INFORMS 2021 Program Book

INFORMS Anaheim 2021

ME32

ME32 CC Room 208B In Person: Revenue Management General Session Chair: Rene A. Caldentey, The University of Chicago, Chicago, IL, 60637-1656, United States 1 - Tight Guarantees for Multi-unit Prophet Inequalities and Online Knapsack Jiawei Zhang, New York University, New York, NY, 10012-1106, United States, Jiashuo Jiang, Will Ma Prophet inequalities are a fundamental tool for comparing the performance of online vs. offline algorithms. In the basic setting of k-unit prophet inequalities, the celebrated algorithm of Alaei (2011) with a performance guarantee of 1- 1/sqrt{k+3} has been applied in online advertising, healthcare scheduling, and revenue management. Despite its wide applicability for rounding an LP solution, the tightness of this guarantee for a given k has remained unknown. In this paper we resolve this question, characterizing the tight bound using differential equations and deriving the best-known guarantee for k-unit prophet inequalities. In the generalization of the online knapsack problem, we also derive an improved and tight guarantee of 1/(3+1/e^2)~0.319, by bypassing the splitting of large vs. small items in our analysis. 2 - Posted Price Versus Auction Mechanisms in Freight Transportation Marketplaces Sungwoo Kim, Georgia Tech, Atlanta, GA, United States, He Wang, Xuan Wang We consider a truckload transportation marketplace in which a platform serves an intermediary to match shippers, who make payment to the platform for transportation services, with carriers, who book loads and get compensation from the platform for transporting the loads. The objective of the platform is to design a policy that specifies how to set prices for shippers and payments to carriers, as well as how carriers and loads should be matched, in order to maximize its long- run average profit. This research analyzes theoretical performances of posted price, auction, and hybrid mechanisms which combine posted price and auction mechanisms. ME33 CC Room 209A In Person: Experimental Design for Marketplaces General Session Chair: Chen Chen, Booth School of Business, The University of Chicago, Durham, NC, 27708-9972, United States Chair: Rad Niazadeh, Chicago Booth School of Business, Stanford, CA, 94305-5008, United States 1 - Balancing Covariates in Randomized Experiments with the Gram- schmidt Walk Design Christopher Harshaw, Yale University, New Haven, CT, United States, Fredrik Sävje, Daniel Spielman, Peng Zhang In the design and analysis of Randomized Control Trials, it is widely accepted that balancing pre-treatment covariates between the treatment groups may lead to improved precision of treatment effect estimates when the covariates are correlated with outcomes. However, we argue that there is a fundamental trade- off between efficiency gained by covariate balance and robustness of these estimates.We present the Gram-Schmidt Walk Design, which allows experimenters to optimally navigate this trade-off. The design utilizes recent advances in algorithmic discrepancy theory [Bansal et al 2019]. We provide a tight analysis of the design, including non-asymptotic bounds on the variance and tails of the Horvitz-Thompson estimator. Based on these results, we develop estimators for non-asymptotic confidence intervals. 2 - Near-optimal Experimental Design for Networks: Independent Block Randomization Chen Chen, Booth School of Business, The University of Chicago, Chicago, IL, 27708-9972, United States, Ozan Candogan, Rad Niazadeh We consider the problem of designing a randomized experiment for a network of users. A decision-maker uses an unbiased Horvitz-Thompson estimator to estimate the total market effect of the treatment and chooses an optimal joint distribution of randomized assignments to minimize the worst-case variance of this estimator. For networks that can be partitioned into densely connected communities by ignoring only a small number of connections, it is near-optimal to assign all users in the same community to the same variant. We develop a family of independent block randomization (IBR) experiments, and we show these policies are asymptotically optimal when the number of communities grows large and no community size dominates the rest. Beyond the asymptotic regime, the IBR experiment is 7/3-approximation for any problem instance.

ME34 CC Room 209B In Person: Trust in the Digital Economy and Social Media General Session Chair: Yangyan Shi, Auckland, New Zealand 1 - Reverse Logistics Capability for Sustainable Development Olivia Lee, Macquarie University, Sydney, Australia The purpose of this paper is to present a conceptual framework of reverse logistics capability in the pharmaceutical industry. Today, sustainability is becoming increasingly important for all companies, across all industries including the pharmaceutical industry. In addition, reverse logistics plays a vital role in recycled material flows and waste management in supporting sustainable 2 - Risk Factor Analysis of the Time to Progression for Alzheimer’s Disease Patients Robin Qiu, Penn State (The Pennsylvania State University), Malvern, PA, United States In this talk, we show a predictive model for probable Alzheimer’s disease (AD) patients to enable risk factor analyses, aimed at providing medical professionals and caregivers a tool so that personalized care becomes possible. As AD patients’ time to progression generally accelerates over time, a predictive model should account for a patient staging status to determine his/her personalized and precise care. The Global Staging Clinical Dementia Rating score is used to define AD patients’ progression stages. Hence, we present cluster-based predictive models, enabling supporting decision-making on the guidance of personalized care and disease management on an individual patient basis. 3 - Improving Social Media Presence of Firms under Budget Constraints: A Multi-method Approach Mayukh Majumdar, PhD Candidate, Mays Business School, College Station, TX, United States, Subodha Kumar, Chelliah Sriskandarajah The use of social media platforms by firms to promote their products among the public has received major attention among researchers and practitioners, especially the image-based content given the widespread availability of multimedia-based platforms. In this study, we examine the role of post features in driving user engagement and the operational value in the analysis of social media content. We use a combination of deep learning method, econometric approach, and optimization framework to offer relevant managerial insights under budget constraints. 4 - Artificial Intelligence-as-a-service in Healthcare Youakim Badr, Associate Professor, The Pennsylvenia State University, Malvern, PA, United States, Robin Qiu Artificial Intelligence-as-a-Service typically refers to off-the-shelf software packages that are offered by third party vendors to implement AI solutions with minimal investment. Healthcare is likely to be the one market where AIaaS can truly have impacts that affect human lives with a large spectrum of applications, ranging from drug discovery, personalized medicine, personal health virtual assistants, survival analysis, and surgery robots, etc.In this study, we survey the current status of AIaaS in healthcare and identify common setbacks, including a lack of data exchange, data privacy and security, regulatory compliance, ethic and fairness, patient and provider adoption. ME36 CC Room 210B In Person: Emerging Transportaion and Transportation-Enabled Services General Session Chair: Neda Mirzaeian, Carnegie Mellon University, Pittsburgh, PA, 15217-1249, United States 1 - Multimodal Transportation Alliance Design with Endogenous Demand: Large-scale Optimization for Rapid Welfare Gains Kayla Spring Cummings, Massachusetts Institute of Technology, Cambridge, MA, 02134-1506, United States, Vikrant Vaze, Ozlem Ergun, Cynthia Barnhart We present an incentive-aligned collaborative pricing structure for competing urban transportation operators to jointly maximize system welfare. The alliance requires no change in infrastructure and captures behavior of all strategic agents in the system. We demonstrate the alliance’s utility on a full-scale case study of the Greater Boston Area which integrates 10 different datasets.

96