Fully Simplified Multivariate Normal Updates in Non-Conjugate Variational Message Passing
Matt P. Wand; 15(Apr):1351−1369, 2014.
AbstractFully simplified expressions for Multivariate Normal updates in non-conjugate variational message passing approximate inference schemes are obtained. The simplicity of these expressions means that the updates can be achieved very efficiently. Since the Multivariate Normal family is the most common for approximating the joint posterior density function of a continuous parameter vector, these fully simplified updates are of great practical benefit.