the resulting policy is often at the mercy of the data used to parameterize the MDP. Often times, the transition probabilities must be estimated from external sources, and it can be unclear which source is most appropriate to use. In this talk, we describe a new method for mitigating the effects of parameter ambiguity by allowing for multiple models of the MDP parameters. We discuss properties of optimal policies, the complexity, and solution methods. We highlight the merits of this approach using a case study in the context of cardiovascular disease management. 3 - Establishing an Optimal IRT Threshold for Cystic Fibrosis S. Saloumeh Sadeghzadeh, PhD Candidate, Virginia Tech, Cystic Fibrosis (CF) is one of the most prevalent genetic diseases in the United States. Newborn screening for CF allows for early diagnosis, and can substantially improve outcomes. Most first-tier screening for CF in the United States is performed via the IRT test, which measures the level of a CF-related biomarker. However, newborn testing practices for CF that involve the IRT test vary significantly among the states. We determine optimal testing schemes for newborn CF screening. Our case study demonstrates the value of the proposed testing schemes in newborn screening. 4 - Optimal Set of Interventions for Childhood Obesity: Benefit and Cost Parisa Zahiri, M.Sc., Tehran Azad University, Tehran, Iran, Islamic Republic of, zahiriparisa86@gmail.com Childhood obesity is a public health crisis in many other countries. The cause of childhood obesity is complex, involving biological, behavioral, psychosocial, demographic, economic, community, and environmental factors. The goal of this project is to create a model to estimate the effect of each intervention on childhood obesity over time considering a broad range of risk factors widely accepted in the literature and adherence. The most cost-effective set of interventions is then investigated. Co-Chair: Mario Eduardo Villanueva, PhD, Texas A&M University, College Station, TX, United States, me.villanueva@tamu.edu 1 - Global Optimization based on Edge-Concave Underestimators M.M. Faruque Hasan, PhD, Texas A&M.University, College Station, TX, United States, hasan@tamu.edu A new relaxation technique is developed for the deterministic global optimization of twice-differentiable continuous problems. Instead of using a convex underestimator, the relaxation is based on an underestimator which is edge- concave. The underestimator is constructed by subtracting a positive quadratic expression. We use the linear facets of the vertex polyhedral convex envelope of the edge-concave underestimator to obtain a linear programming (LP)-based relaxation of general nonconvex functions. The method will be presented with theoretical results and will be compared with convexification/ underestimation techniques such as BB and its variants through test examples. 2 - Global Optimization of Differential Games Mario Eduardo Villanueva, Texas A&M.University, Research This talk presents novel strategies of formulating 2-stage dynamic games in the form of a generalized optimal control problem (OCP) for systems with two control inputs. The first control input is chosen by Player 1. Player 2 plays after Player 1 by choosing the second control input. Both players have to satisfy state- and control constraints or loose the game. In addition, one of the players tries to minimize an objective that is maximized by the other player. First, we discuss differences and similarities to standard robust OCPs. Then, we outline numerical strategies for analyzing and solving dynamic games by using algorithms from the field of set-valued propagation and dynamic global optimization. Parkway, College Station, TX, 77845, United States, me.villanueva@tamu.edu, Xuhui Feng, Boris Houska 428 Harding Ave, Blacksburg, VA, 24060, United States, saloumeh@vt.edu, Ebru Korular Bish, Douglas R. Bish SC75 372D Global Optimization Sponsored: Optimization, Nonlinear Programming Sponsored Session Chair: Boris Houska, borish@shanghaitech.edu.cn

372B Blockchain for Supply Chain and Logistics Invited: Operations Research and the Future of Computing Invited Session Chair: Robin Lougee, IBM Research, Yorktown Heights, NY, 10598, United States, rlougee@us.ibm.com Co-Chair: Chandra Narayanaswami, IBM Research, NY, United States, chandras@us.ibm.com 1 - The Systemic Adoption of Blockchain with in the Aviation Ecosystem Robert J. Rencher, The Boeing Company, Seattle, WA, 98008, United States, robert.j.rencher@boeing.com This presentation is intended to provide to the participants a background on aviation industry supply chain as it exists today, challenges it is facing - in speed, trust, repetitive work, complexity, provenance, disputes, etc. We will discuss the following: 1) Why Blockchain is of interest to the aviation industry. 2) A primer on the state of managing information in the aviation ecosystem. 3) Challenges for Blockchain adoption by the aviation industry and 4) Value propositions blockchain brings to the aviation industry. 2 - Blockchain Technology for Ocean Container Shipping Jeffrey L. Howard, INTTRA, Parsippany, NJ, 07504, United States, jeff.howard@inttra.com Jeff Howard, SVP and Chief Product Officer of ocean shipping e-commerce leader INTTRA, will discuss his observations of the use and potential benefits of Blockchain technologies in the ocean shipping industry and logistics. Howard will also share his observations of challenges facing the ocean container shipping industry including complex, multi-party transactions, the need for digitization, issues with trust across transactions and parties, transaction velocity and workflow, as well as the desire to balance transaction confidentiality and security with increased visibility to required workflow milestones. 3 - Introduction to Blockchain Technology and Supply Chain Use Cases Chandra Narayanaswami, Principal Research Staff Member, IBM.Research, 1101 Kitchawan Road, Yorktown Heights, NY, 10598, United States, chandras@us.ibm.com A Permissioned Blockchain provides an immutable, shared ledger that is kept consistent using distributed and resilient consensus protocols, along with inherent provisions for data separation, privacy and security among the participants. Executable smart contracts orchestrate automation and efficiency. Additional analytics offer insights on aspects such as compliance and quality. Together it provides what is needed for an efficient next generation supply chain, with provable provenance, for ordering, tracking, and maintaining assets and services need to operate in distributed cross-organizational settings where partial lack of trust among organization is common. 372C Computational Methods in Healthcare Sponsored: Computing Sponsored Session Chair: Amin Khademi, Clemson University, Central, SC, 29630, United States, khademi@clemson.edu 1 - Infectious Disease Modeling in a Meta-Population Ceyda Yaba, Clemson University, 833 Old Greenville Highway, Apt 614, Clemson, SC, 29630, United States, cyaba@g.clemson.edu, Burak Eksioglu, Amin Khademi We studied an infectious disease without cure in a meta-population over a geographical region that is divided into subpopulations. Each individual is either susceptible or infected; and the rate of infection is independent for every subpopulation. We model this scenario as a Markov Decision Process, where the actions are defined as the number of people to quarantine in every subpopulation under a capacity constraint for the entire population. We solve this model by Approximate Dynamic Programming approach in which a column generation algorithm is used. The results from the column generation are then compared to a simulation model to offer bounds for the value functions of the Markov Decision Process. SC74

94