Let's start talking about perceptrons. In this section, we will discuss the perceptron algorithm, particularly as an instance of online learning. We will also look at a demonstration of training the perceptron, and show you what online training looks like.
On the surface, the perceptron classifier resembles a support vector machine (SVM). It is a linear classifier and predicts that all observations lying on a particular side of a hyperplane belong to a particular class. However, perceptrons are not SVMs—that is, they do not try to maximize the gap between classes and they are not designed to train data in a batch, where all data is perceived at once and no future data is used. Batch learning uses all available data for training at once. In comparison, the classifier fits data in the best way that it can.
All algorithms that we usually...
