Speaker: Avishek Chakraborty (University of Arkansas)
Title: Efficient Spatial Models for Datasets over Large Lattices
Abstract: In this talk, we shall focus on spatial models where association depends on adjacency instead of distance. Such models have broad applicability including datasets indexed by geographical units or observations connected through a graph or network. Depending on the size of the graph, these datasets can be very large. We shall discuss the standard Bayesian model for this setting and its limitations. Then, we are going to develop alternative methods utilizing spectral properties of the Laplacian of the adjacency matrix. We shall begin with a low-rank expansion for modeling spatial dependence and subsequently refine it to improve the accuracy of parameter estimation. Connection of the proposed models with the standard approach will also be discussed. Simulation studies and real-world data analysis will be presented to illustrate benefits of proposed methods with respect to predictive performance and computational attributes. (joint work with Md Kamrul Hasan Khan and Ghadeer Mahdi)
