Types of Machine Learning

Machine learning can be classified into three categories based on the characteristics of the
data that is provided and the training methodology:

• Supervised learning
• Unsupervised learning
• Reinforcement learning

Supervised learning

With supervised learning, the machine is trained using a set of labeled data, where each
element is composed of given input/outcome pairs. The machine learns the relationship
between the input and the outcome, and the goal is to predict behavior or make a decision
based on previously given data. For example, we can provide the machine with the
following input to get specific specific outcomes:

• A set of integer numbers (or letters), and then train it to recognize a handwritten number or letter.
• A set of musical notes, and then teach it to recognize the name and the associated pitch.
• Pictures of animals with their names, and then train it to identify a given animal.
• A list of movies that a person has watched, and then train it to determine whether that person will like some other movie.
• A number of e-mails received in your inbox, and then train it to distinguish spam messages from legitimate ones.
•  A list of web-browsing habits, and then teach it to provide search suggestions accordingly.

Unsupervised learning

With unsupervised learning, the machine is trained with unlabeled data and the goal is to
group elements based on similar characteristics or features that make them unique. These
groups are often referred to as clusters. Here we are not searching for a specific, right, or
even approximate single answer. Instead, the accurateness of the results is given by the
similarities in the characteristics or behavior between members of the same group when
compared one to another, and the differences with the elements of another group.
To illustrate, we will use a variation of some of the preceding supervised-learning
examples. If you provide the machine with the following:

• A set of handwritten numbers and letters, it can help you divide the set with numbers in one group and letters in another
• A number of pictures with only one person in each, it can help you group them based on ethnicity, hair or eye color, and so on
• A list of items bought from an online store, it can help you determine the shopping habits and group them by geographical location or age

Reinforcement learning

Reinforcement learning is similar to unsupervised learning in that the training dataset is unlabeled, but differs from it in the fact that the learning is based on rewards and punishments–for lack of better introductory terms–that indicate how closely or otherwise a given element matches a certain grouping condition.
To illustrate, let's return for a moment to the game of Checkers, and picture yourself playing against a machine that is using a reinforcement-learning algorithm. As the computer plays more and more games, the games that are won are used to reinforce the validity of the moves that were made. This is done by assigning a score to each move in a winning game. Moves that result in the capture of a
checker of your opponent get a high score (or reward), whereas those that end up with the
opponent capturing yours get a low score (or punishment). As this process is repeated over
and over again the machine can come up with a set of high-score moves that guarantee a
winning strategy.