INFORMS 2021 Program Book

INFORMS Anaheim 2021

TB15

TB12 CC Room 304D In Person: Railroad Maintenance General Session Chair: Faeze Ghofrani, Penn State Altoona, PA, United States 1 - Automatic Train Dispatching: A Real-life Application in the Greater Oslo Region Carlo Mannino, SINTEF Digital, Oslo, Norway, Giorgio Sartor, Andreas Nakkerud, Oddvar Kloster, Christian Schulz, Bjørnutar Leberget, Giorgio Grani Serving more than a million residents, the railway network of the Greater Oslo Region is composed of several lines incident to the large Oslo central station (Oslo S). An ongoing project with Bane NOR, the Norwegian infrastructure manager aims at developing a system to dispatch trains for the entire region. A prototype of such system is currently being tested by Bane NOR dispatchers. It uses mathematical optimization and decomposition to find optimal schedules (every few seconds) based on the real-time train positions and the network status. To our knowledge, this is the largest real-life application of automatic train dispatching in Europe. 2 - Inspection Technologies for Reliable Railway Transportation Systems Faeze Ghofrani, Assistant Teaching Professor, Pennsylvania State University, Altoona, PA, 16803, United States This study delivers an in-depth review of the state-of-the-art technologies relevant to inspection technologies giving emphasis to their use in railroad systems. The review not only looks at the research being carried out but also investigates the commercial products available for railroad systems inspection. It continues further to identify the methods suitable to be adopted in a moving vehicle detection system. Even though flaw detection has been a well-researched area for decades, an in-depth review summarizing all available technologies together with an assessment of their capabilities has not been provided in the recent past according to the knowledge of the authors. As such, it is believed that this study will be a good source of information for future researchers in this area. TB13 CC Room 201A In Person: Models of Learning and Experimentation in OM General Session Chair: Vikas Deep, Northwestern University, Evanston, IL, United States 1 - Markdown Pricing under Unknown Demand Su Jia, Carnegie Mellon University, Pittsburgh, PA, United States Dynamic pricing problem has been extensively studied recently. Naturally, people formulate such problems as variants of the MAB problems. Although bandits problems have been well understood from the theoretical perspective, bandit based pricing policies are rarely deployed in practice, mainly because they often overlooked some practical constraints. In this work, we consider the dynamic pricing problem under the monotonicity constraint, i.e. markdown pricing. We provide a complete settlement of the problem by providing simple, efficient markdown policies, with best possible theoretical guarantees, for each of those settings. 2 - Discriminative Learning via Adaptive Questioning Vikas Deep, Northwestern University, Evanston, IL, United States, Sandeep Juneja, Achal Bassamboo, Assaf Zeevi We consider the problem of designing an adaptive sequence of questions of varying degree of hardness thatoptimally classify a candidate’s ability into one of several categories or discriminative grades. A candidate’sability is modeled as an unknown parameter, which, together with the difficulty of the question asked,determines the probability with which s/he is able to answer a question correctly. The learning algorithm isonly able to observe noisy responses to its queries. We consider this problem from a fixed confidence-based -correct framework, that in our setting seeks to arrive at the correct ability discrimination at the fastestpossible rate while guaranteeing that the probability of error is less than a pre-specified and small .

TB14 CC Room 201B In Person: Reinforcement Learning with Engineering Applications II General Session Chair: Mohammad Dehghanimohammadabadi, Northeastern University, Boston, MA, 02115-5005, United States Co-Chair: Ashwin Devanga, Northeastern University, Boston, MA, 02120, United States 1 - Building Data Mining Test Environments to compare performance of different Algorithms including Reinforcement Learning Ashwin Devanga, Northeastern University, Boston, MA, United States, Mohammad Dehghanimohammadabadi We are working on a project called DM-Gym, an open-source python library for developing reinforcement learning (RL) algorithms to address data mining (DM) problems, and a testbed for creating multiple DM environments for RL such as regression, and classification. DM-Gym provides a new toolkit for the machine learning community to explore the capabilities of RL in solving data mining benchmarks. 2 - Reinforcement Learning Applications in Engineering using MATLAB Mohammad Dehghanimohammadabadi, Northeastern University, Snell Engineering Center, Boston, MA, 02115-5005, United States, Sahil Belsare, Rifat Sipah The project demonstrates RL’s capabilities of providing insightful results to the Engineering problems using MATLAB Environment. Tutorials from different domains such as HVAC control, Financial Portfolio management, Cart-Pole control, and Robotics are designed to (i) teach RL concepts, (ii) provide guidelines to create and develop engineering problem environments, and (iii) apply different RL and deep RL technique to solve them. 3 - The Third-Party Logistics Provider Freight Management Problem: A Reinforcement Learning Approach Amin Abbasi Pooya, University of Kansas, Lawrence, KS, 66045, United States, Michael Lash Third-party logistic (3PL) providers act as external entities that provide companies with partial/full logistics services. Contracting to 3PL providers allows companies to focus on their primary business objectives, while maintaining a consistent flow of products through their supply chains. We propose a framework that captures the salient objectives involved in the so-called “freight management problem” (FMP) faced by 3PL providers. To solve the FMP, we adopt Reinforcement Learning (RL) approaches, which permit the efficient learning of policies without making restricting assumptions. We find that RL methods substantially outperform adopted heuristics on simulated and real-world data. TB15 CC Room 201C In Person: Advances in Customers Behavior Analytics and Modeling General Session Chair: Yichen Ding, University of Iowa, Iowa City, IA, 52246-2872, United States Co-Chair: Amin Hosseininasab, University of Florida, Gainesville, FL, 32611-1942, United States 1 - From Favorited to Fear: An Empirical Investigation of Customer Emotions and Behavior of Online Customers After Data Breaches John N. Angelis, University of Maine, Orono, ME, 24504, United States, Rajendran Murthy, Tanya Beaulieu, Joseph Miller Previous empirical papers on data theft crimes often ask respondents to imagine that a well-known company has been hacked and then measures their response. We improved this design by first having respondents pick a favorite free or paid website and then presenting them with a data breach scenario involving their account or the entire site. Using automated textual analysis, we discover that only fear has a significant effect on breached customer behavior, and the customers who most likely to express positive sentiment pre-breach do not significantly differ in their post-breach behavior. 2 - Behavioral Analysis of Consumer Return Policy Decisions Han Oh, Mays Business School, Texas A&M University, College Station, TX, United States, Huseyn Abdulla, Rogelio Oliva We investigate consumer return policies recognized and studied by operations management scholars as an important managerial decision in a retail environment. Our research investigates, through randomized experiments, the behavioral aspects of return policy decisions and their interaction with other operational decisions.

103