Fast Low-Rank Bayesian Matrix Completion with Hierarchical Gaussian Prior Models