RBM is a generative stochastic neural network. By saying generative, it indicates that the network models the probability distribution over its set of inputs. And being stochastic means neurons have random behavior when activated. A general diagram of RBMs is depicted as follows:
In general, an RBM is composed of one input layer that is more commonly called the visible layer (v1,v2, v3, v4 in the diagram), and one hidden layer (h1, h2, h3, h4, for example). An RBM model consists of weights W = { 
} that are associated with the connection between the visible layer and the hidden layer, as well as bias a = 
for the visible layer, and bias b = 
for the hidden layer.
There is obviously no output layer in RBMs, and hence the learning is very different from that in feedforward networks, as outlined here:
- Instead of reducing the loss...