INFORMS 2021 Program Book

INFORMS Anaheim 2021

ME17

3 - Stochastic Optimization of Area under Precision-Recall Curve with Provable Convergence Qi Qi, The University of Iowa, Iowa City, IA, United States, Tianbao Yang In this work, we propose a principled technical method to optimize area under Precision-Recall Curve (AUPRC) for deep learning. Our approach is based on maximizing the averaged precision (AP), which is an unbiased point estimator of AUPRC. We cast the objective into a sum of dependent compositional functions with inner functions dependent on random variables of the outer level. We propose efficient adaptive and non-adaptive stochastic algorithms with provable convergence guarantee under mild conditions by using recent advances in stochastic compositional optimization. Extensive experimental results on graphs and image datasets demonstrate that our proposed method outperforms prior methods on imbalanced problems. To the best of our knowledge, our work represents the first attempt to optimize AUPRC with provable convergence. 4 - Safeguarding Data Privacy in the Era of Artificial Intelligence Kwan-Yuet Ho, United States Because knowing individuals’ preferences is a lucrative business, personalization has become a common task among data science teams in various commercial and government sectors. The inevitable use of personal data puts people’s privacy at risk. Some hackers might simply steal the data illegally. Some employees who are working on the data and machine learning models might make inappropriate use of the data for their own purpose. Sometimes even if the data are turned “anonymous,” the use of artificial intelligence can sometimes reveal the identity of the individuals represented by the data. In this talk, I will talk about how to protect the privacy of everyone in three aspects: 1) the system architecture design; 2) the rule-based removal of the sensitive information; and 3) the use of machine learning models to further eliminate information about the individuals. 5 - Customers Privacy Protection Behaviors in E-commerce Transactions: The Role Of Privacy Concerns, Privacy Risks, and Privacy Policies. Katia Guerra, University of North Texas, Denton, TX, United States, Vess Johnson This research investigates the impact of information privacy concerns on the awareness of privacy risks and on the awareness of organizational privacy policies toward the adoption of customers’ privacy protection behaviors in the context of e-commerce transactions. We develop a research model and we test it by employing a Qualtrics online survey questionnaire. We use crowdsourcing through Mechanical Turk to collect data. The contribution of this study is to shed the light on the mechanisms that lead consumers to take privacy protection behaviors rather than to interrupt e-commerce transactions in the light of privacy risks, on one side, and organizational privacy policies, on the other side. ME17 CC Room 202A In Person: Decision Models for Resilient Network Operations General Session Chair: Mathieu Dahan, ISyE Georgia Tech, Atlanta, GA, 30309, United States 1 - Network Inspection Against Strategic Attacks Using Heterogeneous Detectors Bobak McCann, Georgia Institute of Technology, Atlanta, GA, United States, Mathieu Dahan We consider a network inspection game, in which a defender positions heterogeneous detectors according to a probability distribution in order to detect multiple attacks caused by a strategic attacker. We assume the defender has access to multiple types of detectors that can potentially differ in their accuracy and cost. The objective of the defender (resp. attacker) is to minimize (resp. maximize) the expected number of undetected attacks. We provide a full analytical characterization of the Nash equilibria for this game under the assumption that each component in the network can be monitored from a unique detector location. Then, we determine the optimal detector investment for the defender that guarantees a target detection level in the worst case. 2 - Resilient Hyperconnected Intercity Parcel Delivery Network Design Onkar Kulkarni, ISyE Georgia Tech, Atlanta, GA, United States, Yaarit Cohen, Mathieu Dahan, Benoit Montreuil In this work, we study a tri-level optimization model to design a hyperconnected intercity parcel delivery network that is capable to withstand coordinated disruptions. In the first stage, hubs are positioned, and service routes are set up. Then, a fictitious adversary disrupts one or more routes. Finally, the flow of parcels is selected as to avoid the disruptions. The overall aim is to design a network that minimizes the total distance travelled by parcels after worst-case disruptions. We propose an approach to approximately solve this challenging problem by decoupling the first stage from the others. We validate our approach by designing networks across China, analyze their efficiency and resilience under various worst-case scenarios.

3 - A Study of Projection-free Gradient-based Algorithms for Zero-sum Linear Quadratic Games Arnesh Sujanani, Georgia Institute of Technology, Atlanta, GA, United States, Vidya Muthukumar, Mathieu Dahan Zero-sum linear quadratic games can be formulated as nonconvex-nonconcave min-max problems. Recent work has established that under certain conditions, stationary points of the objective are Nash equilibria. Furthermore, such work has established that projected nested gradient methods, under such assumptions, achieve global sublinear convergence to Nash equilibria. We build upon this work and study the landscape of gradient based algorithms for linear quadratic games. In particular, we study the convergence, or lack thereof, of projection-free gradient-based algorithms for linear quadratic games. Our analysis primarily relies on establishing conditions under which the updates of gradient-based algorithms remain stable. 4 - Operations Planning and Risk Management in Covid-19 Surveillance Testing Programs Hannah Wilborn Lagerman, ISyE Georgia Tech, Atlanta, GA, United States, Mathieu Dahan, Pinar Keskinocak The success of a surveillance testing program relies on managing risks which threaten to disrupt its operations. In this work, we aim to identify and mitigate such risks for Covid-19 testing laboratories that conduct pooled as well as diagnostic testing. Using machine learning and simulation, we quantify the laboratory’s testing capacity, evaluate bottleneck improvements, and analyze laboratory’s sensitivity to variability in supply chain conditions. We apply this approach to Georgia Tech’s Covid-19 testing laboratory to improve its operations and inventory management as a function of disease positivity rate, pool size, and predicted testing demand. ME18 CC Room 202B In Person: Environmental Sustainability Analysis and Strategies General Session Chair: Xichen Sun, College Station, TX, 77845, United States 1 - Searching for the Synergy Between Environmental Sustainability and Economic Sustainability Sidi Deng, Research Assistant, Purdue University, West Lafayette, IN, United States, Yuehwern Yih, John W. Sutherland This presentation concentrates on the synergy between environmental sustainability and economic sustainability. The presenter will provide an overview of some innovative methodologies that incorporate environmental factors into the framework of economic evaluation. These methods fall into three categories, which will be introduced in three sections:(1) Integrating techno-economic assessment (TEA) and life cycle analysis (LCA) framework. An overview of a TEA template adopted by Purdue and The University of Arizona will be given.(2) Including external cost factors into the cost-benefit analysis. This section highlights a dynamic price model that analyzes the impact of environmental regulation/policies on the associated market.(3) A discussion on the relationship between economic growth and environmental sustainability. 2 - Process Optimization for Bioleaching Lithium-ion Batteries Majid Alipanah, University of Arizona, Tucson, AZ, United States, Hongyue Jin, Qiang Zhou, Caitlin McNamara, Vicki Thompson, Yoshiko Fujita, David Reed By 2040, end-of-life lithium-ion batteries (LIBs) are expected to reach approximately 8 million tonnes per year. Several critical materials found in LIBs, such as lithium, cobalt, and nickel, have limited resources and are mined in only a few countries. Bioleaching technology has the potential to recycle EOL LIBs in a cost-effective and environmentally friendly manner. This study tried to optimize the bioleaching process for value recovery from EOL LIBs through integrating techno-economic assessment and response surface methodology. The result of the optimization showed the economic feasibility of the process and the life cycle assessment showed its sustainability by the lower environmental footprint compared to other leaching methods 3 - On the Complementarity Between Servicizing and Remanufacturing: Economic and Environmental Implications Xichen Sun, Texas A&M University, College Station, TX, United States, Rogelio Oliva, Tharanga Rajapakshe The recent shift in customer demand from product ownership to other alternatives has motivated manufacturers to offer servicizing a business strategy that sells a product’s functionality as a service. This trend is observed in the industries where remanufacturing is prevalent. Offering servicizing facilitates remanufacturing by providing a stable flow of high-quality returns, thus reducing remanufacturing-related costs. Further, servicizing can satisfy customer demand with reduced production, reinforcing remanufacturing’s positive effect on the environment. To investigate these complementarities, we consider a profit- maximizing manufacturer who markets new and remanufactured products and explores the possibility of introducing servicizing. We develop stylized models that capture the main product attributes for remanufacturing and servicizing.

92