Summary

In this chapter, we discussed the perceptron. Inspired by neurons, the perceptron is linear model for binary classification. The perceptron classifies instances by processing a linear combination of features and weights with an activation function. While a perceptron with a logistic sigmoid activation function is the same model as logistic regression, the perceptron learns its weights using an online, error-driven algorithm. The perceptron can be used effectively in some problems. Like the other linear classifiers that we have discussed, the perceptron separates the instances of positive and negative classes using a hyperplane. Some datasets are not linearly separable; that is, no possible hyperplane can classify all the instances correctly.

In the following chapters, we will discuss two models that can be used with linearly inseparable data: ANN, which creates a universal function approximator from a graph of perceptrons, and the support vector machine, which projects the data onto...