A1089
Title: Theoretical and computational challenges for space-time models
Authors: Hao Zhang - Michigan State University (United States) [presenting]
Abstract: Spatiotemporal data are usually huge in size, and the corresponding covariance matrix is of high dimension. Theoretically, it is shown that this covariance matrix is ill-conditioned if the number of spatial locations is huge and constrained in a bounded domain. In that case, it is forced to seek an approximation of the likelihood and the best linear unbiased prediction. Some approximation methods and a new attempt to apply deep learning to the prediction of spatiotemporal data are discussed.
