INFORMS 2021 Program Book

INFORMS Anaheim 2021

SD24

SD24 CC Room 205A In Person: Data-Driven Healthcare Operations Management Flash Session Chair: Jing Dong, Columbia University, New York, NY, 10027-6945, United States 1 - Adaptive Clinical Trial Designs with Surrogates: When Should We Bother? Arielle Elissa Anderer, The Wharton School, PA, 19096-2455, United States, Hamsa Sridhar Bastani, John Silberholz Surrogate outcomes have long been used in clinical trials when the true outcome of interest is expensive, time consuming, or otherwise difficult to measure. In this work we propose optimal adaptive clinical trial designs that integrate surrogate and true outcomes, and we analytically and empirically characterize regimes where our designs are especially beneficial. 2 - Prediction-driven Surge Planning with Application in the Emergency Department Yue Hu, Columbia University, New York, NY, 10027-3203, United States, Carri Chan, Jing Dong Optimizing emergency department (ED) nurse staffing decisions to balance the quality of service and staffing cost can be extremely challenging, especially when there is a high level of uncertainty in patient-demand. Increasing data availability and continuing advancements in predictive analytics provide an opportunity to mitigate demand-rate uncertainty by utilizing demand forecasts. In this work, we study a two-stage prediction framework that is synchronized with the base (made months in advance) and surge (made nearly real-time) staffing decisions in the ED. We quantify the benefit of the more expensive surge staffing. We also propose a near-optimal two-stage staffing policy that is straightforward to interpret and implement. Lastly, we develop a unified framework that combines parameter estimation, real-time demand forecasts, and staffing in the ED. SD25 CC Room 205B In Person: Idea to Product to Business General Session Chair: Hallie Sue Cho, PhD, Vanderbilt University, Nashville, TN, 37203, United States 1 - The Role of Learning Mechanisms on Pivoting Success in Start-ups Stylianos Kavadias, Margaret Thatcher Professor of Innovation & Entrepreneurial pivots have become almost necessities for start-up companies to survive and succeed. However, it remains unclear how should entrepreneurs learn which options to pursue during these pivots. We build an evolutionary model based on the cultural evolution tradition and we identify circumstances where imitation (social learning) benefits more than the often advocated learning based on experiments (so called “scientific entrepreneurialism”). We identify how different learning mechanisms aid or hinder entrepreneurs during their pivoting efforts. 2 - Best in Class: The Effect of Relative Perceived Quality on Demand in the U.S. Automobile Industry Hallie Cho, Vanderbilt University, Nashville, TN, United States This paper explores which factors drive consumers to consider certain products together and which factors set apart the ultimate choice from the rest. Based on co-occurring product mentions in online customer reviews, we find which products are often considered together. Using aggregated customer review measures as a proxy for product quality, we investigate how relative quality amongst similar products influences market shares in a competitive market. Our findings help to understand what drives optimal distinctiveness from the consumer’s perspective and have important implications for an automobile manufacturer’s product strategy. 3 - Ai-assisted Multimodal Evaluation System for Design Assessment Chenxi Yuan, Northeastern University, Boston, MA, United States Design concept evaluation is a key process in new product development with a significant impact on the product’s success. In view of limited and biased concept evaluation caused by subjective judgment of designers, we propose a deep multimodal regression model as a potentially disruptive way to bridge this gap. Specifically, we develop a deep neural network enabling accurate and scalable prediction of overall and the attribute-level performance ratings of design concepts from product images and descriptions. We test and validate the model through experiments on a large footwear dataset with low MSE loss and high accuracy. Growth, University of Cambridge, Judge Business School, Cambrige, CB2 1AG, United Kingdom, Konstantinos Ladas

SD26 CC Room 206A In Person: Global Optimization and Computing Applications General Session Chair: Anna Svirsko, United States Naval Academy, Annapolis, MD, 21401, United States 1 - Minimization of a Particular Singular Value Michael C. Rotkowitz, Amazon, Palo Alto, CA, United States We consider the problem of minimizing a particular singular value of a matrix variable, neither the largest nor the smallest, with the matrix subject to various constraints. This simply stated but unstudied problem arises in control theory, where it serves as the main obstacle to computing metrics for stabilizability, controllability, and robustness. Prior work found fast methods for obtaining upper bounds, but scalable lower bounds remained elusive. We show how to achieve convex upper and lower bounds for this problem, beginning with using a Difference of Convex (DC) formulation. 2 - Risk Based Allocation of COVID-19 PPEs under Supply Shortages Gohram Baloch, University of Waterloo, Waterloo, ON, Canada, Fatma Gzara, Samir Elhedhli We consider a resource allocation problem for personal protective equipment (PPEs) by integrating government supply and procurement decisions with healthcare facilities’ PPE usage policy. We present a modelling framework to make these decisions simultaneously to minimize both infection risk and monetary cost to the government. We derive closed-form expressions under different objective criteria to present easy-to-use policies to decision-makers. A mixed-integer quadratically constrained program (MIQCP) is also proposed to handle real-life PPE distribution planning problems. An Ontario-based case is built to derive managerial insights. 3 - Two-stage Distributionally Robust Optimization in Natural Disaster Management Mohamed El Tonbari, ISyE Georgia Tech, Atlanta, GA, United States, Alejandro Toriello, George L. Nemhauser We are motivated by natural disaster applications where data is limited. We solve a two-stage distributionally robust optimization model with a Wasserstein ambiguity set, where the first stage is a facility location problem and the second stage is a fixed-charge transportation problem. We develop a column and constraint generation algorithm and handle the presence of binary variables in the second stage by leveraging the structure of our support set and of the second stage value function. 4 - Convexification of Disjoint Bilinear Programs Hyun-Ju Oh, Graduate Student, Purdue University, West Lafayette, IN, United States, Mohit Tawarmalani In this paper, we construct a hierarchy of relaxations for disjoint bilinear programs using double description (DD) method. We show that, at each level, our relaxations are at least as tight as the corresponding lift-and-project (L&P) relaxation. In contrast to L&P relaxations, our relaxations converge to the convex hull of the bilinear set in a finite number of iterations. We discuss various ways to tighten relaxations using vertex decompositions and discuss applications of our results to max-min problems. 5 - Creating Equitable Communities Through Natural Disaster Recovery Anna C. Svirsko, Assistant Professor, United States Naval Academy, Annapolis, MD, United States, Daphne Skipper, Tom Logan, Tommy Reeder, Christina Domanowski When a natural disaster strikes, a resident’s ability to access services such as food and gas, are crucial to rebuilding the community. Instead of simply rebuilding, a natural disaster can be used as a catalyst to create equitable communities where residents have increased access to these services. We develop an integer programming model that determines a recovery plan after a natural disaster which looks to provide access as quickly as possible while also considering equitable access in the long-term. We solve both the deterministic and robust model with data from Hurricane Florence to demonstrate the advantages of including uncertainty in the recovery process.

38