Rates of Convergence for Sparse Variational Gaussian Process Regression

Date June 9, 2019
Authors David Burt (University of Cambridge), Carl Edward Rasmussen (University of Cambridge), Mark van der Wilk

We prove that our Gaussian process approximations can work well with much less computational requirements than what was known before. Essentially, if the opposite was proved to be true, this would make large scale GP models much more computationally expensive, bringing into question whether GPs would ever be useful for large datasets.

View the paper