Tutorial: Variational Inference and Expectation Maximization

Graphical models are easily cosntructed to describe structure 
of virtually any problem. However, inference in most models is 
intractable. In this lecture, various approximate techniques 
that speed up inference are presented and compared on a single 
graphical model. Also, algorithms such as Bayesian inference, 
EM, Variational inference, Belief propagation, and Gibbs sampling 
are derived starting from a single energy function.