INFORMS 2021 Program Book
INFORMS Anaheim 2021
SB43
2 - What Went Wrong? Signals and Mechanisms for Unintended Consequences in AI Madhav Sharma, Oklahoma State University, Stillwater, OK, United States, David P. Biros, Corey Baham The technologies that we have come to know as Artificial Intelligence (AI) are becoming general-purpose technologies. However, that impact has not been entirely positive. This research seeks to uncover common signals and mechanisms that lead to unintended consequences in AI. Using a grounded theory approach, we propose a unifying theoretical framework for unintended consequences in AI projects. We analysed 840 quotes from key informants about 30 unique AI cases using multiple news articles for each case. Our analysis of media discourses revealed signals of intended actions concerning the implementation of AI tools, which led to unintended consequences through various linking mechanisms. 3 - Dynamic Probabilistic Deep Learning Forecasting Models for Traffic Congestion Pedro Cesar Lopes Gerum, Cleveland State University, Cleveland, OH, United States We provide a new framework for the dynamic prediction of traffic density distributions using deep learning. The proposed models provide accurate transient distributions by extending and improving state-of-the-art sequential deep learning models. Moreover, they are flexible and work for both distributional and quantile forecasts. The models are validated using three different data sets, and the results suggest that the proposed models are general and can outperform an extensive list of common models for probabilistic time series forecasting. 4 - A Reinforcement-learning Approach to Credit Collections Michael Mark, EPFL, Cugy Vd, Switzerland, Naveed Chehrazi, Thomas A. Weber This paper develops a dynamic reinforcement-learning agent capable of finding high-quality policies for the practice of debt collections. At its core, the agent effectively learns how to control a stochastic self-exciting point process in order to maximize an asynchronously obtained reward. Because we use a general formulation of the problem as an agent-environment interaction our results are readily extensible beyond the presented application. Furthermore, with the growing need for interpretable machine-learning models we augment the learning procedure with a monotonicity regularizer which makes learned policies intuitively understandable for human decision makers. SB43 CC Room 213A In Person: Socio-economic Impacts of Digital Technology General Session Chair: Nils Van den Steen, Southern Methodist University Cox School of Business, Dallas, TX, 75275 1 - Dynamic Pricing to Balance On-demand Vehicle Rental Networks: Empirical Evidence from Carsharing Karsten Schroer, University of Cologne, Cologne, Germany, Muhammed Demircan, Wolfgang Ketter Dynamic pricing has been proposed as a theoretically appealing way to manage demand and supply imbalances that occur in on-demand vehicle rental networks. Literature on imperfect information markets and platform lock-in, however, indicates that price adjustment may fail to materially influence demand in such systems. We resolve this tension by means of a large-scale econometric investigation of transaction-level carsharing data. Our analysis of the causal effect of price on vehicle utilization paints a nuanced picture. We show that price premiums reduce vehicle utilization, while price reductions have no significant effect indicating highly asymmetric price sensitivity. Finally, we derive recommendations for improving the effectiveness of dynamic pricing policies in rental networks. 2 - The Peril of Free Product Sampling on Online Crowdfunding Platforms Zibo Liu, University of Washington, Seattle, WA, 98105-5835, United States, Weijia You, Yong Tan Crowdfunding market has developed fast recent years. However, the problem of information asymmetry in the market is still an issue. In this paper, we study the impact of a novel mechanism in crowdfunding market, free product sampling, on crowdfunding projects. Leveraging a rich data set from a large online crowdfunding platform in China, we construct a structural model considering both demand side and supply side of the market. We find that sampling campaign hurts crowdfunding projects in every stage of the campaign, namely Application Stage, Trial Stage, and Report Stage. Backers’ strategic delay and the negative impact of revealed information on projects results in this surprising result. Our research fills the research gap of free product sampling in crowdfunding market and provides meaningful managerial implications to both fundraisers and crowdfunding platforms. 3 - Digital Technology Choices by Buyers and Sellers for Open
Market Business-to-Business Transactions Nils Van den Steen, SMU Cox School of Business, Dallas, TX, United States, Steve Muylle, Amit Basu, Willem Standaert Firms that use digital technologies to support their transactions in an open market, where any firm can access the market as a buyer or a seller, can make those digitalization decisions independently. As a result, candidate transaction counterparties in an open market could end up missing each other or fail to realize the full benefits from their digitalization efforts because they made different choices. Based on an empirical analysis of buyer-seller dyads this study examines the relation between the technology choices by the buyer and the seller in an open market transaction and the benefits that both firms realize from their digitalization efforts.
SB44 CC Room 213B In Person: Supply Chain Management and Revenue/Yield Management Contributed Session
Chair: Sophia Huang, Vistex, Inc, Kent, WA, 98032, United States 1 - Contingency Planning for Combined Adaptation of Healthcare and Commercial Supply Chains for a Pandemic Response Oleg Gusikhin, Ford Motor Company, Dearborn, MI, United States, Xingyu Li, Dmitry Ivanov, Kathryn E. Stecke During the COVID-19 pandemic, severe shortages have been observed in healthcare production entailing ad hoc supply chain (SC) adaptation by using capacities of commercial companies. Time delays, high preparation efforts, and long shortage periods have been seen during these adaptations. We hypothesize that some collaborative preparedness to the contingent structural SC adaptation with development of upfront plans for combining commercial and healthcare SCs should result in higher efficiency and effectiveness. Using optimization and simulation methodology, the value of collaborative structural adaptation is examined in the context of different industrial infrastructures. 2 - Multiechelon, Multicommodity Supply Chain Design with Uncertain Demand From a Climate Change Mitigation Perspective Reza Alizadeh, University of Oklahoma, Norman, OK, United States, Janet K. Allen, Farrokh Mistree According to the US EPA, companies with a supply chain (SC) generate about 42% of greenhouse gas. Thus, designing a green supply chain (GSC) is a reasonable solution to mitigate climate change. To design a GSC, we model the SC as a network of customers, stores, and warehouses. The number and location of stores are determined to find a low-cost and low emission configuration. A multi- echelon, multi-commodity SC with different warehouses and stores is designed. Using socio-spatial data, demand is predicted. The multi-echelon multi- commodity supply chain distribution and inventory systems are then considered in the proposed model for different carbon policies. 3 - Strengthening the Resilience of Seaport Terminals for Disruption Management Weimar Ardila, University of South Florida, Tampa, FL, United States, Alex Savachkin, Devashish Das, Daniel Romer o The implementation of resilience strategies is essential to ensure the regular operation of supply chains. Nonetheless, many actions that can increase resilience conflict with traditional business goals. The main research objective is to propose an initial approach for a Markov Decision Process (MDP) formulation to enhance a system’s response by minimizing the cost of implementing actions that can reduce a system’s total recovery time. A testbed based on the seaport terminals operation in Barranquilla (Colombia) will be used to validate and assess this approach’s performance. 4 - Lift Estimates and Schedule Optimization for Trade Promotion Planning Sophia Huang, Senior Data Scientist, Vistex, Inc., Chicago, IL, United States, Maarten Oosten Trade promotions consist of the multi-level promotional activities of a manufacturer. The manufacturer tries to promote their products to the consumers but since the retailers are the intermediate parties that sell and promote the products they must be incentivized to execute the promotions. After all, if the promotion is not attractive for the retailer, there won’t be a promotion. Therefore, the manufacturer should model the behavior of the consumers as well as that of the retailers. In this presentation we discuss the challenges this poses in both the estimation of the promotion effects as well as the optimization of the promotion schedule and propose models that address these challenges. SB45 CC Room 213C In Person: Influence of Supply Chain Practices on Emerging Economies
14
Made with FlippingBook Online newsletter creator