Informs Annual Meeting 2017

TA37

INFORMS Houston – 2017

TA37

6 - Capacity Planning Analysis for Amazon.com Lockers Naser Nikandish, Assistant Professor, California State University, Fullerton, 800 North State College Boulevard, Fullerton, CA, 92831, United States, n.nikandish@gmail.com, Lawrence W. Robinson Amazon.com’s lockers are this company’s innovation to control their increasing fulfillment costs by decreasing last-mile distribution costs and to improve their shoppers’ experience by bringing convenience and safety to the package delivery process. In this project, we investigate online shoppers’ package retrieval behavior from an Amazon.com locker location and research its impact on capacity decisions at the Amazon.com locker locations. 351F Recent Progress in Efficient Algorithms in Convex Optimization Sponsored: Optimization, Nonlinear Programming Sponsored Session Chair: Qihang Lin, University of Iowa, Iowa City, IA, 52245, United States, qihang-lin@uiowa.edu 1 - New Results for Sparse Methods for Logistic Regression and Related Classification Problems Robert Michael Freund, MIT, MIT Building E62-567, 77 Massachusetts Ave., Cambridge, MA, 02139, United States, rfreund@mit.edu, Paul Grigas, Rahul Mazumder We propose and analyze methods in logistic regression that induce sparse solutions. We introduce a novel condition number associated with the data for logistic regression that measures the degree of non-separability of the data, and that naturally enters the analysis and the computational properties of methods. We take advantage of very recent new results in first-order methods for convex optimization due to Bach and others, to present new computational guarantees for the performance of methods for logistic regression. In the high-dimensional regime in particular, these guarantees precisely characterize how the methods impart both data-fidelity and implicit regularization, for any dataset. 2 - Sliding Techniques for Large-scale Optimization Guanghui (George) Lan, Georgia Institute of Tehnology, Newberry, FL, 32669, United States, george.lan@isye.gatech.edu We provide an overview of gradient/operator sliding techniques which lead to signifiant improvement on optimization algorithms. This includes savings on gradient computation, accesses to dataset and communication costs etc. 3 - Feasible Level Set Method for Constrained Convex Optimization Problems Negar Soheili Azad, University of Illinois-Chicago, 601 S.Morgan Street, University Hall 2416, Chicago, IL, 60607, United States, nazad@uic.edu, Qihang Lin, Selvaprabu Nadarajah First order methods have emerged as an effective solution strategy for tackling large scale convex optimization problems. The success of these methods has been greatest for such problems with simple feasible sets. We propose a novel feasible level set method that extends the applicability of these methods to convex optimization problems with potentially complicated constraint sets. 4 - Searching the Growth Rate: Adaptive SVRG Methods under Error Bound Conditions Qihang Lin, University of Iowa, Iowa City, IA, 52245, United States, qihang-lin@uiowa.edu, Yi Xu, Tianbao Yang Error bound condition (EBC), an intrinsic property of an optimization problem, can be exploited for developing algorithms with surprisingly fast global convergence. However, many existing approaches utilizing EBC requires knowing a growth parameter in the statement of EBC, which is usually very difficult to estimate, leading to a gap between the theoretical convergence and practical implementation. To address this issue, we propose a variance reduced stochastic gradient method that automatically searches for this unknown parameter on the fly of optimization, while still maintaining almost the same convergence rate just as the parameter is known. TA36

352B Design and Operation of Interdependent

Infrastructure Systems Sponsored: Service Science Sponsored Session Chair: Ann Suhaimi, Northeastern University, Somerville, MA, 02145, United States, nmsuhaimi@gmail.com Co-Chair: Jacqueline Griffin, Northeastern University, Boston, MA, 02115, United States, jacqueline.griffin@gmail.com 1 - Effect of Constructing Utility Tunnels on Interdependent Critical Infrastructure Resilience to Malicious Attacks Min Ouyang, Huazhong University of Science and Technology, Wuhan, China, min.ouyang@hust.edu.cn, Min Xu, Liu Hong We propose a bilevel attacker-defender model to analyze the effect of constructing utility tunnels on interdependent critical infrastructure resilience under the worst-case malicious attack. Results show that designing utility tunnels may largely reduce system resilience to the worst-case malicious attack. But if the utility tunnels are designed with sufficient redundancy or well hardened, interdependent system resilience could be well improved. 2 - An Interdiction-based Approach to Identify Damage in Interdependent Critical Infrastructures N. Orkun Baycik, PhD Student, Rensselaer Polytechnic Institute, 110 8th Street Center for Industrial Innovation, Suite 5015, Troy, NY, 12180, United States, baycin@rpi.edu, Thomas Sharkey We study the problem of determining the damage in multiple interdependent infrastructures given outage reports from the customers (or service receivers). Demand nodes within these infrastructure networks may have an outage either from damage that occurred within its infrastructure or due to cascading failures across infrastructures. We present a network interdiction based approach to identify the order in which components should be inspected in order to determine the damage within these networks based on the outages reported by customers. 3 - Employing a Mixed-strategy for Effective Response to Vulnerable Populations in Short-notice Disasters Rana Azghandi, Northeastern University, 360 Huntington Avenue, Boston, MA, 02115, United States, rana.azghandi@gmail.com, Jacqueline Griffin A mixed integer programming model is developed for applying a mixed-strategy framework to respond to needs of vulnerable populations in short-notice disasters. The mixed-strategy approach focuses on the joint deployment of resources for sheltering-in-place and evacuation assistance. Moreover, social features, such as the physical and emotional needs of target population, and physical features, such as infrastructure systems, will affect the optimal strategy. We examine the effect of these features on the effectiveness of the mixed strategy policy. 4 - Emergency Response Operations Accounting for Interdependent Infrastructures Ann Suhaimi, Northeastern University, 30 Adams St, Apt 7, Boston, MA, 02145, United States, nmsuhaimi@gmail.com, Jacqueline Griffin Demand for emergency response personnel is tightly coupled with the operation of other infrastructures, including power and communication systems, as well as the behaviors of citizens, particularly with respect to evacuation. Additionally, the demand for these services changes dynamically throughout an extreme event (e.g. snowstorm, hurricane, etc.). Using MIP model, we examine effects of coordinated decision making across infrastructures on the efficiency and effectiveness EMS deployment, accounting for the dynamics of demand for services. 5 - Optimizing the Resilience of Interdependent Infrastructure Networks under Pre-and Post-event Uncertainty Andres David Gonzalez, Assistant Professor, The University of Andres L. Medaglia, Mauricio Sánchez-Silva, Andrew J. Schaefer Optimizing the resilience of any given infrastructure system requires optimizing decisions for pre-event enhancement and damage mitigation, as well as for post- event efficient recovery. Moreover, it is imperative to consider the multiple interdependencies that these infrastructure systems depict, which directly impact their vulnerability and recoverability. In this work, we present a comprehensive optimization framework that allows optimizing both pre- and post-event decisions, based on the stochastic Interdependent Network Design Problem (sINDP), while taking into account diverse sources of uncertainty, associated with the failure of components and the demands and costs involved. Oklahoma, Norman, OK, 73071, United States, andres.gonzalez@ou.edu, Leonardo Dueñas-Osorio,

283