New algorithm for parallel inference without block partitioning.
problem Slow inference in probabilistic models due to single-variable updates.
method Lower bounds ELBO with forest mixture bound for parallel inference.
result Algorithm converges faster for 'forest-like' models.