Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

POSTER SESSION

24 - Difference-of-convex (DC) Functions offer a Unified Framework for High-dimensional Sparse Estimation Shanshan Cao, PHD Student, Georgia Institute of Technology, 500 Tech Parkway, Atlanta, GA, 303320435, United States, Xiaoming Huo Under the linear regression framework, we study the variable selection problem when the underlying parameter is sparse. With difference-of-convex (DC) penalties, which includes most existing non-convex penalties, such as SCAD, MCP, etc, we consider the directional-stationary solutions. We show that under mild conditions, a certain subset of d-stationary solutions have the ideal statistical properties: asymptotic estimation consistency, model selection consistency, and efficiency. This work shows that DC is a nice framework to offer a unified approach to the existing works when non-convex penalty is involved. Our theoretical results bridge the communities of optimization and statistics. 25 - Decentralized Mixed Integer Programming Theory and Application Innetworked Microgrids Planning A method for formulating and solving a networked microgrid planning problem is performed. Several forces are motivating the development of decentralized microgrid operations, including growing interest in decentralized control frameworks. The method extends the alternating direction method of multipliers (ADMM) along with several refinements to mitigate traps in local optimality that result from the nonconvexity of networked microgrid planning. The scalability observed so far suggests that this method is a practical option for use with large networks and may provide a significant benefit for computational speed. 26 - Learning the Best Metaheuristic Parameters Set for Car- passenger Matching Problem in on Demand Mobility Services Arslan Ali Syed, Bavarian Motor Works, Schroefelhofstr 12, Muenchen, 81375, Germany, Klaus Bogenberger Metaheuristics provide good suboptimal solution to an optimization problem in a reasonable time, but extensive effort is required to select a good set of parameters to obtain reasonable performance. In this work we propose that a neural network could be trained that takes specific features of the problem instance and outputs the best parameters for a specific metaheuristic algorithm. We extract various features for car-passenger matching problem to train a neural network that depending on problem instance returns the best metaheuristic parameters set. 27 - Dynamic Car-passenger Matching Based on Tabu Search Metaheuristics Marvin Erdmann, PhD, BMW, Munich, Germany On Demand Mobility is a concept that would lead to an enhanced use of shared mobility services. To avoid a decline of convenience for the costumers my work’s focus is the realization of a Metaheuristic which matches requests and vehicles in order to find a near-to-optimal solution for the system. 28 - A Variable Neighbourhood Descent Heuristic for Conformational Search Using a Quantum Annealer Brad Woods, 1QBit, 4436 Nanaimo St., Vancouver, BC, V5N5J3, Canada, Dominic Marchand, Moslem Noori, Gili Rosenberg, Austin Roberts Discovering the low-energy conformations of a molecule is of great interest to computational chemists with potential applications in ab-initio materials design and drug discovery. In this paper, we propose a variable neighbourhood search heuristic for the conformational search problem. Using the structure of the molecule, neighbourhoods are chosen to allow for efficient optimization, and also allowing the application of a quantum annealer for this step of the iteration. The proposed method can adapt to the size and topology of the available quantum annealer chip through careful definition of neighbourhoods, making it scalable with respect to future hardware specifications. 29 - M-estimator for Robust Regression Revisited: Robustness and Tractability Tradeoffs Ruizhi Zhang, Georgia Institute of Technology, Georgia Institute of Technology, Atlanta, GA, United States We investigate two important properties of M-estimator, robustness and tractability, in linear regression when the data are contaminated by outliers. Specifically, robustness means the statistical property that the estimator should always be close to the true parameters regardless of the distribution of the outliers, whereas tractability means the computational property that the estimator can be computed efficiently even though the objective function of the M- estimator can be non-convex. By learning the landscape of the empirical risk, we show under mild conditions, many M-estimators enjoy robustness and tractability properties simultaneously when the percentage of outliers is small. 30 - High Dimensional Global Optimization via Optimization of Communicating Low Dimensional Complimentary Subspaces Logan Michael Mathesen, Arizona State University, 699 S. Mill Ave, Tempe, AZ, 85281, United States, Giulia Pedrielli Global optimization suffers the curse of dimensionality. High dimensional search is dominated by the assumption of low effective dimensionality, where few dimensions impact function value, with sophisticated algorithms searching for and exploiting a projection, or creating random embeddings. Masoud Barati, University of Pittsburgh, Benedum Hall 3700 O’Hara Street, Pittsburgh, PA, 15261, United States

We avoid assuming low effective dimensionality and high dimensional modeling by optimizing sets of complimentary subspaces (that collectively exhaust the full space). Enabling intelligent information sharing amongst subspace optimizations, guiding one another to new optimal global projection locations. 31 - Multi-attribute Diffusion Models for Cross-sectional Networks Brennan Antone, Northwestern University, 2145 Sheridan Road, Evanston, IL, 60208, United States, Alina Lungeanu, Noshir Contractor We propose a statistical model for stochastic diffusion processes involving the simultaneous spread of multiple attributes based on patterns of social connections and interaction frequency. We apply this model to the spread of contraception use in rural Kenya. Our model showed the spread of contraception use was dependent upon the diffusion of different beliefs (attitude towards contraception use, subjective norms, perceived behavioral control) throughout the population, the most influential of which was attitude towards contraception use. We discuss how data-driven optimization, using posterior distributions obtained from our model, can be applied to plan social interventions. 32 - Recent Advances on Solving Generalizations of the Single Item Capacitated Lot-sizing Problem Kartik G. Kulkarni, Virginia Tech, 1215 J, Progress Street NW, Blacksburg, VA, 24060, United States, Manish Bansal In this talk, we present our recent advances on solving a generalization of the classical single-item economic lot-sizing problem where the total production capacity in each period can be the summation of some binary multiples of several capacity modules of different sizes. We also develop a new algorithm for the lot- sizing problem with piecewise concave production costs and concave holding costs. Finally, we present results of our computational experiments performed on the aforementioned problems. 33 - Different Decompositions in Random Coefficient and Mixed Integer Programming Kaichen Hsu, National Tsinghua University, No. 101, Section 2, Kuang-Fu Road, Hsinchu, Taiwan 30013, R.O.C., Hsinchu, Taiwan Decompositions are the approach to solving large-scale optimization programs. This approach decomposes the original optimization problem into the master problem and the subproblem. Then the master problem and the subproblem can be solved more efficiently on an iterative basis. We have an original problem which is to estimate a BLP-type random coefficients discrete choice model, a popular utility function used in economics in recent 20 years, formulated as a constrained optimization with mixed integer variables. This study attempts to solve this model more efficiently by various decompositions. 34 - A Trilevel Optimization Model for Resilient Transportation Network Design Mohammad Rahdar, Assistant Professor, St. Ambrose University, Davenport, IA, United States, Lizhi Wang, Guiping Hu We propose a trilevel optimization model for transportation network design, which improves the resiliency of the network against uncertain disruptions. The middle and bottom levels are the network interdiction problem, in which we identify the worst-case scenario disruptions that could lead to a maximal cost to the transportation system. The top level takes the system perspective, which designs the optimal strategy to expand the existing transportation network so that it confronts the worst-case scenario disruptions in the most resilient manner. We also designed an iterative algorithm to solve the trilevel optimization model. 35 - Optimal Placement of a Small Order in a Diffusive Limit Order Book Hyoeun Lee, 725 S. Wright St, Champaign, IL, 61820, United States, J. Figueroa-Lopez, R. Pasupathy We study the optimal placement problem of a stock trader who wishes to clear his/her inventory by a predetermined time horizon by using a limit order or a market order. For a diffusive market, we characterize the optimal limit order placement policy and analyze its behavior under different market conditions. In particular, we show that, in the presence of a negative drift, there exists a critical time horizon such that, for any time horizon longer than the critical time horizon, there exists an optimal placement, which, contrary to earlier works, is different from one that is placed infinitesimally close to the best ask, such as the best bid and second best bid. We also propose a simple method to approximate the critical time horizon and the optimal order placement. 36 - Integration of Variable Micro-grid Energy to Attain Carbon Neutral Manufacturing: A Machine Learning Approach Fei Sun, Texas State University, Austin, TX, 78750, United States, Tongdan Jin, Clara Novoa This study aims to formulate a stochastic model that minimizes one-time investment and operation costs when variable renewable generators and hybrid energy storage systems are integrated into large manufacturing facilities. Various design constraints are considered, including zero carbon footprint, reliability of power supply, and battery state of charge. The objective function is a mixed integer linear programming (MILP) model. The results can provide a reference for the capacity optimization of wind, solar and energy storage systems. The variation of microgrid costs for time of use rate and battery’s depth of discharge levels are analyzed as well.

179