Informs Annual Meeting 2017

WB72

INFORMS Houston – 2017

2 - Steps to Success in Training Neural Networks by Second-order Algorithms Xi He, Lehigh University, 223 Summit Street, Bethlehem, PA, 18015, United States, xih314@lehigh.edu We move forward from the common used stochastic gradient based methods to stochastic Hessian-based approaches. They show capabilities to escape saddle points which are crucial in training neural networks, and also help achieve faster convergence rate and better model generalization. We investigate Hessian evolution along iterations to provide insights on neural network training. It motivates us to add elegant extra line-search and momentum to our proposed methods for better optimization performance. Numerical experiments show that our proposed methods can be more successful in practice at optimizing neural networks than several other fine-tuned first order learning algorithms. 3 - Applying Multidisciplinary Design Optimization to Service Network Design Problems Leah J. Ruckle, PhD Candidate, Georgia Institute of Technology, 270 Ferst Drive, Atlanta, GA, 30332, United States, leruckle@gatech.edu, Dimitri Mavris Multidisciplinary Design Optimization (MDO) is an optimization field that takes a “divide-and-conquer” approach to solving large and complex optimization problems. It is primarily used within the aerospace and mechanical engineering communities for the design of complex physical systems such as aircraft, automobiles and wind turbines. Up to this point, MDO has not been applied to logistical systems such as the Service Network Design Problem (SNDP). This presentation studies the suitability of applying multidisciplinary design optimization approaches to SNDPs and compares their performance to more traditional decomposition methods. 4 - Aircraft Recovery Problem Considering Airport Capacity and Maintenance Flexibility Xianfei Jin, Sabre, Southlake, TX, United States, Xianfei.Jin@sabre.com, Xiongwen Qian, Zhe Liang, Lei Zhou, Sureshan Karichery, Xuehua Lu Disruptions have a huge financial impact on the airline industry. When disruptions happen, airlines need to re-schedule flights and re-assign aircraft in real time with minimized recovery cost. In most previous works, airport capacity and flexible maintenance are not considered simultaneously via an optimization approach. To bridge this gap, we propose a column generation framework to solve the problem. Computational experiments show that the framework solves real- life problems with small, usually zero, optimality gap in very short run time. Chair: Jean-Paul Arnaout, Lebanese American University, School of Engineering-101F, P.O. Box 36, Byblos, Lebanon, jparnaout@lau.edu.lb 1 - Minimizing Vessel Service Costs in Berth Scheduling via Evolutionary Computation with Deterministic Parameter Control Maxim A. Dulebenets, Florida A&M.University-Florida State University, 2300 Bluff Oak Way, Apt. 8408, Tallahassee, FL, 32311, United States, mdlbnets@gmail.com This study focuses on the seaside operations of marine container terminals and proposes a novel Memetic Algorithm with a deterministic parameter control to facilitate the berth scheduling and minimize the total vessel service cost. Unlike canonical Evolutionary Algorithms, the developed algorithm uses a local search heuristic and adjusts the mutation rate values based on a deterministic parameter control strategy for more efficient exploration and exploitation of the search space. Computational experiments demonstrate efficiency of the developed algorithm. 2 - Storage Location Assignment for Rectangular Items Beneath a Gantry Crane Roya Ghorashi, University of Wisconsin- Milwaukee, 3200 N Cramer St, EMS1062, Milwaukee, WI, 53211, United States, ghorashi@uwm.edu, Matthew Petering Storage location assignment problems are critical to increasing productivity in many industries. In this paper we consider the problem of locating rectangular items in a temporary storage area beneath a gantry crane. We develop a mixed- integer linear programming model of the problem and devise a metaheuristic method that quickly obtains good solutions to large problem instances. Experimental results obtained via CPLEX and C++ demonstrate the effectiveness of the proposed methods. WB72 372A Optimization, Network Contributed Session

3 - Cross Dock Scheduling of Inbound and Outbound Trucks for Fresh Products Fei Pan, East China University of Science and Technology, 130# Meilong Road, Shanghai, 200237, China, 1004841702@qq.com Cross dock is a warehouse management concept which can reduce inventories, lead time and customer response time. One of the most fundamental issues is how well the trucks can be scheduled. Aiming at the feature of fresh products, we address three objectives respectively to optimize, the total operation time, the total deterioration of fresh products and the total products passing through temporary storage. Furthermore, we propose a mathematical model and design a genetic algorithm for real size applications. Through comparing the optimal scheduling sequence for trucks based on three objectives, we conclude that optimization of total deterioration can make three objectives low simultaneously. 4 - Solving the Modular Fixed Charge Transportation Problem – A Polyhedral Approach Faiz Hamid, Assistant Professor, Indian Institute of Technology, Kanpur, 208016, India, fhamid@iitk.ac.in, Yogesh K. Agarwal In this paper we study a variant of the fixed-charge transportation problem where the capacity can be installed only in integer multiples of a fixed module capacity. The problem is referred as modular fixed charge transportation problem in this paper. The problem arises frequently in practice, for example in supply chain vehicle dispatching problems. However, the problem itself has not been studied in the literature so far to the best of our knowledge. We formulate the problem as a mixed integer linear program, propose different valid inequalities to strengthen the formulation, and present some preliminary computational results. 5 - A Worm Optimization Algorithm to Minimize the Makespan on Unrelated Parallel Machines Jean-Paul Arnaout, Associate Professor, Gulf University for Science & Technology, West Meshref, 00965, Kuwait, jeanpaularnaout@gmail.com This paper addresses the unrelated parallel machine scheduling problem with setup times, with an objective of minimizing the makespan. A Worm Optimization (WO) algorithm is introduced and is applied to this NP-hard problem. The novel WO is based on the behaviors of the worm, which is a nematode with only 302 neurons. Nevertheless, these neurons allow worms to achieve several intricate behaviors including finding food, interchanging between solitary and social foraging styles, alternating between dwelling and roaming, and entering a type of stasis/declining stage. WO’s performance is evaluated by comparing its solutions to solutions of other known metaheuristics. Chair: Phebe T. Vayanos, University of Southern California, Los Angeles, CA, 90089, United States, phebe.vayanos@usc.edu 1 - Simulation Optimization via Parametric Robust Optimization Chaithanya Bandi, Northwestern University, 2001 Sheridan Road, Suite 566, Evanston, IL, 60208, United States, c- bandi@kellogg.northwestern.edu In this work, we build on data driven uncertainty sets and parameteric robust optimization to simulate and optimize complex stochastic systems such as queueing networks. 2 - Worst-case Demand Distributions in Vehicle Routing Mehdi Behroozi, Assistant Professor, Northeastern University, Boston, MA, 02115, United States, m.behroozi@neu.edu, John Gunnar Carlsson A recent focal point in research on the vehicle routing problem (VRP) is the issue of robustness in which customer demand is uncertain. In this paper, we conduct a theoretical analysis of the demand distributions whose induced workloads are as undesirable as possible. We study two common variations of VRP in a continuous approximation setting: the first is the VRP with time windows, and the second is the capacitated VRP, in which regular returns to the vehicle’s point of origin are required. 3 - Two-stage Robust Optimization with Decision-dependent Information Discovery Jiachuan Chen, University of Southern California, Los Angeles, CA, United States, jiachuac@usc.edu, Phebe T. Vayanos We consider two-stage robust optimization problems with decision-dependent information discovery in which some of the here-and-now decisions correspond to (costly) measurement variables that determine the portion of the uncertain parameters that will be observable in the recourse stage. We propose a solution approach based on the popular K-adaptability framework. We illustrate the performance of our approach on several stylized problems, including a variant of the capital budgeting problem from the literature. WB74 372C Robust and Data-Driven Optimization Sponsored: Computing Sponsored Session

500