Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB84

One bottleneck in analyzing extreme events is that, by its own definition, tail data is often scarce. Conventional approaches fit data using justified parametric distributions, but the inherent bias-variance tradeoff in the parametric fitting can hinder the estimation reliability. We discuss approaches using distributionally robust optimization as a nonparametric alternative that, through a different conservativeness-variance tradeoff, can mitigate some of the statistical challenges in estimating tails. We discuss the solution approaches and statistical performances compared to the conventional methods. 2 - Distributionally Robust Expectation using Dominance Information Ruiwei Jiang, University of Michigan, 1205 Beal Ave., Ann Arbor, MI, 48109, United States, Yuanyuan Guo This talk discusses the expectation of a random function when the distributional information of the uncertain parameters consists of moment (e.g., mean, covariance, support) and probabilistic dominance information. We find that the expectation in this setting can be bounded using conic programming. Finally, we demonstrate the theoretical results via case studies on appointment scheduling. 3 - Faster Rates of Convergence of Stochastic Gradient Descent for Wasserstein Distributionally Robust Optimization Karthyek Murthy, Columbia University, Department of IEOR, 500 West 120th Street, New York, NY, 10027, United States, Jose Blanchet, Fan Zhang Among the various notions of distributionally robust optimization (DRO) schemes being investigated to beat “optimizer’s curse when performing optimization under uncertainty, Wasserstein distance based DRO has gained much attention recently because of its relationship with machine learning algorithms that successfully employ regularization. In this talk, we shall see how these Wasserstein based DRO formulations can be solved almost as fast, and in some cases even faster(!) than, the respective nonrobust schemes for a large class of useful models. Specifically, we establish faster rates of convergence by studying the strong convexity of robust objective functions. 4 - Distributionally Robust Hypothesis Testing Rui Gao, Georgia Institute of Technology, 755 Ferst Drive NW, ISyE Main Building, Atlanta, GA, 30332-0205, United States We develop an approach to hypothesis testing problems that find the optimal test for deciding an observation belongs to a certain family of distributions. Such family of distributions is a non-parametric data-driven set of hypotheses based on Wasserstein distance. Leveraging tools from distributionally robust optimization and chance-constrained optimization, we provide convex reformulations of such problems which render a nearly optimal test. n TC02 North Bldg 121B Joint Session OPT/Practice Curated: Stochastic Integer Program: Theory and Applications Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Jie Zhang, Virginia Tech, Blacksburg, VA, 24060, United States, 1 - Stochastic Programming Models for Coordinated and Non- Coordinated Natural Gas and Power Systems Dan Hu, Iowa State University, Ames, IA, 50010, United States, Sarah M. Ryan With the growing proportion of electric power generated from natural gas, greater coordination between the two energy systems is gaining importance. We formulate separate but interacting stochastic programs to model the existing non- coordinated daily operations of the natural gas and electric power systems and a single stochastic program for a coordinated alternative. Variable renewable energy generation and demand for gas other than for electric power generation are the uncertain quantities. Through numerical studies we assess the benefits of coordination in terms of the distribution of the daily operational cost. 2 - Polyhedral Analysis of the Single-node Fixed-charge Network Flow Problem under Uncertainty David Mildebrath, Rice University, Houston, TX, United States, Victor Gonzalez, Mehdi Hemmati, Andrew J. Schaefer We consider a two-stage extension of the single-node fixed-charge network flow (SNFCNF) problem with uncertain demands and capacities. The stochastic SNFCNF problem is an important problem arising throughout the fields of network design, vehicle routing, and elsewhere. We present the first polyhedral results for the associated polytope, and establish fundamental connections between facets of the single-scenario (i.e. deterministic) SNFCNF polytope and its stochastic extension. Finally, we provide computational examples to quantify the degree to which knowledge of the structure of each single-scenario sub-problem improves the solution to the extensive form of the stochastic problem.

n TB84 Hyatt, Russell Disaster and Disruption Management Contributed Session Chair: Shawn Bhimani, School of Business, University of Leicester, 918 Garrison Ridge Blvd, England, 37922, United Kingdom 1 - Performance of Alternative Linear Objective Functions in Effectiveness Fairness Trade Off Gokalp Erbeyoglu, Bogazici University, Bogazi i Universitesi Endustri, Muhendisligi Bolumu Bebek, Istanbul, Turkey, Umit Bilge The quality of a humanitarian response plan is judged by two contradicting objectives: timely and fair satisfaction of demand. In this study, we provide alternative objectives that can focus on effectiveness and fairness to distribute relief items in disaster response. In our multi-depot multi-vehicle routing model with split deliveries, tours are open. Also, the demand occurrences are not limited to the initial period. The proposed objective functions capture both the timeliness and satisfaction levels in a way that their progression is important, rather than their final values. The performance is investigated in comparison to various related performance measures in the literature. 2 - Applying Backup Agreements with Penalty Scheme under Credit Guarantee Mechanism Cheng-feng Wu, Hubei University of Economics, No. 8 Yangqiaohu Road Jiang-xia Wuchang, Wuhan, Hubei, 430205, China This study examines the issue of the strait of capital shortage in SME supplier and supply risks from the buyer’s perspective. The buyer builds a credit guarantee mechanism in which a SME supplier is guaranteed to finance the shortfall in capital from the financial institution.This study considers the backup agreement with penalty scheme in which the buyer places the reservation quantity at the backup supplier and pays the penalty for the reservation that has not been ordered. The buyer considers the backup agreement with penalty scheme as its sourcing strategy to manage its supply basis and increase its profits under credit guarantee mechanism. 3 - A Methodology for Infrastructure Resilience Metric Assessment Mohammad Najarian, University of Houston, 4722 Calhoun Road, Engineering Building 2, Houston, TX, 77204-4008, United States, Gino J. Lim Natural and manmade disasters are unavoidable and they may incur extravagant costs. Resilience enhancement activities, which aim at reducing the disasters’ impact, rely on resilience metrics to find areas for improvement and measure the effectiveness of enhancements. Several resilience metrics have been suggested . However, the question is that whether they reveal the characteristics determined by the underlying framework or not. We address this question by defining a valid resilience metric and a novel methodology for assessing the validity. This methodology combines the experimental design methods and statistical analysis techniques that can be utilized to examine the resilience metrics. 4 - How Supply Chains Break Shawn Bhimani, University of Leicester, England, 37922, United Kingdom We present an enhanced framework of supply chain vulnerability through an analysis of 200 years of supply chain failures. This builds on previous literature in supply chain management, disruptions and risk mitigation. Through our extensive analysis of past failure types and causes, we offer an updated framework to better understand, prevent and predict catastrophic supply chain failures. This is useful for continuity of operations in corporate, military, and humanitarian supply chains, which can save money and lives. It also has implications for supply chain resiliency, robustness and the potential inducement of disruptions.

Tuesday, 12:05PM - 1:35PM

n TC01 North Bldg 121A

Data-driven Distributionally Robust Optimization Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Rui Gao, Georgia Institute of Technology, Atlanta, GA, 30332-0205, United States 1 - Enhancing Statistical Performances in Extreme Event Analysis via Distributionally Robust Optimization Xinyu Zhang, Columbia University, NYC, NY, 10025, United States, Henry Lam, Clementine Mottet

306