Gaussian Mixture Models, Expectation Maximization & KL Divergence