Introducing machine learning

Machine learning is a computer science research area that deals with methods to identify and implement systems and algorithms by which a computer can learn, based on the examples given in the input. The challenge of machine learning is to allow a computer to learn how to automatically recognize complex patterns and make decisions that are as smart as possible. The entire learning process requires a dataset as follows:

• Training set: This is the knowledge base used to train the machine learning algorithm. During this phase, the parameters of the machine learning model (hyperparameters) can be tuned according to the performance obtained.
• Testing set: This is used only for evaluating the performance of the model on unseen data.

Learning theory uses mathematical tools that are derived from probability theory of and information theory. This allows you to assess the optimality of some methods over others.

There are basically three learning paradigms that will be briefly discussed:

• Supervised learning
• Unsupervised learning
• Learning with reinforcement

Let's take a look at them.

Supervised learning

Supervised learning is the automatic learning task simpler and better known. It is based on a number of preclassified examples, in which, namely, is known a prior the category to which each of the inputs used as examples should belong. In this case, the crucial issue is the problem of generalization. After the analysis of a sample (often small) of examples, the system should produce a model that should work well for all possible inputs.

The set consists of labeled data, that is, objects and their associated classes. This set of labeled examples, therefore, constitutes the training set.

Most of the supervised learning algorithms share one characteristic: the training is performed by the minimization of a particular loss or cost function, representing the output error provided by the system with respect to the desired possible output, because the training set provides us with what must be the desired output.

The system then changes its internal editable parameters, the weights, to minimize this error function. The goodness of the model is evaluated, providing a second set of labeled examples (the test set), evaluating the percentage of correctly classified examples and the percentage of misclassified examples.

The supervised learning context includes the classifiers, but also the learning of functions that predict numeric values. This task is the regression. In a regression problem, the training set is a pair formed by an object and the associated numeric value. There are several supervised learning algorithms that have been developed for classification and regression. These can be grouped into the formula used to represent the classifier or the learned predictor, among all, decision trees, decision rules, neural networks and Bayesian networks.

Unsupervised learning

In unsupervised learning, a set of inputs is supplied to the system during the training phase which, however, contrary to the case supervised learning, is not labeled with the related belonging class. This type of learning is important because in the human brain it is probably far more common than supervised learning.

The only objects in the domain of learning models, in this case, are the observed data inputs, which often is assumed to be independent samples of an unknown underlying probability distribution.

Unsupervised learning algorithms are used particularly used in clustering problems, in which given a collection of objects, we want to be able to understand and show their relationships. A standard approach is to define a similarity measure between two objects, and then look for any cluster of objects that are more similar to each other, compared to the objects in the other clusters.

Reinforcement learning

Reinforcement learning is an artificial intelligence approach that emphasizes the learning of the system through its interactions with the environment. With reinforcement learning, the system adapts its parameters based on feedback received from the environment, which then provides feedback on the decisions made. For example, a system that models a chess player who uses the result of the preceding steps to improve their performance is a system that learns with reinforcement. Current research on learning with reinforcement is highly interdisciplinary, and includes researchers specializing in genetic algorithms, neural networks, psychology, and control engineering.

The following figure summarizes the three types of learning, with the related problems to address:

Figure 1: Types of learning and related problems