INFORMS 2021 Program Book

INFORMS Anaheim 2021

WC28

2 - Self-guided Approximate Linear Programs Parshan Pakiman, Information and Decision Sciences, University of Illinois at Chicago, Chicago, IL, 60605, United States, Selvaprabu Nadarajah, Negar Soheili, Qihang Lin Solving real-world Markov decision processes using reinforcement learning (RL) methods requires selecting approximation architectures (i.e., basis functions) and tuning parameters, limiting their deployment in practice. We develop an RL framework that solves a convergent sequence of approximate linear programs (ALPs), largely side-steps basis function selection by sampling random basis functions. It also self-tunes state-relevance weights, which are parameters that determine the approximation quality across states. Our methodology facilitates implementation, has strong theoretical guarantees, and outperforms existing control policies on two business applications. WC30 CC Room 207D In Person: Statistical Methods for Quality and Engineering General Session Chair: Rui Tuo, Texas A & M University, College Station, TX, 77845- 7399, United States 1 - High-dimensional Change-point Detection using Generalized Homogeneity Metrics Xianyang Zhang, Texas A&M University, College Station, TX, United States This work focuses on detecting abrupt distributional changes in high-dimensional data generating distribution. We develop a nonparametric methodology to detect an unknown number of change-points in an independent sequence of high- dimensional observations and to test for the significance of the estimated change-point locations. Our approach essentially rests upon nonparametric tests for the homogeneity of two high-dimensional distributions. We construct a single change-point location estimator via defining a cumulative sum process in an embedded Hilbert space. Subsequently, we combine our statistics with the idea of wild binary segmentation to recursively estimate and test for multiple change- point locations. The method is further extended to incorporate external graph information. 2 - Optimal Maintenance Planning and Budget Prioritization of the Deteriorating Co-located Road and Water Systems with Interdependencies Hung Quoc Nguyen, University of South Florida, Tampa, FL, United States, Noha Abdel-Mottaleb, Shihab Uddin, Qiong Zhang, Qing Lu, He Zhang, Mingyang Li For co-located interdependent systems, such as transportation & water systems (TS & WS), proactive and joint maintenance is one of the promising solutions to reduce costs and improve overall serviceability. Many of existing maintenance works for interdependent systems often focus on short-term reactive maintenance instead of long-term proactive maintenance planning. We propose a model-based analytics framework for long-term maintenance planning and budget prioritization of TS and WS with a high level of co-location by accounting for their interdependencies and varied spatial heterogeneity. A case study on a sub- region in the City of Tampa is provided to show the benefits of the proposed work. WC31 CC Room 208A In Person: Revenue Management with Customer Choice General Session Chair: Mika Sumida, Cornell Tech, Cornell Tech, New York, NY, 10128- 5805, United States 1 - Price Discrimination with Fairness Constraints Adam Elmachtoub, Columbia University, 560 Riverside Dr Apt 15a, New York, NY, 10027-3241, United States, Maxime Cohen, Xiao Lei Price discrimination allows sellers to increase their profits, but it also raises several concerns in terms of fairness which has received extensive attention from media, industry, and regulatory agencies. In this paper, we consider the problem of setting prices for different groups under fairness constraints. We first propose four definitions: fairness in price, demand, consumer surplus, and no-purchase valuation. We analyze the pricing strategy of a profit-maximizing seller and the impact of imposing fairness on the seller’s profit, consumer surplus, and social welfare.

WC28 CC Room 207B In Person: Learning and Decision-making with Contextual Information General Session Chair: Rui Gao, University of Texas at Austin, Austin, TX, 78712-1277, United States Co-Chair: Luhao Zhang 1 - Contextual Chance-Constrained Programming Hamed Rahimian, Clemson University, 2145 Tech Dr # C210, Clemson, SC, 60208-0884, United States, Bernardo Kulnig Pagnoncelli Uncertainty in classical stochastic programming models is often described solely by independent random parameters, ignoring their dependence on multidimensional features. We describe a novel contextual chance-constrained programming formulation that incorporates features, and argue that solutions that do not take them into account may not be implementable. Our formulation cannot be solved exactly in most cases, and we propose a tractable and fully data- driven approximate model that relies on weighted sums of random variables. Borrowing results from quenched large deviation theory we show the exponential convergence of our scheme as the number of data points increases. We illustrate our findings on real and synthetic data. 2 - Residuals-based Distributionally Robust Optimization With Covariate Information Rohit Kannan, Los Alamos National Laboratory, University of Wisconsin Madison, Los Alamos, NM, United States, Guzin Bayraksan, James Luedtke We consider data-driven approaches that integrate a machine learning prediction model within distributionally robust optimization (DRO) given limited joint observations of uncertain parameters and covariates. Our framework is flexible in the sense that it can accommodate a variety of learning setups and DRO ambiguity sets. We investigate the asymptotic and finite sample properties of solutions obtained using Wasserstein, sample robust optimization, and phi- divergence-based ambiguity sets within our DRO formulations, and explore cross-validation approaches for sizing these ambiguity sets. Through numerical experiments, we validate our theoretical results, study the effectiveness of our approaches for sizing ambiguity sets, and illustrate the benefits of our DRO formulations in the limited data regime even when the prediction model is misspecified. 3 - Optimal Policies for Robust Big Data Newsvendor Luhao Zhang We consider a robust big newsvendor problem that seeks an optimal end-to-end policy, under the Wasserstein robust framework that hedges against data uncertainty on demand and covariates.We develop an equivalent linear programming reformulation, by proving the optimality of a novel Shapley policy.This provides the first policy optimization framework that produces a robust optimal policy without restricting the policy class while still maintaining tractability. Numerical experiments on real and synthetic datasets demonstrate the competitive performance of our proposed policy. WC29 CC Room 207C In Person: Recent Advances in Reinforcement Learning General Session Chair: Parshan Pakiman, University of Illinois-Chicago, Chicago, IL, 60605, United States 1 - Biodiversity Preservation via Adjustable Robust Optimization Yingxiao Ye, University of Southern California, Los Angeles, CA, United States, Christopher Doehring, Angelos Georghiou, Hugh Robinson, Phebe Vayanos To protect biodiversity against human impact, existing methods purchase lands to maximize the value of the protected area with the given budget. However, budget is usually received progressively over time, and also, the existing models cannot capture the uncertainty in development. We propose a multistage, robust optimization problem with a data-driven uncertainty set to minimize the biodiversity loss due to human impact. We prove that the problem can be reformulated into a robust problem with exogenous objective uncertainty. The numerical results based on real data show that the proposed method outperforms the MARXAN, a conservation planning software, in 90% cases.

164