BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:High-Dimensional Bayesian Geostatistics - Sudipto Banerjee (UCLA)
DTSTART:20170123T160000Z
DTEND:20170123T170000Z
UID:TALK70609@talks.cam.ac.uk
CONTACT:Quentin Berthet
DESCRIPTION:With the growing capabilities of Geographic Information System
 s (GIS) and user-friendly software\, statisticians today routinely encount
 er geographically referenced data containing observations from a large num
 ber of spatial locations and time points. Over the last decade\, hierarchi
 cal spatial-temporal process models have become widely deployed statistica
 l tools for researchers to better understanding the complex nature of spat
 ial and temporal variability. However\, fitting hierarchical spatial-tempo
 ral models often involves expensive matrix computations with complexity in
 creasing in cubic order for the number of spatial locations and temporal p
 oints. This renders such models unfeasible for large data sets. In this ta
 lk\, I will present some approaches for constructing well-defined spatial-
 temporal stochastic processes that accrue substantial computational saving
 s. These processes can be used as "priors" for spatial-temporal random fie
 lds. Specifically\, we will discuss and distinguish between two paradigms:
  low-rank and sparsity and argue in favor of the latter for achieving mass
 ively scalable inference. We construct a well-defined Nearest-Neighbor Gau
 ssian Process (NNGP) that can be exploited as a dimension-reducing prior e
 mbedded within a rich and flexible hierarchical modeling framework to deli
 ver exact Bayesian inference. Both these approaches lead to algorithms wit
 h floating point operations (flops) that are linear in the number of spati
 al locations (per iteration). We compare these methods and demonstrate the
 ir use in a number of applications and\, in particular\, in inferring on t
 he spatial-temporal distribution of air pollution in continental Europe us
 ing spatial-temporal regression models in conjunction with chemistry trans
 port models.\n\nThis is based upon joint work with Abhirup Datta (Johns Ho
 pkins University) and Andrew O. Finley (Michigan State University)
LOCATION:MR12\, Centre for Mathematical Sciences\, Wilberforce Road\, Camb
 ridge.
END:VEVENT
END:VCALENDAR