2016 INFORMS Annual Meeting Program

WA14

INFORMS Nashville – 2016

WA13 104C-MCC Recent Developments in Optimization Software Sponsored: Optimization, Computational Optimization and Software Sponsored Session Chair: Hans Mittelmann, Arizona State University, Box 871804, Tempe, AZ, 85287-1804, United States, mittelmann@asu.edu 1 - Selected Benchmark Results Hans Mittelmann, Arizona State University, mittelmann@asu.edu We will present results from selected benchmarks we are maintaining in both continuous and discrete optimization by both commercial and open source software. 2 - Recent Advances In The New Mosek 8 Andrea Cassioli, MOSEK, andrea.cassioli@mosek.com Erving Anderson In this talk we present the new features and improvements in the new MOSEK 8 solver. Improved presolving, automatic dualization for conic quadratic problems and other new developments in the core routines has lead to a significant improvements of the solver performance both in terms of speed and accuracy. Computational results will be presented and discussed. 3 - Recent Advances In The SCIP Optimization Suite Gregor Hendel, Zuse Institute Berlin, Takustrasse 7, Berlin, 14195, Germany, hendel@zib.de The general-purpose branch-and-cut solver SCIP is one of the fastest noncommercial software tools for solving mixed integer linear optimization problems. In this talk, we will give an overview of algorithmic advances in the upcoming release with a special focus on new and extended primal heuristics of SCIP. 4 - UG[PIPS-SBB, MPI]: A Massively Parallel Branch-and-bound Solver For Stochastic Mixed-integer Programs Yuji Shinano, Zuse Institute Berlin, shinano@zib.de, Lluis-Miquel Munguia, Geoffrey Malcolm Oxberry, Deepak Rajan PIPS-SBB is a LP-based branch-and-bound solver using a distributed-memory simplex algorithm that leverages the structure of stochastic mixed-integer programs (MIPs). However, it does not parallelize its branch-and-bound tree search. The Ubiquity Generator (UG) is a general framework for the external parallelization of mixed-integer programming solvers. It has been used to develop ParaSCIP, a massively parallel version of the academic constraint integer programming solver SCIP. In this talk, we will introduce a parallel solver ug[PIPS- SBB, MPI] in which PIPS-SBB’s branch-and-bound tree search is parallelized on top of the parallel solution of the LP relaxations. Chair: 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 1 - Inventory Control Policy For A Periodic Review System With Expediting Yi Tao, Assistant Professor, Guangdong University of Technology, 161 Yinglong Road, Tianhe District, Guang Zhou, 510520, China, kenjimore@gmail.com, Loo Hay Lee, Ek Peng Chew, Gang Sun, Charles Vincent We study a periodic review inventory system where two modes, regular and fast mode, are available to obtain replenishment. A firm can choose fast mode with shorter lead time at a higher cost when necessary. A two-replenishment-mode model, with random expediting points is established and an ordering policy (S,e) which replenishes the inventory to S in every cycle and expedites part of the order using fast mode when the inventory drops below e, is proposed. A simulation optimization based heuristic which uses infinitesimal perturbation analysis (IPA) method and gradient search algorithm is employed to find the best (S,e). Numerical experiments have shown our new policy outperforms existing policies. WA14 104D-MCC Inventory Management Contributed Session

4 - Optimization Of Cascading Processes In Multiscale Networks With Stochastic Interactions Oleg A Prokopyev, University of Pittsburgh, 1037 Benedum Hall, Pittsburgh, PA, 15261, United States, droleg@pitt.edu, Juan Borrero, Pavlo Krokhmal We study the problem of optimal cascade propagation in a network, where the cascade’s spread depends on a vector of given attributes. Given there are costs associated with changing the attributes’ values, a decision-maker desires to minimize the time until all network nodes receive influence, subject to a budget constraint. To this end, we propose a stochastic optimization model based on Markov chains. Under simple assumptions, we derive analytical solutions for the optimal budget allocation in terms of a minimum spanning arborescence on an auxiliary graph. These results establish that optimal solutions have a hierarchical structure, and show that they can be found in polynomial time. WA12 104B-MCC Optimization in Cyber Defense Sponsored: Optimization, Integer and Discrete Optimization Sponsored Session Chair: Les Servi, MITRE, 202 Burlington Road, Bedford, MA, 01730, United States, lservi@mitre.org Co-Chair: Doug Altner, MITRE Corporation, 7525 Colshire Drive, McLean, VA, 22102, United States, daltner@mitre.org 1 - A Supply Chain Network Game Theory Model Of Cybersecurity Investments With Nonlinear Budget Constraints Shukla Shivani, University of Massachusetts, Amherst, MA, United States, sshukla@som.umass.edu, Anna B Nagurney, Patrizia Daniele In our paper, we develop a supply chain network game theory model consisting of retailers that compete noncooperatively to maximize their expected profits and reduce network vulnerability by determining their optimal product transactions as well as cybersecurity investments subject to nonlinear budget constraints that include the cybersecurity investment cost functions. 2 - Optimal Scheduling Of Cybersecurity Analysts For Minimizing Risk Rajesh Ganesan, George Mason University, Fairfax, VA, 22030- 4422, United States, ashah20@masonlive.gmu.edu, Ankit Shah, Sushil Jajodia, Hasan Cam The talk presents a generalized optimization model for scheduling cybersecurity analysts to minimize risk (a.k.a maximize significant alert coverage by analysts) and maintain risk under a pre-determined upper bound. The paper tests the optimization model and its scalability on a set of given sensors with varying analyst experiences, alert generation rates, system constraints, and system requirements. 3 - Dynamic Scheduling Of Cybersecurity Analysts For Minimizing Risk Using Reinforcement Learning Rajesh Ganesan, George Mason University, Fairfax, VA, 22030- 4422, United States, ashah20@masonlive.gmu.edu, Ankit Shah, Sushil Jajodia, Hasan Cam The talk presents a reinforcement learning-based stochastic dynamic programming optimization model that incorporates the estimates of future alert rates and responds by dynamically scheduling on-call cybersecurity analysts to minimize risk (a.k.a maximize significant alert coverage by analysts) and maintain the risk under a pre-determined upper bound. 4 - A Two-stage Stochastic Shift Scheduling Model For Cybersecurity Workforce Optimization With On Call Options Doug Altner, MITRE Corporation, McLean, VA, United States, daltner@mitre.org, Les Servi This talk proposes a two-stage stochastic program for optimizing staffing and shift scheduling decisions at a 24/7 cybersecurity operations center with three shifts per day, several staffing and scheduling constraints, uncertain workloads and on call staffing options. We then show how near optimal solutions can be obtained in a few minutes without a full branch-and-price implementation.

367