In this section, we will discuss a very famous generative model that, in an unsupervised scenario, can be employed in order to perform the dimensionality reduction of input dataset X, drawn from a predefined data-generating process. As this book has no particular prerequisites and the mathematical complexity is quite high, we are going to briefly introduce the concepts, without proofs, nor a deep analysis of the structures of the algorithms. Before discussing deep belief networks (DBN), it's necessary to introduce another model, the Restricted Boltzmann Machine (RBM), which can be considered a building block for a DBN.
Unsupervised deep belief networks
Restricted Boltzmann Machines
This network, also known as Harmonium...