CFE 2019: Start Registration
View Submission - CMStatistics
Title: Blockwise Euclidean likelihood for estimation of space-time covariance model using OpenCL in GPGPUs Authors:  Victor Morales-Onate - Facultad Latinoamericana de Ciencias Sociales (Ecuador) [presenting]
Federico Crudu - University of Siena (Italy)
Moreno Bevilacqua - Universidad de Valparaiso (Chile)
Abstract: A spacetime blockwise Euclidean likelihood method is proposed for the estimation of covariance model when dealing with large spacetime Gaussian data. The method uses moment conditions coming from the score of the pairwise composite likelihood. A feature of this approach is that it is possible to obtain computational benefits with respect to pairwise likelihood depending on the choice of the spatiotemporal blocks. We also study the asymptotic properties of the proposed estimator. In order to speed-up computation we consider a general purpose GPU implementation using OpenCL. We illustrate the advantages of our methodology by means of a simulation study highlighting the computational gains of the OpenCL GPU implementation. Finally, we apply our estimation method to the Irish wind speed data.