Prof Geoffrey Hinton led the University of Toronto team that won the Netflix competition for the best collaborative filtering to predict user ratings for movies using RBM (https://en.wikipedia.org/wiki/Netflix_Prize). The detail of their work can be accessed from their paper: http://www.cs.toronto.edu/~hinton/absps/netflixICML.pdf.

The output from the hidden units of one RBM can be fed to visible units of another RBM, the process can be repeated to form a stack of RBMs. This results in Stacked RBMs. Each stacked RBM is trained independently assuming that others do not exist. A large number of stacked RBMs form Deep Belief Networks (DBN). DBNs can be trained using both supervised or unsupervised training. You will learn more about them in the next recipe.