Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TE69

4 - Tree-structured Data Clustering Derya Dinler, Middle East Technical University, ODTU Endustri Muhendisligi Bolumu, Cankaya, Ankara, 06800, Mustafa Kemal Tural, Nur Evin Ozdemirel We consider a clustering problem in which data objects are rooted trees with unweighted or weighted edges and propose a k-means based algorithm which repeats assignment and update steps until convergence. The assignment step utilizes Vertex Edge Overlap to assign each data object to the most similar centroid. In the update step, each centroid is updated by considering the data objects assigned to it. For the unweighted edges case, we propose a Nonlinear Integer Programming (NIP) formulation to find the centroid of a given cluster and solve the formulation to optimality with a heuristic. When edges are weighted, we also provide an NIP formulation for which we have a heuristic not guaranteeing optimality. n TE68 West Bldg 105C Joint Session QSR/Practice Curated: Spatio-Temporal Data Analysis and Applications II Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Shyam Ranganathan Co-Chair: Asif Sikandar Iquebal, MS, Texas A&M University, TX, United States 1 - Optimal Sentinel Placement on a Network to Infer Transmission Probability of a Contagion Samarth Swarup, PhD, Virginia Polytechnic Institute and State University, Blacksburg, VA, 24060, United States, Yihui Ren, Stephen Eubank We study the problem of optimal sentinel placement on a network to estimate the probability of transmission of a contagion. We assume that only one partial observation of the outbreak is available, corresponding to locations (nodes) where sentinels are placed. The problem is to choose the sentinel locations, given the graph and source node. Our approach relies on the mathematical theory of network reliability, which can be used to estimate the probability distribution of the transmission rate given a partial observation and the network structure. We cast the problem in an entropy maximization setting and develop a greedy approach for sentinel placement. Results on two real-world networks are shown. 2 - Planar Random Graph Representations of Spatio-temporal Evolution of Surface Morphology Ashif S. Iquebal, Texas A&M University, College Station, TX, 77840, United States, Satish Bukkapatnam, Soundar Kumara We present a random planar graph representation to quantify the spatiotemporal evolution of surface morphology during finishing processes. We show that the proposed representation captures the complex interflow among neighboring asperities during finishing, and establishes an efficient endpoint criterion, i.e., surface quality improves only until each asperity interflows with six neighbors. 3 - A New Nonhomogenous Poisson Process Model for Spatiotemporal Event Data Yifei Yuan, University of Arizona, 651 E. Camino Lujosa, Tucson, AZ, 85704, United States, Yinwei Zhang, Jian Liu Spatially distributed events occurred over time are usually modeled by spatiotemporal point processes with nonstationary occurrence rate. Conventional methods based on prespecified B splines may not be accurate if the knot locations are not properly defined. A free knot B spline approach is proposed to optimally model the spatiotemporarily dependent occurrence rate, with the knot locations and spline coefficients estimated simultaneously. A real-world case study demonstrates the effectiveness of the proposed method. 4 - Numerical Simulation of the Conduction and Propagation of Spatiotemporal Electrodynamics in Complex Systems Bing Yao, The Pennsylvania State University, 801 B6 Southgate Dr, University Park, PA, 16801, United States, Hui Yang Heart disease is a major threat to human health. The key to improving the cardiac care services is to develop a better understanding of cardiac activity. Computer simulations facilitate quantitative elucidation of heart functions. Here, we propose a novel method to simulate cardiac spatiotemporal electrodynamics. We project the 3D heart into a 2D graph by the method of t-distributed stochastic neighbor embedding. Then, a moving-least-square mesh-free method is proposed to simulate the nonlinear electrodynamics on the 2D graph. Simulation results show that this model efficiently simulates the cardiac electrodynamics, and will be an effective tool for optimal medical decision-making.

5 - An Adaptive Approach for Fusion of High-accuracy with Low-accuracy Data Mostafa Reisi Gahrooei, PhD, Georgia Institute of Technology , GA, United States Available data fusion techniques concentrate only on data fusion strategies without providing guidelines on how to sample high-accuracy (HA) data. This work addresses the problem of selecting HA data adaptively and sequentially so when it is integrated with the low-accuracy (LA) data a more accurate surrogate model is achieved. n TE69 West Bldg 106A Spatio-temporal Modeling and Analysis Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Aziz Ahmed Co-Chair: Yu Ding, Texas A&M University, College Station, TX, 77843-3131, United States 1 - Statistical Modeling of Spatio-temporal Data Arising from Stochastic Convection-diffusion Processes Xiao Liu, University of Arkansas This talk will describe a holistic spatio-temporal modeling and analysis framework for large-scale spatio-temporal data sets arising from a broad class of physical convection-diffusion processes. 2 - Spatio-temporal Modeling of Adverse Birth Outcomes Due to Heat Effects Shyam Ranganathan, Virginia Polytechnic Institute and State University, VT, Blacksburg, VA, United States, Christopher Grubb, Samarth Swarup, Julia Gohlke, Yuhao Wu Observational studies to estimate effects that are rare in the population often report small effect sizes. Spatio-temporal models with random effects and spatial clustering can help improve estimates by disentangling confounders. We develop a novel spatio-temporal model and address an important problem in health effects of heat waves. Adverse birth outcomes such as low birth weight and preterm birth due to heat effects have been studied for some time. We use anonymized birth records data from Virginia to present our results for this problem using our model. 3 - Combining Constrained Bayesian Optimization And Spatio- temporal Surveillance For Optimal Water Quality Monitoring Network Design Junzhuo Chen, Georgia Institute of Technology, Atlanta, GA, 30318, United States The goal of designing an optimal water quality monitoring network for river systems is to identify the location of a finite number of sensors that minimizes the expected detection delay of a contaminant event. In practice, data collected by sensors are subject to measurement errors, which makes a statistical procedure necessary to control the false alarm rates when using such streaming data for contaminant detection. We propose a combined framework of Spatio-Temporal surveillance and Bayesian Optimization for robust sensor network design in the presence of sensor measurement error. 4 - Spatio-temporal Modeling & Analysis For Wind Farm Data. Ahmed Aziz, Texas A&M University, 2500 Central Park Ln Unit 2406, College Station, TX, 77840, United States Wind data are known to exhibit dependencies across a broad spectrum of spatial and temporal scales. In this talk, we present several findings related to the spatio- temporal modeling and analysis of local wind dynamics using turbine-specific wind farm data. We discuss various concepts such as space-time interaction, flow- dependent asymmetry and regime-switching dynamics, and present novel ways to incorporate them into spatio-temporal models. Finally, we discuss the implications of such findings on the wind farm operational analytics such as wind speed forecasting and power estimation. 5 - Spatiotemporal Transfer Learning for 3D Dynamic Field Modeling Di Wang, Peking University, Beijing, China, Xi Zhang We propose a 3D field estimation method using limited sensor observations. A dynamic field transfer learning approach, to identify spatiotemporal correlation by integrating a multitask learning framework into an autoregressive model, is developed. Our proposed approach is verified through a real case study of grain thermal field estimation.

403