2016 INFORMS Annual Meeting Program

TA94

INFORMS Nashville – 2016

3 - System Optimal Transit Assignment With Flow-dependent Dwell Times Alireza Khani, University of Minnesota, 136 Civil Engineering Building, 500 Pillsbury Drive S.E., Minneapolis, MN, 55455, United States, akhani@umn.edu Travel time of transit vehicles between stops is constant with respect to passenger flow. However, dwell time at stops changes by more boarding and alighting of passengers. This makes the transit assignment an asymmetric problem. To model this phenomenon, an optimization problem is developed and its mathematical properties are investigated.

4 - Balancing Flexibility And Inventory In Workforce Planning With Learning And Uncertain Demand Silviya Valeva, University of Iowa, 710 Carriage Hill, Apt 5, Iowa City, IA, 52246, United States, silviya-valeva@uiowa.edu, Barrett Thomas, Mike Hewitt Explicitly modeling human learning in task assignment problems offers opportunities for better decision making that can result in both increased revenue and decreased lost sales. We present an assignment model incorporating both individual learning and uncertainty in demand and compare its performance to several myopic models. Our results demonstrate that cross training the workforce can be a successful way to hedge against uncertain demand. We further explore the use of practice assignments and inventory building as ways of creating capacity and preparing to meet future demand. 5 - Approximation Algorithms For Capacitated Perishable Inventory Systems With Positive Lead Times Xiting Gong, Assistant Professor, The Chinese University of Hong Kong, Room 506, William M. W. Mong Engineering Building, Shatin N.T., Hong Kong, xtgong@se.cuhk.edu.hk, Xiuli Chao, Cong Shi, Chaolin Yang, Huanan Zhang, Sean Zhou Perishable inventory system with positive lead time and finite ordering capacity is an important but notoriously difficult class of problems in both analysis and computation. Its optimal control policy is extremely complicated, and no effective heuristic policy has been proposed in the existing literature. In this paper, we develop an easy-to-compute approximation policy for this class of problems and show that it admits a theoretical constant-factor worst-case performance guarantee under most demand models of practical interest. Our numerical study shows that the proposed policy performs consistently well. TA94 5th Avenue Lobby-MCC Technology Tutorial: Frontline/ SAS-GAP/EDU Technology Tutorial 1 - Frontline: AnalyticSolver.com: Data Mining, Simulation And Optimization In Your Web Browser Daniel Fylstra, Frontline Systems, Inc., Incline Village, NV, Daniel@solver.com AnalyticSolver.com is the new, simple, point-and-click way to create and run analytic models using only your web browser - that also works interchangeably with your spreadsheet. Whether you need forecasting, data mining and text mining, Monte Carlo simulation and risk analysis, and conventional and stochastic optimization, you can “do it all” in the cloud. We’ll show how you can upload and download Excel workbooks, pull data from SQL Server databases and Apache Spark Big Data clusters, solve large-scale models, and visualize results - without leaving your browser. If you’re more comfortable working on your own laptop or server, we’ll show how you can do that, too. 2 - SAS: Analysis Of a Presidential Debate Using SAS Text Analytics André de Waal, Global Academic Program, Cary, NC, 27513, United States During the last year of a presidential term in the United States of America, the race to the White House has everybody excited. News channels and newspapers provide “expert” analysis of day to day events. However, many of the expert opinions are biased and reflect a commentator’s political viewpoint or affiliation. Can text mining be used to look at the data objectively and cut through the political rhetoric? In this talk, a script of one of the 2016 presidential debates is analyzed with SAS Text Miner. An attempt is made to look at the data “objectively” and to let the data speak. Words are counted and stemmed, documents are grouped into clusters, topics are identified and candidates are analyzed while trying to determine what separates one candidate from the rest of the field. Although it is impossible to predict using text mining alone who will win the presidential election, text mining could provide some insight into the election process (of which the debates are an integral part) that is not generally available to the general populace and might influence their choice of presidential candidate.

TA90 Broadway D-Omni Opt, Stochastic II Contributed Session

Chair: Xiting Gong, Assistant Professor, The Chinese University of Hong Kong, Room 506, William M. W. Mong Engineering Building, Shatin N.T., Hong Kong, xtgong@se.cuhk.edu.hk 1 - Modeling AS/RS Travel In Order Picking Applications Jingming Liu, Research Assistant, 1986, 920 N Leverett, Apt 824, Fayatteville, AR, 72701, United States, jl011@uark.edu, John A White An order picking operation is modeled as an M|G|1 queueing problem, with the S/R machine being the server and order picking stations being customers. Service time is the time required for an automated storage and retrieval machine to travel from an order picking station to a random storage location and, then, to an order picking station. To obtain the variance for service time, the density function is derived. Because the two travel times are statistically dependent random variables, results obtained previously for S/R travel do not apply. Random demands for replenishment of order picking stations and Chebyshev travel by the S/R machine are assumed. 2 - Stochastic Quasi-newton Methods For Non-strongly Convex Problems: Convergence And Rate Analysis Farzad Yousefian, Assistant Professor, Oklahoma State Univeristy, 317D Engineering North, School of Industrial Engineering & Management, Stillwater, OK, 74074, United States, farzad.yousefian@okstate.edu, Angelia Nedich, Uday Shanbhag Motivated by applications in machine learning, we consider stochastic quasi- Newton (SQN) methods. Traditionally, the convergence of SQN schemes relies on strong convexity. To our knowledge, no rate statements exist in the absence of this assumption. We consider merely convex problems and develop an SQN scheme where both the gradient mapping and the Hessian approximation are regularized and updated in a cyclic manner. Under suitable assumptions on the stepsize and regularization parameters, the convergence is shown in both almost sure and mean senses and the rate of convergence is derived in terms of function value. The empirical results on a binary classification problem are promising. 3 - Optimization With Reference-based Almost Stochastic Dominance Jian Hu, Assistant Professor, University of Michigan - Dearborn, 4901 Evergreen Rd., Dept. of IMSE,, Dearborn, MI, 48128, United States, jianhu@umich.edu Stochastic dominance is a preference relation of uncertain prospect defined over a class of utility functions. While this utility class represents basic properties of risk aversion, it includes some extreme utility functions rarely characterizing a rational decision maker’s preference. We introduce reference-based almost stochastic dominance (RSD) rules which well balance the general representation of risk aversion and the individualization of the decision maker’s risk preference. We also propose RSD constrained stochastic optimization model and develop an approximation algorithm based on Bernstein polynomials.

263