Graph representations of deep learning models

Deep learning is often used to learn models consisting of parameterized probability distributions. They are described using directed and undirected graphical models. Undirected graphical models can also be expressed as energy-based models, a particular case of which is known as the Restricted Boltzmann machine.