B0217
Title: On the use of mini-batching for fitting Gaussian processes
Authors: Matthew Heaton - Brigham Young University (United States) [presenting]
Abstract: Gaussian processes (GPs) are highly flexible, nonparametric statistical models that are commonly used to fit nonlinear relationships or account for correlation between observations. However, the computational load of fitting a Gaussian process makes them infeasible for use on large datasets. To make GPs more feasible for large datasets, the focus is on the use of mini-batching to estimate GP parameters. Specifically, both approximate and exact minibatch Markov chain Monte Carlo algorithms are outlined that substantially reduce the computation of fitting a GP by only considering small subsets of the data at a time. This methodology is demonstrated and compared using various simulations and real datasets.
