Informs Annual Meeting 2017

KEYNOTE

INFORMS Houston – 2017

2 - An Alternate Way of Real Options Valuation with Loss Function Kyongsun Kim, Auburn University, Auburn, AL, United States, kimkyon@auburn.edu, Chan Park Adopting the financial option pricing method for real options implicitly assumes the project value follows a geometric Brownian motion. Because of many practical difficulties of formulating real option strategies based on financial option framework, we propose an alternative method for valuing non-financial assets with the standardized loss function that provides practically identical results with the well-known Black-Scholes formula. 3 - Obtaining Lower Bounds for Multi-stage Risk-averse Stochastic Mixed-integer Programs Ge Guo, Iowa State University, 3004 Black Engineering Building, Iowa State University, Ames, IA, 50011-2164, United States, geguo@iastate.edu, Sarah M.Ryan Traditional stochastic programming is risk-neutral in the sense that it is concerned with the optimization of an expectation criterion. In many applications, however, the decision makers are more concerned about large losses. Thus, risk-averse optimization has attracted attention in stochastic programming. We present a method to obtain convergent and tight lower bounds from the Progressive Hedging algorithm (PHA) for multi-stage risk-averse stochastic mixed-integer programs focusing on time-consistent risk measures. This method can assess solution quality for PHA and also integrate with exact algorithms that rely on lower bounds. We report computational results on lot sizing instances. 4 - Identifying Error- and Attack-resilient Clusters in Networks under Decision-dependent Uncertainties Hossein Dashti, hdashti@email.arizona.edu, Pavlo A. Krokhmal Network robustness is crucial in many areas such as energy and defense. Possible disruptions may affect network’s functionality. One of the key robustness requirements is every pair of nodes are connected through number of short intermediate links. Moreover, having multiple paths makes such a cluster more robust. The cluster represents a R-robust 2-club, which is a subgraph with at least R disjoint paths of length at most 2 connecting each pair of nodes. Uncertainty manifested as stochastic errors/attacks in different nodes. If one can reinforce the network components against future threats, the goal is to determine optimal reinforcements that would yield a cluster with minimum risk of disruptions. Ye Wang, University of Southern California, 3767 Clarington Avenue, Apt 206, Los Angeles, CA, 90034, United States, wang141@usc.edu Wasserstein distance is a statistical metric that describes a distance function between two probability distributions. In our research, we consider the entropy maximization problem and the highest density region optimization problem in which we will search through all distributions whose Wasserstein distance to the empirical distribution of a given set of data points is sufficiently small. 2 - Robust Binary Linear Programming under Implementation Uncertainty Jose E. Ramirez, PhD Student, Texas A&M.University, 3131 TAMU, College Station, TX, 77843, United States, ramirez.jose@tamu.edu, Victor J. Leon This paper studies robust binary linear programming in the presence of uncertainties that may prevent the implementation of solutions exactly as prescribed. A formal model of this type of uncertainty, termed implementation uncertainty, is presented and used to develop a robust binary linear programming formulation under implementation uncertainty. The solution approach involves a mixed-binary linear program reformulation used in conjunction with identified structural properties, allowing the generation of a set of optimal solutions. Three methods to control the level of conservatism are developed including a cardinality-constrained reformulation of the original problem. 3 - A Robust Optimisation Approach to the Manpower Planning Problem Gar Goei Loke, Public Service Division, 9 Jalan Grisek, Singapore, 419440, Singapore, e0012863@u.nus.edu, Melvyn Sim A Robust Optimisation approach is applied to the manpower planning problem of optimal recruitment and progression through a hierarchy, under constraints of budget, productivity, headcount and managerial span of control. The model also incorporates employees’ time-in-grade as a determinant, and is endogenic in this sense. A Satisficing approach was adopted, which guarantees that the probability of violation of constraints is controlled up to T time periods. The effectiveness and robustness of the model is tested against existing methods in the literature and compared. WA83 382C Optimization, Stochastic Contributed Session Chair: Mehdi Behroozi, Northeastern University, Boston, MA, United States, m.behroozi@neu.edu 1 - Applications of Wasserstein Distance in Distributionally Robust Optimization

4 - A Robust Optimization Approach to Transshipment Problem under Demand Uncertainty Nafiseh Ghorbani Renani, Graduate Research Assistant, The University of Oklahoma, 519 South University Blvd, Apt 21, Norman, OK, 73069, United States, Nafiseh.ghorbani.renani-1@ou.edu In each country, there are a number of food supply companies which are in charge of distributing food properly throughout cities. In fact, the main duty of these organizations is preventing of the food shortage. In such a company, the role of network flow management has become bold. Since demand is uncertain in this situation, two different probabilistic linear programming models have been suggested by using stochastic method and robust optimization method to find the optimum solution. This study is mainly focusing on a multi-period robust optimization model for transshipment of N products among N cities for a food supply company. 5 - Data Driven Robust Optimization for Logistics and Transportation Problems Mehdi Behroozi, Assistant Professor, Northeastern University, Boston, MA, 02115, United States, m.behroozi@neu.edu, John Gunnar Carlsson Using Wasserstein metric in defining the ambiguity set in robust optimization allows us to circumvent common overestimation that arises when other procedures are used. In this paper, we consider a distributionally robust version of the Euclidean travelling salesman problem in which we compute the worst-case spatial distribution of demand against all distributions whose Wasserstein distance to an observed demand distribution is bounded from above. We compare the advantages of this method with other approaches for describing the region of uncertainty, such as taking convex combinations of observed demand vectors or imposing constraints on the moments of the spatial demand distribution. Keynote GBCC- General Assembly B, Level 3 Keynote: A Probabilistic Theory of Deep Learning Invited Session 1 - A Probabilistic Theory of Deep Learning Rich Baraniuk, Rice University A grand challenge in machine learning is the development of computational algorithms that match or outperform humans in perceptual inference tasks that are complicated by nuisance variation. For instance, visual object recognition involves the unknown object position, orientation, and scale in object recognition while speech recognition involves the unknown voice pronunciation, pitch, and speed. Recently, a new breed of deep learning algorithms have emerged for high- nuisance inference tasks that routinely yield pattern recognition systems with near- or super-human capabilities. But a fundamental question remains: Why do they work? Intuitions abound, but a coherent framework for understanding, analyzing, and synthesizing deep learning architectures has remained elusive. We answer this question by developing a new probabilistic framework for deep learning based on the Deep Rendering Model: a generative probabilistic model that explicitly captures latent nuisance variation. By relaxing the generative model to a discriminative one, we can recover two of the current leading deep learning systems, deep convolutional neural networks and random decision forests, providing insights into their successes and shortcomings, a principled route to their improvement, and new avenues for exploration. Keynote GBCC- Grand Ballroom A, Level 3 Keynote: Smarter Tools for (Citi)Bike Sharing: Cornell Rides Tandem with Motivate Invited Session 1 - Smarter Tools for (Citi)Bike Sharing: Cornell Rides Tandem with Motivate David Shmoys & Shane Henderson Cornell’s School of Operations Research and Information Engineering (ORIE) has been working with the bike-sharing company Motivate with an emphasis on its New York system Citi Bike since it began operations in 2013. Cornell ORIE provides data analysis and advice about strategy and operations to Motivate, which operates most of the leading bike-sharing programs in the United States. We will describe a suite of models and algorithms that provide data-driven decision-making tools not just for operations but also for strategic system planning. Wednesday, 9:40 - 10:30AM

471