Abstract
Variational Bayes (VB) is an approximate method in the Bayesian estimation for parametric models, and provides analytical forms for the posterior distributions. However, since it cannot express essentially parameter correlations in approximate posterior distributions, this is the drawback of VB approach in terms of interval estimation. In this paper, we relax the assumption on the posterior distributions in the VB context, and develop a new VB scheme that can express the posterior distribution with parameter correlation.
