Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TA07

4 - Convergence Rate of Distributed Random Projections Thinh Doan Doan, Georgia Institute of Technology, 755 Ferst Drive, Atalanta, GA, 30332, United States

4 - Computational Evaluation of New Models for Tree Ensembles Optimization

Jean-Philippe P. Richard, University of Minnesota, Industrial and Systems Engineering, 111 Church Street S.E., Minneapolis, MN, 55455, United States, Jongeun Kim, Bijan Taslimi, Mohit Tawarmalani Tree ensemble models are widely used to predict the value of a dependent variable a s a function of independent variables. When the independent variables are controllable, the problem of optimizing the value of the dependent variable by adjusting the values of the independent variables naturally arises. Models have recently been proposed for such three ensembles optimization problems.In this talk, we introduce new models, formulations and heuristics for this problem and compare provide a numerical evaluation of their performance. n TA07 North Bldg 123 Supply Chain and Transportation Networks Sponsored: Optimization/Network Optimization Sponsored Session Chair: Pedro Cesar Lopes Gerum, Rutgers University, 100 Canterbury Court, Piscataway, NJ, 08854, United States 1 - Optimizing Medical Supply Distribution in Rural Regions Considering Uncertainty Larissa P. G. Petroianu, PhD Student, University of Washington, 1703 Harvard Avenue, apartment 6, Seattle, WA, 98122, United States, Zelda B. Zabinsky, Mauricio G. C. Resende Delivery of medical supplies to rural regions is a challenge. Uncertainty contributes to making this a complex task. In the rainy season, roads are often flooded and other means of transportation are needed. We seek to improve the efficiency of distribution, considering the cost of transportation, uncertainties of demand and weather, transportation modes, and transit times. We construct four models: deterministic, robust, chance constraint, and two-stage stochastic program. We present numerical results and discuss differences between the models and how each one addresses uncertainty. 2 - A Comparative Study of Supply Network Robustness: Topological Sensitivity Analysis for Random and Targeted Disruptions Hyunwoo Park, The Ohio State University, 2100 Neil Ave, 632 Fisher Hall, Columbus, OH, 43210, United States, Yusoon Kim Based on 38 empirically derived models of supply networks representing multiple real-world industries, we conducted a sensitivity analysis of supply network robustness. We propose a new measure of supply network robustness to disruption when there are multiple sources and multiple sinks in a network. In the analysis, we simulate both random and targeted disruptions and explain varying vulnerabilities across different supply networks based on the topology (i.e., underlying network structure) and stage-level operational attributes such as cost and time. 3 - The European Hinterland Transport Network as a Complex Network - How is Robustness Affected by the Multi-mode Structure? Camill Harter, PhD Candidate, Erasmus University Rotterdam, Burgemeester Oudlaan 50, T09-08, Rotterdam, 3062PA, Netherlands, Rob A. Zuidwijk, Otto Koppius Robustness of transport networks has widely been studied for unimodal networks. Hinterland container transport, however, comprises multiple interlinked networks formed by different transport modes, which impacts robustness. On the one hand, alternative transport modes provide backup capacities, on the other hand, disruption can cascade easier across interlinked networks. We show that robustness against random and targeted failure differs substantially between multimodal networks and their unimodal counterparts and assess how this generalizes to other multimodal networks. We use a unique dataset containing all intermodal services scheduled in the European hinterland from 2016-2018. 4 - Describing the Distribution of Traffic Density on Corridors Affected by Non-recurrent Congestion Using Minimal Statistical Information Pedro Cesar Lopes Gerum, Rutgers University, 173 Walnut Court, Highland Park, NJ, 08904-1931, United States, Melike Baykal-Gursoy, Marcelo Ricardo Figueroa We investigate reliability measures to assess the performance of roadway traffic density subject to non-recurrent incidents through applied queuing theory. For this, we consider stationary analytical solutions for the number of customers with Markov-modulated service rates. We validate these parametric distributions for peak and non-peak hours, and then generate explicit dependencies on traffic density and incident parameters, thus avoiding the use of costly simulation. Our model will contribute to the design of management tools for roadway traffic, and to incident mitigation, leading to safer and more efficient movement of people and goods.

Stochastic gradient descent is a common algorithm used in machine learning, especially when the loss function is in a separable form. Here, we consider SGD for constrained convex optimization problems, where the constraint is expressed as an intersection of a large number of convex sets. We study a distributed version of the random projections since a centralized approach is too slow for very large- scale problems. Under a mild regularity condition on the convex sets, we show that the rate of convergence of distributed SGD with distributed random projections is the same as that of distributed SGD applied to a problem with no constraints, except for a factor which captures the regularity assumption. 5 - Motivation Dynamics: A Nonlinear Saddle Point Algorithm for Dynamically Satisfying Statically Unsatisfiable Constraints Paul Reverdy, Assistant Professor, University of Arizona, 1130 N. Mountain Ave., P.O. Box 210119, Tucson, AZ, 85721, United States We propose a dynamical system that can be interpreted as a saddle-point algorithm with a nonlinear Lagrange multiplier dynamics. The system is designed to solve constraint satisfaction problems in Euclidean space. When the constraints are reconcilable, i.e., when there exists a feasible solution, the system converges to the feasible solution. When no such solution exists, we derive conditions under which the system exhibits a stable limit cycle where it periodically satisfies the various constraints. n TA06 North Bldg 122C Advances in MINLP Sponsored: Optimization/Global Optimization Sponsored Session Chair: Jean-Philippe P. Richard, University of Florida, Gainesville, FL, 32611-6595, United States 1 - Stronger Polyhedral Relaxations for Polynomial Optimization Problems Aida Khajavirad, Carnegie Mellon University, 1405 Second Ave, Apartment 3S, New York, NY, 10021, United States, Alberto Del pia We consider the Multilinear set defined by a collection of multilinear terms over the unit hypercube. Such sets appear in factorable reformulations of many types of MINLPs. Utilizing an equivalent hypergraph representation for the Multilinear set, we derive various types of facet defining inequalities for its polyhedral convex hull and present a number of tightness results based on the acyclicity degree of the underlying hypergraph. Subsequently, we detail on the complexity of corresponding separation problems and embed the proposed cut generation algorithm at every node of the branch-and-reduce global solver BARON. Extensive computational results will be presented. 2 - Improved Representations of the Quadratic Linear Ordering Problems Boshi Yang, Clemson University, Clemson, SC, United States, Audrey Nicole DeVries, Warren P. Adams The quadratic linear ordering problem (QLOP) seeks a least-cost permutation of n objects where a cost is incurred based on the relative ordering and on products of pairwise orderings. We obtain three results for the QLOP. First, we provide a lifting theorem about the facets of QLOP. Second, we obtain the convex hull representation for n=3 objects with only half the number of restrictions as recent work. Third, we provide the convex hull representation for n=4 objects, expressed in terms of five families of facets. Computational results are presented to show the advantage of our reduced number of constraints in a solution strategy. 3 - Computational Experimentation with Branching Strategies for Global Optimization of Nonlinear Programs and Mixed Integer Nonlinear Programs Current global optimization solvers rely predominantly on branch-and-bound algorithms for the solution of nonconvex nonlinear programs (NLPs) and mixed- integer nonlinear programs (MINLPs). The efficiency of these algorithms depends to a large extent on the techniques used for partitioning the search space. In this work, we investigate novel branching strategies for nonconvex NLPs and MINLPs. We integrate these strategies into the global optimization solver BARON and analyze their impact by conducting an extensive computational study on a large collection of problems selected from publicly available test sets. Carlos Jose Nohra Khouri, Carnegie Mellon University, Department of Chemical Engineering, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States, Nikolaos Sahinidis

245