Estimating the parameters of a probabilistic directed graphical model from
incomplete data remains a long-standing challenge. This is because, in the
presence of latent variables, both the likelihood function and posterior
distribution are intractable without further assumptions about