Informs Annual Meeting 2017

SC18

INFORMS Houston – 2017

4 - Effects of Inventory Efficiency, Productivity and Responsiveness on Operational Performance Gulsah Hancerliogullari, Istanbul Technical University, Macka Campus Faculty of Management, Department of Industrial Engineering, Istanbul, 34367, Turkey, gulsah_2207@hotmail.com, Emra H. Koksalmis Inventory management has been a well-studied area of research in operations management. The purpose of this study is to examine the relationship between inventory management performance including inventory efficiency, inventory productivity and inventory responsiveness, and firm operational performance. We use financial data for publicly listed various U.S. industries available at Standard & Poor’s Compustat database using Wharton Research Data Services (WRDS). 5 - Dual-index Policies for Two-echelon Inventory Systems with Dual Delivery Modes and Batch Orders Qiang Wang, University of Science and Technology of China, Hefei, China, qiangwan@mail.ustc.edu.cn, Jie Wu, Chaolin Yang, Xiang Ji In this paper, we consider a periodic-review, infinite horizon two-echelon inventory system with dual delivery modes. These two modes provide different lead times and incur different costs. The downstream stage faces random customer demands. The objective is to find inventory policies that minimize the expected long-run average cost of the system over an infinite planning horizon. Due to the complex system structure, the optimal inventory policy is unknown. In this article, we study a class of simple inventory policies, called dual-index echelon-base-stock policy, which has been proved to be near-optimal for the single-echelon inventory system with dual-sourcing. 340A Operations Management and Learning in Applied Probability Sponsored: Applied Probability Sponsored Session Chair: Kuang Xu, Stanford University, 655 Knight Way, Stanford, CA, 94305, United States, kuangxu@gmail.com 1 - Dynamic Resource Allocation under Information Loss Yuan Zhong, University of Chicago / Booth School of Business, 5807 S.Woodlawn Avenue, Chicago, IL, 60637, United States, Yuan.Zhong@chicagobooth.edu, Kuang Xu We consider the problem of efficient dynamic resource allocation under information loss, in general stochastic processing systems. Information loss is modeled as an information channel. We characterize the capacity region under various channel models, and analyze the role of memory in policy design. 2 - Spectrum Estimation from a Few Entries Ashish Kumar Khetan, University of Illinois, 104 South Matthews Avenue, Urbana, IL, 61820, United States, khetan2@illinois.edu, Sewoong Oh Singular values of a data provide insights on the structure of the data. However, in many applications, we only get a partial observation. Under such scenarios, we consider the fundamental problem of recovering spectral properties of the underlying matrix from a sampling of its entries. We are particularly interested in recovering the spectrum, which is the set of singular values, and also in recovering a spectral sum function, which is an aggregate sum of a function applied to each of the singular values. We propose first estimating the Schatten k- norms of a matrix, and then applying Chebyshev approximation to the spectral sum function or applying moment matching to recover the spectrum. 3 - On the Performance of the M/coxian-2/n System with Preemption Xin Liu, Arizona State University, Tempe, AZ, United States, liuxincell@gmail.com This presentation discusses the performance of the M/Coxian-2/N system, where the mean service time in the second phase is longer than that in the first phase. The preemption policy assigns the priority to jobs in the first phase. When there are phase-1 jobs in the queue, the policy terminates phase-2 jobs in service and allocates freed servers to phase-1 jobs. The terminated phase-2 jobs are put back to the queue. We use the fluid solution to approximate the number of phase-1 jobs and the number of phase-2 jobs in the system; and use Stein’s method to quantify the approximation error. We will discuss the implications of these bounds on the performance of the M/Coxian-2/N system with preemption. SC17

4 - A Lower Bound on the Queueing Delay in Resource Constrained Load Balancing

Martin Zubeldia, MIT, Cambridge, MA, United States, mar.zubeldia@gmail.com, David Famarnik, John Tsitsiklis

We consider a distributed service model where jobs with unit mean, but general distribution, arrive and are immediately dispatched to one of several queues associated with n servers. We assume that the dispatching decisions are made by a central dispatcher endowed with a finite memory, and with the ability to exchange messages with the servers. We develop a novel approach to show that every symmetric dispatching policy with a message rate proportional to n, and with a memory of the order of log(n) bits, results in an expected queueing delay which is bounded away from zero, uniformly in n.

SC18

340B Stochastic Optimization in Applied Probability Sponsored: Applied Probability Sponsored Session Chair: Alessandro Arlotto, Duke University, Durham, NC, 27708, United States, aa249@duke.edu 1 - Process Flexibility for Multiperiod Production Systems Yuan Zhong, University of Chicago / Booth School of Business, 5807 S.Woodlawn Avenue, Chicago, IL, 60637, United States, yz2561@columbia.edu, Cong Shi, Yehua Wei We consider process flexibility in a multi-period make-to-order production (MTO) system. First, using a new chaining notion, termed the Generalized Chaining Gap (GCG), we prove that in a general system with high utilization, a sparse flexibility structure with m+n arcs is needed to achieve similar performance as full flexibility, where m and n are the number of plants and of products, respectively. We also provide a simple and efficient algorithm for finding such sparse structures. Moreover, we show that the requirement of m+n arcs is necessary, as for some systems, even the best flexibility structure with m+n-1 arcs cannot achieve the same asymptotic performance as full flexibility. 2 - Robust Data-driven Stochastic Programming from a Bayesian Perspective Harsha Honnappa, Purdue University, 315 N. Grant Street, West Lafayette, IN, 47906, United States, honnappa@purdue.edu, Prateek Jaiswal, Vinayak Rao We study the problem of solving a two stage stochastic program with recourse, where the expectation is with respect to a Bayes predictive distribution. We consider two settings: first, a ‘classical’ Bayesian setting identifying the optimal decision when the likelihood is assumed to be known precisely. Second, an ‘expert’ posits a likelihood and her uncertainty about this assessment as a Kullback-Liebler ball centered at the likelihood. We then perform a minimax optimization to identify the the set of optimal decisions that are robust against the experts uncertainty. In either case, we provide large sample analysis of the effect of learning the predictive distribution on the value of the program. 3 - Escaping the Local Minima via Simulated Annealing: Optimization of Approximately Convex Functions Tengyuan Liang, University of Chicago, Chicago, IL, United States, Tengyuan.Liang@chicagobooth.edu, Alexandre Belloni, Alexander Rakhlin, Hariharan Narayanan We consider the problem of optimizing an approximately convex function over a bounded convex set using only function evaluations. The problem is reduced to sampling from an approximately log-concave distribution using the Hit-and-Run method, which is shown to have the same complexity as sampling from log- concave distributions. In addition to extend the analysis for log-concave distributions to approximate log-concave distributions, the implementation of the 1-dimensional sampler of the Hit-and-Run walk requires new methods and analysis. The algorithm then is based on simulated annealing which does not relies on first order conditions which makes it essentially immune to local minima. 4 - A O(log N)-optimal Policy for the Dynamic and Stochastic Knapsack Problem with Equal Values Xinchang Xie, Duke University, xinchang.xie@duke.edu, Alessandro Arlotto We study a dynamic and stochastic knapsack problem in which a decision maker is sequentially presented with n items and needs to select which items to include in a knapsack with fixed capacity. Arriving items have non-negative, independent sizes with common continuous distribution, and every decision is terminal upon arrival. The goal is to maximize the expected number of selections under capacity constraint. An adaptive online policy is proposed and proved to be within O (log n) of the optimal, under mild regularity conditions on the distribution. We also discuss how the distribution of the number of selections under this adaptive policy compares with the number of selections under the optimal policy.

73