Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

PLENARY

Sunday, 3:10PM - 4:00PM

4 - Distributionally Robust Dual Dynamic Programming Daniel Duque, Northwestern University, 2145 Sheridan Rd, Evanston, IL, 60208, United States, David Morton We consider a multi-stage stochastic linear program that lends itself to solution by stochastic dual dynamic programming (SDDP). In this context, we consider a distributionally robust variant of the model with a finite number of realizations in each stage, and distributional robustness is with respect to the probability mass function governing candidate realizations. We describe a computationally tractable variant of SDDP to handle this model using the Wasserstein distance to characterize distributional uncertainty. n SD02 North Bldg 121B Optimizing Policies in Systems Under Uncertainty Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Omid Nohadani, Northwestern University, Evanston, IL, 60208- 3119, United States 1 - Performance Bounds for Multistage Distributionally Robust Optimization Eojin Han, Northwestern University, 2145 Sheridan Road, Tech C210, Evanston, IL, 60208, United States, Chaithanya Bandi, Omid Nohadani Distributionally robust optimization for multistage problems are known to be computationally challenging. Several approximative policies have been proposed, however, without bounds on their performance. For these problems, we study finitely adaptable policies and provide performance guarantees along with tractable implementation. The proposed bounds are established only based on the geometry of the underlying ambiguity set. Therefore, these bounds are general and independent of other problem parameters. We examine their performance on a newsvendor problem, allowing further insights. 2 - Sustainable Inventory with Robust Periodic-affine Policies and Application to Medical Supply Chains Omid Nohadani, Northwestern University, 2145 Sheridan Road, Technological Institute M233, Evanston, IL, 60208-3119, United States, Eojin Han, Chaithanya Bandi We introduce a new class of adaptive policies called periodic-affine policies, that allow to optimally manage large-scale newsvendor networks under demand uncertainty. These policies are data-driven and can model, e.g., correlation. We then generalize the model to multi-product settings and multi-period problems. We show that these policies are sustainable, i.e. time consistent. This approach is tractable and free of distributional assumptions, hence, suited for real-world applications. We provide efficient algorithms and demonstrate them on one of India’s largest pharmacy retailers. 3 - Distributionally Robust Newsvendor with Variation Distance: Calibration under Limited Information and Application to Reserving Operating Room Time Hamed Rahimian, Northwestern University, 1971 Neil Avenue, 210 Baker Systems Building, Evanston, IL, 60208, United States, Guzin Bayraksan, Tito Homem-de-Mello We use distributionally robust stochastic optimization to model a general class of newsvendor problems. We form the ambiguity set by all demand distributions whose total variation distances from a nominal distribution are bounded by a level of robustness. For this problem, we characterize the optimal solution and the regions of demand that are critical to the optimal cost. We also establish quantitative relationships between the distributionally robust model and the corresponding risk-neutral and classical robust optimization models. We use our analyses to recommend an appropriate level of robustness for an operating room time reservation problem. 4 - Data-Driven Distributionally Robust Stochastic Programming for Correlated Data Mihai Anitescu, Argonne National Laboratory, Argonne, IL, United States, Xialiang Dou We present a distributionally robust formulation of a stochastic optimization problem for non-i.i.d vector autoregressive data. We use the Wasserstein distance to define robustness in the space of distributions and we show, using duality theory, that the problem is equivalent to a tractable, finite concave-convex saddle point problem. The performance of the method is demonstrated on both synthetic and real data

n Plenary West Bldg 301AB Plenary: Networks to Save the World: OR in Action Plenary Session Chair: Timothy R. Anderson, Portland State University, Mail Code ETM, Anna B. Nagurney, University of Massachusetts-Amherst, Isenberg School of Management, Dept of Operations & Information Mgmt, Amherst, MA, 01003, United States Networks have evolved to become the foundational framework for a wide range of applications, beginning with classical transportation and logistics problems. In this talk, I will feature network-based results in multiple areas such as: perishable product supply chains from food to healthcare, disaster relief, cybersecurity, and the new Internet, and illustrate how netwORks are transforming our understanding of the world and enhancing our futures. P.O. Box 751, Portland, OR, 97201, United States 1 - Networks to Save the World: OR in Action

Sunday, 4:30PM - 6:00PM

n SD01 North Bldg 121A Joint Session OPT/Practice Curated:

Applications of Stochastic Programming Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Amanda G. Smith, University of Wisconsin-Madison, Madison, WI, 53706, United States 1 - Leveraging Decomposition Methods to Design Robust Policies for Markov Decision Processes Lauren N. Steimle, University of Michigan, 3261 Bolgos Circle, Ann Arbor, MI, 48105, United States, Brian T. Denton Markov decision processes (MDPs) are commonly used to derive optimal sequential decision-making policies. However, these policies can underperform if the true model parameters differ from the estimates used in the optimization process. To address this issue, the Multi-model MDP has been proposed as a way to find a policy that performs well with respect to multiple models of the MDP parameters. Finding a policy that maximizes a weighted value across the models can be viewed as a two-stage stochastic program, and leveraging this connection, we develop exact and approximate solution methods that can be used to generate MDP policies that are robust to deviations in model parameters. 2 - Designing Resilient Distribution Systems Under Natural Disasters Sadra Babaei, Oklahoma State University, 322 Engineering North, Industrial Engineering & Mgmt, Stillwater, OK, 74078, United States This talk proposes a distributionally robust model for designing a distribution power system to withstand the risk of disruptions imposed by natural disasters. Using the moment information of asset failures, we build an ambiguity set of probability distributions of system contingencies. On the one hand, we consider the stochasticity of natural disasters and provide a less conservative configuration than that by a robust optimization approach. On the other hand, our model considers distributional ambiguity and so is more reliable than stochastic programming. We recast the model as a two-stage robust optimization formulation and solve it using the Column-and-Constraint Generation framework. 3 - Optimizing Flux Bound Change Decisions In Metabolic Engineering Amanda Smith, University of Wisconsin-Madison, Fitchburg, WI, 53711, United States, James Luedtke Bilevel mixed-integer programming models are frequently used to solve metabolic engineering problems. However, these models tend to push cells close to infeasibility, and solutions may yield non-viable organisms. To overcome this drawback, we investigate three new bilevel MIP models that attempt to offer a compromise between desired output and organism viability. First, we introduce a bi-objective top-level problem. We then augment this model by incorporating enzyme kinetics and conclude with a stochastic extension that accounts for uncertainty in cellular behavior.

90