mean-field variational inference is a method for approximate bayesian posterior inference. It approximates a full posterior distribution with a factorized set of distributions by maximizing a lower bound on the m