Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WC06

3 - An Application of Doubly Stochastic Nonhomogeneous Poisson Process for Detecting Abnormalities Joonho Bae, Korea Advanced Institute of Science and Technology, 291, Daehak-ro, Yuseong-gu, Daejeon, Korea, Republic of, Seung- hoon Lee, Woojin Cho, Jinkyoo Park The anomaly detection problem can be considered with a stochastic counting process. The event of interest is firstly defined as an exceedance of a numerical threshold of the data. Log Gaussian Cox Process, a doubly stochastic nonhomogeneous Poisson process is used to fit the occurrence pattern of the events. The probability value for the data realization within a fixed time interval under the trained intensity function, as a normality score, is calculated and checked if it is below a pre-determined criterion, which is regarded as an anomaly. The robustness of the model is also verified for the abrupt change within the data. 4 - Using Analytics to Improve Patron Engagement at the Los Angeles Philharmonic Michele J. Fisher, Northwestern University, Chicago, IL, United States, Shelley de Leon, Erin Po, Scott Kennedy One hundred years ago, the Los Angeles Philharmonic was founded as L.A.’s first permanent symphony orchestra. A century later, the LA Phil is one of the most dynamic music organizations in the world. It combines a commitment to the future with a fresh eye on the past. The Phil has been using analytics of historical giving data to target future fundraising efforts. Our team analyzed demographics, ticketing, and giving history for patrons and built models to predict the likelihood of a donation and the associated amount. This is helping the organization get ready for the next century. n WC04 North Bldg 122A Combinatorial Optimization Sponsored: Optimization/Integer and Discrete Optimization Sponsored Session Chair: Rajan Harish Udwani, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States 1 - Continuous Submodular Maximization and Applications Rad Niazadeh, Postdoctoral Researcher, Stanford University, 353 Serra Mall, Gate BLDG, Office 484, Stanford, CA, 94305, United States, Tim Roughgarden, Joshua Wang Submodular optimization has deep roots in various application domains such as machine learning, finance, biology, auctions, or even statistical physics. In this work, we consider the natural generalization of submodularity to continuous domains, termed as Diminishing Return submodularity (DR-SM). We design algorithms with tight approximation guarantees for maximizing such functions over the unit hypercube. We show there exists an algorithm achieving at least 1/2 of the value of the optimal point (tight under standard complexity theoretic assumptions). For the special case of strong DR-SM, i.e. with additional coordinate-wise concavity, we develop a fast binary search 1/2-approximation. 2 - Approximating Maximin Fair Allocations Peter McGlaughlin, University of Illinois Urbana Champaign, Urbana, IL, United States We study the fair division of indivisible goods between agents with additive valuations. We use the recently introduced maximin share as local measure of fairness of an allocation. A maximin share is the maximum value an agent can guarantee himself if he is allowed to partition the goods into N bundles (one for each agent), but is allocated his least preferred bundle. We give a simple, greedy algorithm which provides a 2/3 maximin share allocation, i.e., an allocation where each agent receives a bundle worth at least 2/3 of their maximin share. 3 - Robust Appointment Scheduling Rajan Udwani, MIT, Cambridge, MA, United States, Andreas S. Schulz Designing simple appointment systems that try to achieve both high utilization of expensive medical equipment and personnel, and short waiting time for patients, has long been an interesting problem in healthcare. In this work we consider a robust version of the appointment scheduling problem, and give a simple heuristic that achieves the first constant factor approximation. We also show optimal solutions in various special cases that supersede previous work. For the case where order of patients is interchangeable, and under-utilization costs are homogeneous, we show a simple ratio based heuristic that achieves a 1.06 approximation, improving the 2+e approximation in Mittal et al. (2014).

n WC05 North Bldg 122B

Recent Software Developments and Benchmarks Sponsored: Optimization/Computational Optimization and Software Sponsored Session Chair: Hans Mittelmann, Arizona State University, Tempe, AZ, 85287- 1804, United States 1 - Latest Benchmark Results Hans Mittelmann, Arizona State University, Tempe, AZ, United States We report on the current status of our benchmarking effort for both discrete and continuous, commercial and noncommercial optimization software. 2 - Recent Enhancements to Matlab Optimization Solvers Mary C. Fenelon, MathWorks, 3 Apple Hill Dr., Natick, MA, 01760-2098, United States MATLAB has solvers for linear, quadratic, nonlinear, and mixed-integer linear optimization problems. They can solve both analytical and black-box models, including those with multiple objectives. Recent enhancements to these solvers and guidance on selecting a solver will be presented. 3 - Gurobi 8.0 - What’s New Zonghao Gu, Gurobi Optimization, 3733-1 Westheimer Rd Box 1001, Houston, TX, 77027, United States We will give an overview on new features and improvements in the current Gurobi release. In particular, we focus on the new Cloud and Compute Server enhancements and present our newest performance improvements. 4 - CPLEX Progress in 2018 Andrea Tramontani, IBM Italy, Bologna, 06600, Italy, Xavier Nodet In this talk, we will present the new features in the upcoming CPLEX version, as well as some of the performance improvements that were obtained. n WC06 North Bldg 122C Global Optimization I Sponsored: Optimization/Global Optimization Sponsored Session Chair: Logan Michael Mathesen, Arizona State University, 699 S. Mill Ave, Tempe, AZ, 85281, United States 1 - High Dimensional Global Optimization via Optimization of Complimentary Communicating Low Dimensional Subspaces Logan Michael Mathesen, Arizona State University, 699 S. Mill Ave, Tempe, AZ, 85281, United States, xinsheng li, Kasim Seluck Candan, 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. 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. 2 - Trans-dimensional Seismic Inversion by the Hamiltonian Approach Sen Mrinal, Professor, University of Texas at Austin, 10100 Burnet Road, PRC, Bld 196, Austin, TX, 78758, United States, Reetam Biswas Seismic reflection data are used for characterization of hydrocarbon reservoirs. We have developed a trans-dimension full waveform inversion tecnique in which the number of subsurface parameters is also treated as a variable to be solved for. The problem is set up in a Bayesian framework which draws samples from the posterior probabilty distribution. We make use of the sensitivity information to take large jumps in our model search using the Haniltonian framework and thus we are able to draw several models from the most significant parts of the posterior distribution quickly. These are then used to estimate uncertainty which can be further used in decision making.

471