Geostatistics in the fast lane using R-TensorFlow and GPUs

R-Ladies Canberra will run a joint event with SSA Canberra in May, and we are delighted to have Dr Andrew Zammit Mangion presenting Geostatistics in the fast lane using R-TensorFlow and GPUs. There will be some hands-on exercises so attendees are encouraged to bring their laptops. Please find below the instruction to download TensorFlow.

The talk will take place on the ground level, Hanna Neumann (MSI) building (building 145) on the ANU campus. There will be refreshments in foyer at 5.15 pm, and the talk will start at 6 pm. After the talk, there will be a dinner at China Plate, Kambri precinct ANU. Please RSVP SSA Canberra (ssacanberra@gmail.com) or R-Ladies Canberra (canberra@rladies.org) by 27 May if you would like to attend the dinner.

Abstract:
Geostatistics plays a key role in the analysis and prediction of spatial and spatio-temporal phenomena. Spatial/spatio-temporal model fitting generally involves multiple matrix operations that can be slow to execute on a CPU. RStudio's interface to TensorFlow provides an ideal vehicle for doing spatial model fitting on a GPU in a fraction of the time needed by a CPU.

I will first briefly talk about Google's TensorFlow and the associated R interface. I will then derive the maximum likelihood estimator for the parameters in a simple spatial model and implement maximum-likelihood estimation in R. Finally, I will show how to implement the same fitting procedure using Google's TensorFlow and illustrate the speed improvement when using a GPU. Attendees are invited to bring their own laptop with R and TensorFlow installed (see https://tensorflow.rstudio.com/tensorflow/articles/installation.html).

Biography:
Andrew Zammit Mangion is Senior Lecturer and DECRA Research Fellow in the National Institute for Applied Statistics Research Australia (NIASRA), University of Wollongong, Australia. He is a co-author of two books and he has published over 20 peer-reviewed articles in the field of spatial and spatio-temporal statistics. He is also the author of a number of related R packages available on CRAN, including EFDR, FRK, and IDE. In 2013, Andrew was awarded a US National Academy of Sciences prize for his work on the modelling and prediction of armed conflicts.