An important subfield of machine learning is supervised learning. In supervised learning, we try to learn from a set of labeled data—that is, every data sample has a desired target value or true output value. These target values could correspond to the continuous output of a function (such as y in y = sin(x)), or to more abstract and discrete categories (such as cat or dog).

A supervised learning algorithm uses the already labeled training data, analyzes it, and produces a mapping inferred function from features to a label, which can be used for mapping new examples. Ideally, the inferred algorithm will generalize well and give correct target values for new data.

We divide supervised learning tasks into two categories:

• If we are dealing with continuous output (for example, the probability of rain), the process is called regression.
• If we are dealing with discrete output (for example, species of an animal), the process is called classification.

In this chapter, we focus on the classification problem of labeling images of the GTSRB dataset, and we will use an algorithm called SVM to infer a mapping function between images and their labels.

Let's first understand how machine learning gives machines the ability to learn like humans. Here is a hint—we train them.