Abstract: Gaussian process (GP) models are powerful tools for Bayesian classification, but their limitation is the high computational cost. Existing approximation methods to reduce the cost of GP ...