A neural network is a network of interconnected perceptrons. An example of such a network is shown in the following diagram:

In the preceding diagram, the input layer is fed with input data, which are then passed to the hidden layer. Each unit, or node, in the hidden layer computes an activation based on the input, and then passes it to a final output node. The output node, in turn, computes the final output, based on all of the input from the hidden layer. Although the preceding diagram shows only one hidden layer, there can be multiple hidden layers for a given network. An activation function in the output can be used for categorical data with two classes. There are some additional techniques required to predict multi-label classes. We will learn about one-hot encoding, softmax, and cross-entropy in the following sections.