gaussian processes (GPs) are powerful non-parametric function estimators.
However, their applications are largely limited by the expensive computational
cost of the inference procedures. Existing stochastic or distributed
synchronous variational inferences, although have alleviated thi