2016 INFORMS Annual Meeting Program

WA16

INFORMS Nashville – 2016

2 - A Hybrid Approach To Centralized Drug Inventory Management In A Chain Of Pharmacy Stores Tom Brady, Purdue University, 1401 S US Hwy 421, Westville, IN, 46391, United States, tbradyjr@pnw.edu Inventory analysis has long been a fundamental component of operations research, decision analysis, and operations management. In this paper, we discuss a hybrid approach of inventory policy to pharmaceuticals applied to small pharmacy chain. Decentralized inventory management policy had been in place as the chain grew where the individual store managers were responsible for inventory decisions. After an analysis of the inventory from the chain perspective, it was theorized that by moving certain inventory policy decisions to the chain level rather than the store level could result in potential inventory savings while maintain customer service targets. 3 - Managing Inventory Of Perishable Products at Multiple Locations Fang Liu, Assistant Professor, Nanyang Technological University, S3-B2a-13, 50 Nanyang Avenue, Singapore, 639798, Singapore, liu_fang@ntu.edu.sg, Yun Fong Lim A retailer selling multiple perishable products with random demands over a single season. The warehouses, each with a limited storage capacity, have different store and retrieve costs. Before demand realizes, the products are stored to each warehouse and after demand realizes products are retrieved from the warehouses. We develop a two-dimension table to determine under optimality which warehouses are non-empty, and among the non-empty warehouses, the products are stored following a nested assortment structure. In the retrieval stage, it is optimal to retrieve a product from a warehouse with the lowest retrieve cost that contains the product. 4 - Solving Stochastic Inventory Problems By Integer Programming Reha Uzsoy, North Carolina State University, Dept. of Industrial & Systems Engg, 300 Daniels Hall Camps Box 7906, Raleigh, NC, 27695-7906, United States, ruzsoy@ncsu.edu, Jaap Arts, Ton de Kok, Seza Orcun We propose binary integer programming models to approximately solve finite- horizon stochastic inventory problems with stochastic demand. The models proceed by discretizing the PDF of the demand in each period in a manner emulating numerical integration. The resulting integer programs have a consecutive ones property which provides additional computational efficiency. Cost minimizing service levels are endogenously determined. Computational results are reported for a variety of single-stage stochastic inventory models. Large-scale Stochastic Mixed-integer Programs Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Gabriel Lopez Zenarosa, University of Pittsburgh, 3700 O’Hara Street, Benedum Hall 1048, Pittsburgh, PA, 15261, United States, glz5@pitt.edu 1 - PH-BAB: A Progressive Hedging Based Branch And Bound Algorithm Semih Atakan, PhD Student, University of Southern California, Los Angeles, CA, United States, atakan@usc.edu, Suvrajeet Sen Progressive Hedging (PH) is a well known algorithm for solving multi-stage stochastic convex optimization problems. Most previous extensions of PH for stochastic mixed-integer programs (SMIPs) have been implemented without convergence guarantees. In this talk, we present a new framework that shows how PH can be utilized while guaranteeing convergence to globally optimal solutions of SMIPs. We demonstrate the efficacy of the proposed framework through computational experiments. 2 - New Approaches For The Stochastic Network Interdiction Problem Eli Towle, University of Wisconsin-Madison, etowle@wisc.edu, Jim Luestke We investigate methods to solve the maximum-reliability stochastic network interdiction problem (SNIP). To begin, we introduce a novel reformulation of the SNIP extensive formulation. We then propose a path-based formulation of the SNIP. We present cuts for this new formulation which are dependent on the structure of the given interdicted arc probabilities. To solve this path-based SNIP formulation, we implement a branch-and-cut (BC) algorithm. Computational results demonstrate an improvement of this BC algorithm over a traditional Benders decomposition. WA16 105A-MCC

3 - Two-stage Stochastic Programming For Influence Maximization Hao-Hsiang Wu, Ohio State University, Columbus, OH, 43202, United States, wu.2294@osu.edu, Simge Kucukyavuz Influence maximization problem is to find top-k influential nodes in a social network. Kempe. et al propose the greedy algorithm as a heuristic to solve the linear threshold and independent cascade models for influence maximization. By using the greedy algorithm, only 63% optimal value can be guaranteed. In this project, we formulate influence maximization problem as a two-stage stochastic program, and use a delayed constraint generation algorithm for its exact solution. Our algorithm exploits the submodularity of the influence spread function. We demonstrate the performance of our algorithm on large-scale real-world datasets. 4 - Solving Large-scale Stochastic Programs Using Generalized Value Function Onur Tavaslioglu, PhD Student, University of Pittsburgh, Pittsburgh, PA, United States, ont1@pitt.edu, Andrew J Schaefer, Oleg A Prokopyev This work considers two-stage mixed integer programs with discretely distributed stochastic right-hand sides and objective functions. We present an equivalent superadditive dual formulation that uses the value functions of both stages. We introduce the generalized value function of a mixed-integer program simultaneously parametrized by its objective function and right-hand side. We describe fundamental properties of the generalized value function and three algorithms to calculate it. Then, we present a global branch-and-bound algorithm for solving one stage pure integer and one stage mixed integer program. We conclude with computational results. WA17 105B-MCC Convex and Conic Relaxations in Machine Learning Sponsored: Optimization, Nonlinear Programming Sponsored Session Chair: Amir Ali Ahmadi, Princeton University, 329 Sherrerd Hall, Princeton, NJ, 08540, United States, a_a_a@princeton.edu Co-Chair: Georgina Hall, Princeton University, Sherrerd Hall, Princeton, NJ, 08544, United States, gh4@princeton.edu 1 - A New Perspective On Boosting In Linear Regression Via Subgradient Optimization And Relatives Rahul Mazumder, MIT Sloan School of Management, rahulmaz@mit.edu Boosting is one of the most powerful and popular tools in machine learning/ statistics that is widely used in practice. They work extremely well in a variety of applications. However little is known about many of the statistical and computational properties of the algorithm, and in particular their interplay. We analyze boosting algorithms in linear regression from the perspective modern first-order methods in convex optimization. We derive novel comprehensive computational guarantees for several boosting algorithms, which provide a precise theoretical description of the amount of data-fidelity and regularization imparted by running a boosting algorithm. 2 - Making Sketchy Decisions: Semidefinite Programming With Optimal Storage Madeleine Udell, Cornell University, Ithaca, NY, United States, udell@cornell.edu, Joel Tropp, Alp Yurtsever, Volkan Cevher Is it possible to solve an optimization problem using far less memory than the natural size of the decision variable? In this talk, we propose an affirmative answer to this question when both the problem data and solution have a concise representation. We present an algorithm for provably solving many semidefinite programming problems whose natural size is O(n^2) using no more than O(n) memory.

368