- This event has passed.
Annual Joint SSA Canberra and R-Ladies Event
Andrew Zammit Mangion has been invited to present at the SSA Canberra 2019 meeting schedule. Below is the title and abstract.
Geostatistics in the fast lane using R-TensorFlow and GPUs
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).