Speaker:   

Shinichiro Shirota, Department of Biostatistics, UCLA

Abstract:

Spatial process models for recent geostatistical data entail computations that become intractable as the number of spatial locations become large. Recently, highly computationally scalable and accurate approximation models have been proposed, which is called nearest neighbor Gaussian process (NNGP) models. These models are corresponding to introducing sparsity into precision of spatial Gaussian processes. In this talk, we would survey basic concepts of NNGP, its computational benefits and potential bottlenecks. Then, we propose extended models to solve these bottlenecks and discuss some connections with preceding spatial process models.

This work is a collaborative work with Prof. Andrew Finely in Department of Forestry and Geography, Michigan State University and Prof. Sudipto Banerjee in Department of Biostatistics, UCLA.

Web page:  https://sites.google.com/site/shinichiroshirota/