To start, we will introduce the perceptron model. The perceptron is the simplest neural network model. It can learn a linear mapping based on the input and output when trained on a labeled training dataset. A linear mapping is the sum of a product of weights on a set of input variables, otherwise known as features. The final sum is passed through a step function to select one of the binary values in the case of a classification problem. The following diagram represents a perceptron:

The weights are derived from the training data by a process called learning. The process of learning will be explained later in this chapter. The perceptron uses the unit step function for the output prediction. The final activation output can be 0 or 1, corresponding to the binary class in the training data. While the unit step function is the simplest activation function, we will touch upon other types of activation functions that are widely used in modern deep learning architectures in the following section.