EM in a nutshell

Explaining the EM algorithm in a nutshell
deep learning
probability
technical
Author

Sean Zhang

Published

August 2, 2019

One of the most interesting ideas in machine learning I’ve found is the EM algorithm. The idea behind EM is summarized as follows:


EM Process

Limitations

While the idea of EM is powerful, it is impractical in models where the evaluation of the posterior \(p(Z|\theta, X)\) is impossible (think multi-layer deep neural network with nonlinearity in between).

To make inference about \(Z\) in those cases, we need to resort to another powerful tool (Variational Inference).