A deep belief network (DBN) in Ai & Machine Learning is a generative graphical prototype, or alternatively a class of deep neural network, comprised of several layers of latent variables (“hidden units”), with connections between the levels but not between units within each layer.
DBNs have the following advantages:
Only a tiny labelled dataset is required.
On GPU-powered machines, training time is rather quick.
A DBN can learn to probabilistically reconstruct its inputs when trained on a set of examples without supervision. The layers are then used to detect features. Following this learning stage, a DBN may be trained to do classification under supervision.
DBNs are made up of basic, unsupervised networks such as restricted Boltzmann machines (RBMs) or autoencoders, where eac