Monte Carlo vs. Riemann Reparameterization Trick Mixture of Gaussians - Gradient Ascent vs. EM Unsupervised Learning with Neural Networks