INFORMS 2021 Program Book

INFORMS Anaheim 2021

MC07

3 - Refugee Camp System Aid Allocation with Uncertain Demand and Replenishment Cycles Shima Azizi, Worcester Polytechnic Institute, Worcester, MA, 01545-4285, United States, Erhun Kundakcioglu, Andrew C. Trapp, Cem Deniz Caglar Bozkir, Ali Kaan Kurbanzade Camp-based refugees seek shelter in camps, and urban refugees in nearby areas. Aid distribution to camps should prioritize camp-based refugees, yet share excess inventory with urban refugees when able. Amid uncertainty in demands and replenishments, we derive an inventory policy to govern a camp’s aid sharing with urban refugees. We use the policy to construct expected costs of referring urban refugees elsewhere, depriving camp-based refugees, and holding, and embed them in a cost-minimizing aid allocation problem. Our study reveals insights into humanitarian aid allocation amid uncertainty. MC07 CC Room 303B In Person: Machine Learning in Action General Session Chair: Sara Masoud, Wayne State University, Detroit, MI, 48201-1111, United States Co-Chair: Dan Li, Georgia Institute of Technology, Atlanta, GA, 30309- 4360, United States 1 - An AI Driven Virtual Reality Platform for Human Robot Interaction Ali Kamali Mohammadzadeh, Wayne State University, Detroit, MI, United States, Sara Masoud, Jessica Rajko In this work, a smart virtual platform is developed by integrating a physics-based model with room-scale virtual reality, where a virtual cobot developed in Unity game engine is interacting with human(s) immersed using HTC Vive Pro-Eye Arena Bundle. This platform provides an opportunity for safe data collection, feeding ensemble deep artificial neural networks for intention classification and trajectory tracking. 2 - Autoencoders for Detecting Anomalies in Neonatal MRI Brain Scans Jad Raad, Ford Motor Company, Bloomfield Hills, MI, 48301, United States Anomalydetection in and analysis of adult MRI brain scans is a well studied field, butfew endeavors have been made to apply similar concepts to the neonatal brain.The challenges involved with analyzing the neonatal brain range from practicalto physiological, and cannot be understated, particularly due to the rapidgrowth and development seen in the developing brain. In this talk, we will discussthe successes and failures we’ve encountered in the process of developingautoencoders to detect anomalies in neonatal MRI brain scans. 3 - Pm2.5Forecasting Utilizing Graph Convolutional And LSTM Neural Networks Sara Masoud, Wayne State University, MI, 48201-1111, United States, Ali Kamali Mohammadzadeh, Abd Ali Hussain, Marisa O’Dea, Yaoxian Huang PM2.5, as inhalable particles with maximum diameters of 2.5 micrometers, are the cause of many serious health problems. Here, a PM2.5 forecasting framework is developed by integrating convolutional and recurrent neural networks. Although it is common to use recurrent neural networks to study the temporal behavior of PM2.5, this is the first work to take advantage of the geo-correlation of monitoring stations. Here, graph convolutional neural networks are implemented to exploit the nested structure of the data, composed of different time series of various meteorological factors over different monitoring stations in Michigan, feeding an LSTM model to improve the forecasting accuracy of PM2.5. 4 - Vertical Federated Learning for Anomaly Detection in Multi Component Cyber Physical Systems Paritosh Ramanan, Georgia Institute of Technology, Atlanta, GA, 30305-4240, United States Federated Learning (FL) is a distributed machine learning paradigm that accomplishes large scale learning tasks among multiple user devices with full data privacy. However, classical FL schemes assume homogeneity of features as well as labels across all user devices. In case of large scale multi component systems, classical FL might be infeasible owing to a heterogenous feature set that is scattered across all components/devices. In this talk we present a Vertical Federated Learning framework that eliminates the need to move data for multicomponent systems. Instead, our VFL based approach uses a combination of local and global embeddings to capture interdependencies in the performance at the component level. We demonstrate our results using anomaly detection for a large scale multi component system with a heterogenous feature set at the component level.

MC04 CC Ballroom D / Virtual Theater 4 Hybrid Emerging Issues in Supply Chain Finance and Risk Management Practice Sponsored: MSOM/iForm Sponsored Session Chair: Gill Eapen, Decision Options, LLC, Groton, CT, 06340, United States Co-Chair: Selvaprabu Nadarajah, Information and Decision Sciences, University of Illinois at Chicago, Woodridge, IL, 60517, United States 1 - Emerging Issues in Supply Chain Finance and Risk Management Practice Gill Eapen, Decision Options, Groton, CT, 06340, United States This panel explores emerging topics at the interface of finance, operations, and risk management. The increasing number of shocks (e.g., COVID, weather events) on integrated supply chains has placed a premium on quantifying their impact on physical and financial flows, as well as the need for forward looking design. Thought leaders from academia and industry will discuss this burgeoning theme by considering the interactions between financial and operating risks, the role of risk management, and the value of recent technologies (e.g., blockchains and digital platforms) and practices. 2 - Panelist Volodymyr O. Babich, Georgetown University, Washington, DC, 20057, United States 3 - Panelist John R. Birge, University of Chicago, Chicago, IL, 60637-1656, United States 4 - Panelist Aurelien Ouattara, Amazon Luxenburg, Luxenburg, Germany 5 - Panelist Nicola Secomandi, Carnegie Mellon University, Pittsburgh, PA, 15213-3815, United States MC05 CC Ballroom E / Virtual Theater 5 Hybrid TSL Award II Sponsored: Transportation Science and Logistics Sponsored Session Chair: Mike Hewitt, Loyola University Chicago, Glen Ellyn, IL, 60137- 5246, United States MC06 CC Room 303A In Person: OR for Vulnerable Populations General Session Chair: Shima Azizi, Shrewsbury, MA, 01545-4285, United States 1 - Assessing Transportation Barriers to Opioid Treatment in Tennessee Anna White, Ann Arbor, MI, 48103-2911, United States Lack of access to transportation keeps individuals from accessing regular treatment for opioid misuse disorder. In this work, we evaluate the transportation needs and potential investment solutions to improve access to treatment in Tennessee. We conduct case studies of rural, suburban, and urban regions. This work helps to better understand where transportation investments may be needed to connect individuals to treatment services. 2 - Sex Trafficking Analytics: Data, Predictions, And Interdictions Burcu B. Keskin, University of Alabama, Tuscaloosa, AL, 35406- 4062, United States, Nickolas K. Freeman, Gregory Bott Human traffickers have been using mobile technologies, online classified advertisement sites, and social media but the volume and frequency of ads and the obfuscation tactics complicate the law enforcement investigations. Analyzing over ten million records, our approach combines machine learning models with network theory to understand real/fake posts, identify patterns, predict the movement of the sex trafficking organizations, and inform interdiction efforts.

59