Machine learning means the gathering of a large number of data to identify underlying patterns and then use them to make predictions about new examples.
For example, if a large selection of individuals provide us with the name of their favorite movie, then the computer will be able to find the commonalities between all of those movies. The computer will then make proposals while explaining the path it followed to reach these conclusions.
For example: Those who appreciate romantic movies in general don’t like horror movies. On the other hand, they will love films in which the same actors play.
Simply put, machine learning is a form of artificial intelligence based on mathematical and statistical concepts that allows the computer to learn from data without using explicit programming.
Machine learning is also called artificial learning. Machine learning techniques are essential for the prognosis of the machine to be the most accurate possible. Indeed, there are many approaches that change depending on the category and amount of data.
How does machine learning work?
In general, machine learning consists of 2 phases:
The first phase is the design of a system, also called the learning or training phase, and the estimation of a model from the analysis of the data. This includes estimating a probability density or solving a practical task such as translating speech.
In the second phase, the systems may continue to learn even when they are already in production. After determining the model, the second part of the data useful for carrying out the desired task is tested.
Visual recognition is the most common example of machine learning application. For example, the recognition of an animal in a photograph, the detection of bank fraud or even an autonomous vehicle from Google.
Depending on the information available during the learning phase, the learning is qualified separately. If the data is labeled, it is called supervised learning. If the data is not labeled, then this is unsupervised learning. Machine learning can be used with many types of data: graphs, curves, or feature vectors for example
In supervised learning, the machine learning algorithm finds its operation in the fact that the machine has already defined input and output data from the start of its learning. Thus, the algorithm will be able to study these examples, understand them, and thus develop a prediction model capable then of processing the new data.
To create these prediction models, supervised learning will use classification. The classification implies that we can identify and categorize all the data provided according to their characteristics. Thus, the classification is used in particular in the digital, medical, and banking industries. A most common example of classification is the automatic classification between emails to separate emails from interesting sources from those considered spam.
For example, supervised learning is also very often used in regression techniques. This technique makes it possible to predict continuous variables.
Thanks to the supervised learning of a data model, the machine can predict different data variations such as temperature changes for example
Unlike supervised learning, unsupervised learning has a lot of input data but no output data: the responses are not determined. Among unsupervised learning, we find the technique of clustering.