INFORMS 2021 Program Book

INFORMS Anaheim 2021

SC26

the rising demand, disruption uncertainty, and related ripple effect. Accordingly, this paper studies the influence of disruptions on the logistics industry, considering the existence of third-party logistics services. We investigate related qualitative studies and structural analyses of the logistics network to understand how the ripple effect of disruption influences the supply chain network. 3 - A Disruptive Peer-to-peer Network System for Operations and Fintech in the Food Industry Jinwook Lee, Drexel University, Berwyn, PA, 19312-2512, United States, Sejong Yoon, Lanqing Du, Paul Moon Sub Choi For transparency in the supply chain, we apply and extend the existing notions of distributed ledger technology and internet-of-things to enhance authenticity verification radically. Additionally, we use novel stochastic optimization techniques for optimal rewards and fees of a P2P network. We also model the supply chain as a P2P network to robustly identify changes in a food information network. Using the suitable evaluation of peers’ effort, we calculate optimal rewards based on a utility function of rewards and effort. Furthermore, we evaluate the optimal structure within rewards and fees to maximize participants’ utility while guaranteeing network sustainability. SC26 CC Room 206A In Person: Global Optimization/Optimization with Noisy Intermediate-Scale Quantum 2 General Session Chair: Brandon Augustino, Lehigh University, Landing, NJ, 07850, United States 1 - Characterization and Mitigation of Errors in Quantum Computing via Consistent Bayesian Muqing Zheng, Lehigh University, Bethlehem, PA, 18015, United States Various noise models have been developed in quantum computing studies to describe the propagation and effect of the noise due to the imperfect implementation of hardware. While measurement errors are widely accepted to be modeled classically, the actual behaviors of gate errors are harder to identify. As a result, methods like Randomized Benchmarking (error characterization) and Randomized Compiling (error mitigation) are two that do not require the knowledge of gate error channels by taking the advantage of average behaviors. Different from those existing schemes, in this talk, we are going to stochastically model the error propagation and obtain its probabilistic information. 2 - An Inexact-Infeasible Quantum Interior Point Method (II-QIPM) for Linear Optimization Mohammadhossein Mohammadisiahroudi, Lehigh University, Bethlehem, PA, 18015, United States Quantum Linear System Solvers (QLSAs) have the potential to solve Newton systems in QIPMs much faster than classical solvers w.r.t dimension. However, the use of QLSAs in IPMs comes with many challenges, such as the impact of having ill-conditioned systems and the accuracy of QLSAs. We explore efficient use QLSAs in QIPMs. Accordingly, an II-QIPM is developed to solve LO problems. We also discuss how we can get an exact solution by Iterative Refinement without excessive time of QLSAs. Finally, the results of implementing our quantum method using quantum simulators are analyzed. 3 - Exact Global Optimization of Frame Structures for Additive Manufacturing Oguz Toragay, Research Assistant, Auburn University, Auburn, AL, United States, Daniel F. Silva, Alexander Vinel, Nima Shamsaei The problem of designing the lightest load-carrying planar frame structures for additive manufacturing (AM) is concerned with minimizing the structure weight by selecting discrete design elements and their continuous diameters. We focus on globally optimal solutions for the problem that is known to be computationally challenging, as it combines integer variables and non-convex constraints. We adapt existing formulations to allow for AM constraints and propose a new (non- convex) quadratic version. In numerical experiments, we show that with advanced solvers, the quadratic model performs best, even though it is still restricted to relatively small problem sizes. 4 - Improving Searchability in Evolutionary Algorithms: A Novel SFDM Method Reza Gharoie Ahangar, University of North Texas, Denton, TX, United States, Rebert Pavur This study proposes a search field division method (SFDM) to improve the searchability in evolutionary algorithms (EAs). To validate the proposed technique’s performance, we examine the diversity and exploration-exploitation search behaviors of the SFDM approach. The findings show that the proposed SFDM method can improve the performance of all EAs with different single- modal, multi-modal, and unimodal benchmark functions. This novel approach increases efficiency in EAs to find the best global optimal points to solve real- world industry problems.

SC24 CC Room 205A In Person: Applied Machine Learning in Operation General Session Chair: Gad Allon, University of Pennsylvania, Philadelphia, PA, 19104- 3615, United States 1 - Learning to Recommend Using Non-uniform Data Wanning Chen, Stanford University, Stanford, CA, United States, Mohsen Bayati Learning user preferences for products based on their past purchases or reviews is at the cornerstone of modern recommendation engines. One complication in this learning task is that some users are more likely to purchase products or review them, and some products are more likely to be purchased or reviewed by the users. This non-uniform pattern degrades the power of many existing recommendation algorithms, as they assume that the observed data is sampled uniformly at random among user-product pairs. We design a theory-driven weighted matrix completion method that restores the non-uniformity and, using real data, we show that it boosts the prediction performance of user preferences. 2 - Can AI Impact What We Eat in a Restaurant? Dmitrii Sumkin, INSEAD, Singapore, 138676, Singapore, Pavel Kireyev, Serguei Netessine We analyze 920 outlets of restaurants in Southeast Asia observed for almost over 2 years. They replaced paper menus with tablets, installed kiosks for ordering, and facilitated mobile phone usage to place an order on the website. The check level panel data include the sequence of goods added in a cart with their customization options regarding the dish size and toppings added. Staggered timing of AI implementation that gives recommendations along the ordering process allows identifying the causal impact of AI on customer’s choice. We study whether AI increases the check’s size and assortment and how it depends on the type of recommendation, type of order, the device used for the order, and other factors. 3 - Machine Learning and Prediction Errors in Causal Inference Machine learning is a growing method for causal inference. In machine learning settings, prediction errors are a commonly overlooked problem that can bias results and lead to arbitrarily incorrect parameter estimates. We consider a two- stage model where (1) machine learning is used to predict variables of interest, and (2) these predictions are used in a regression model for causal inference. Even when the model specification is otherwise correct, traditional metrics such as p-values and first-stage model accuracy are not good signals of correct second- stage estimates when prediction error exists. We show that these problems are substantial and persist across simulated and empirical data. We propose general methods to identify when prediction errors are biasing estimates and provide consistent corrections for the case where an unbiased subset of the data is available. SC25 CC Room 205B In Person: Nonlinear Optimization and Applications I General Session Chair: Jinwook Lee, Drexel University, Berwyn, PA, 19312-2512, United States 1 - Accelerating Quadratic Optimization With Reinforcement Learning Bartolomeo Stellato, Assistant Professor, Princeton University, Princeton, NJ, United States, Jeff Ichnowski, Paras Jain, Goran Banjac, Michael Luo, Joseph E. Gonzales, Ion Stoica, Francesco Borrelli, Ken Goldberg First-order methods for quadratic optimization such as OSQP are widely used for machine learning and embedded optimal control, where many related problems must be rapidly solved. These methods face two main challenges: hyperparameter tuning and convergence time to high-accuracy solutions. To address these, we explore how Reinforcement Learning can learn a policy to adapt hyperparameters to accelerate convergence. Our RL policy, RLQP, generalizes well to previously unseen problems with varying dimension and structure from different applications, including the QPLIB, Netlib LP, and Maros-Meszaros problems. RLQP outperforms state-of-the-art QP solvers including Gurobi and OSQP. 2 - Decision Making for the Disrupted Supply Chain Using Timestamped Location Graph Representation Lanqing Du, Drexel University, Philadelphia, PA, United States, Jinwook Lee COVID-19 has posted a profound influence on the flexibility and resilience of the supply chain management. Those factors will urge logistics industries to seek flexible routing strategies to build resilience and mitigate disruptions. Third-party logistics services are widely used within the logistics industry, and it would be considered as one of many essential logistics fulfillment alternatives when facing Daniel Chen, University of Pennsylvania, Philadelphia, PA, United States, Gad Allon, Zhenling Jiang, Dennis J. Zhang

23