Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MB71

Applications using these enhancements may simultaneously exploit two different kinds of parallelism, one within subproblems and another across multiple subproblems. We also discuss other changes to PEBBL for the 2.0 release. 2 - A Constraint Based Snowplow Optimization Framework Joris Kinable, Eindhoven University of Technology, De Lismortel 62, Eindhoven, 5612 AR, Netherlands, Willem van Hoeve, Stephen F. Smith Snow plow optimization is a complex process in which routes for a fleet of vehicles have to be determined, while adhering to vehicle operating restrictions, resource utilization and replenishment constraints. We present an optimization framework for a real-world snow plow optimization problem in Pittsburgh, USA. Feasible routes and schedules are calculated using heuristics and a Constraint Programming approach. Commercial GIS data is used as the underlying data source. Evaluations are performed against commercial snow plow optimization software. In addition, pilot tests with actual vehicles are being conducted. 3 - Team of Teams: Optimal Communication Structures for Concurrent Solvers Oleg Shylo, University of Tennessee, 523 John D. Tickle Engineering Building, 851 Neyland Drive, Knoxville, TN, 37996, United States, Andrii Berdnikov The focus is on the teams of concurrent optimization solvers that can attain high levels of parallelism required by extreme-scale computing. To explore theoretical properties of communication and its impact on computational performance, we model communication between individual solvers by stochastic processes. These theoretical constructs enable efficient evaluation of communication topologies without empirical testing, and are used to define a class of scalable communication design problems. 4 - A State-Space Decomposition Algorithm for the Integrated Last-Mile Transportation Problem David Bergman, University of Connecticut, Storrs, CT, 02451, United States, Arvind U. Raghunathan, John Hooker, Thiago Serra, Shingo Kobori In this talk we discuss a state-space decomposition algorithm for the integrated last-mile transportation problem, which consists of the joint scheduling of passengers in a mass transportation system and limited-capacity last-mile delivery vehicles. The algorithm results in orders-of-magnitude performance gains over previously introduced integer programming models. n MB71 West Bldg 106C Recent Advances in Non-convex Optimization Sponsored: Computing Sponsored Session Chair: Andres Gomez, University of Pittsburgh, Pittsburgh, PA, 15217, United States 1 - Cuts for Polynomial Optimization Chen Chen, The Ohio State University, Columbus, OH, United States, Daniel Bienstock, Gonzalo Munoz We introduce intersection cuts for polynomial optimization via maximal S-free sets (termed outer-product-free sets). Such cuts can be used to generate improved convex relaxations. We also discuss implementation of these cuts in MINLP branch-and-cut software. 2 - Optimality-based Domain Reduction for Inequality-constrained NLP and MINLP Problems Carlos Nohra, Carnegie Mellon University, Pittsburgh, PA, United States, Yi Zhang, Nikolaos Sahinidis, Gang Rong We propose novel optimality conditions for general inequality-constrained NLP and MINLP problems, which fully exploit the monotonicity of objective and constraints. Three reduction algorithms are developed separately for unconstrained, one-constraint, and multi-constraint problems. Through forward and backward bound propagation of subgradients with respect to each decision variable, the reduction procedures are developed and integrated with the global solver BARON. Significant memory and CPU time reductions are observed for many problems. 3 - Continuing Work on Polynomial Optimization Daniel Bienstock, Columbia University, Dept of IEOR, 342 Mudd, New York, NY, 10027, United States We present continuing work on polynomial optimization problems, including cuts and approximate discretization. If time permits we will include a discussion on diagnosing infeasibility.

n MB68 West Bldg 105C Joint Session QSR/DM: Journal of Quality Technology Invited Session Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Xiao Liu Co-Chair: Bianca Maria Colosimo, Politecnico di Milano, Politecnico di Milano, Milan, I-20156, Italy 1 - JQT Invited Session Bianca Maria Colosimo, Politecnico di Milano, Via La Masa, 1, Milan, I-20156, Italy In the JQT session three panelists are invited to present their research works, which have been recently published in JQT. A short discussion will follow to highlight the relevant features of manuscripts the journal is looking for. 2 - 50 Years of the Journal of Quality Technology Fugee Tsung, HKUST, Clearwater Bay Road, Hong Kong, Hong Kong Abstract not available. 3 - Multi-Scale Diagnosis of Spatial Point Interaction via Decomposition of the K Function-Based T2 Statistic Qiang Zhou, University of Arizona, Tucson, AZ, United States Abstract not available. 4 - Two-level Augmented Definitive Screening Designs Abby Nachtsheim, Pennsylvania State University, State College, PA, United States Abstract not available. Chair: Yisha Xiang, Lamar University 1 - Funding Opportunities for QSR Yisha Xiang, Lamar University, 2626 Cherry Engineering Building, Beaumont, TX, United States This panel will discuss funding opportunities that are relevant to the Quality, Statistics and Reliability community, with emphasis on the Operations Engineering, Cyber Manufacturing, and Humans, Disasters and the Built Environment programs in the Division of Civil, Mechanical & Manufacturing Innovation at NSF. An open discussion and Q&A will follow the presentations. Panelists Georgia-Ann Klutke, National Science Foundation, 4201 Wilson Boulevard, Arlington, VA, 22230, United States Robin Dillon-Merrill, National Science Foundation, Alexandria, VA, United States Irina Dolinskaya, National Science Foundation, 2415 Eisenhower Ave, Alexandria, VA, 22314, United States n MB70 West Bldg 106B Topics in Scheduling and Parallel Computing Sponsored: Computing Sponsored Session Chair: Oleg Shylo, University of Tennessee, Knoxville, TN, 37996, United States 1 - Supporting an Additional Dimension of Parallelism in PEBBL 2.0 Jonathan Eckstein, Rutgers University, New Brunswick, NJ, United States We consider some large-scale optimization applications which require branch- and-bound algorithms, but the processing of even a single subproblem requires significant parallelism. This talk describes enhancements to the PEBBL parallel branch-and-bound class library designed to support such applications. n MB69 West Bldg 106A Funding Opportunities for QSR Sponsored: Quality, Statistics and Reliability Sponsored Session

173