Machine Learning is a branch of artificial intelligence. As part of his thesis, the young Malagasy Princy Heritiana RALAIVAO decided to look into this subject and to explain the fundamentals of this field of study.
Machine learning, a complex science
Can a machine learn without human intervention? It would seem that the answer is yes. Of course, the process includes some human involvement, as Machine Learning requires the developer to be involved in the design of its algorithms. This system, which is related to artificial intelligence, allows a machine, in the broadest sense of the term, to learn from previous experiences. In scientific jargon, this is called “learning data”. These are assimilated by the machine using algorithmic and statistical means. For example, a robot that has the ability to move will have to learn to walk and coordinate its movements.
The recurring example to explain how Machine Learning works remains the classification of an email as spam. In order to recognize a spam, the computer must have training data, collected through several mails with precise characteristics (number of words, terms used …), which will be indicated as being or not spam. It will therefore be necessary to determine a function or, more precisely, a “classifier” that will link the characteristics of an email as “Spam” or “Non Spam”. This function will serve as a model and will then allow the machine to predict the class of an email that arrives in the inbox.
Several projects in progress
Machine Learning may seem complex. However, this is a system that many of our machines are already using. For example, a smartphone can automatically recognize an object or a face or predict the “next word” when we write a message. In the vast field of artificial intelligence, Princy Heritiana RALAIVAO has focused on the prediction of subscription termination, or Churn, of an operator. It is based on call traffic data so that decision makers can then determine more effective marketing strategies to retain their subscribers.
Recently, he has also been working on a system that can recommend a movie to an average user. Thus, based on the films that the latter has noted before, he will be able to propose a selection of films likely to please him.
A methodical operation
Princy Heritiana RALAIVAO explains that Machine Learning can be summarized in three steps:
“Data collection, data understanding, and implementation of a learning algorithm.”
The first step is therefore to determine the data that the machine will learn. Step two pushes the machine to understand this data. Step three is to implement an algorithm adapted to the problem. (classification, regression…) Once this is complete, steps four and five can appear. The model will be evaluated by testing it with data other than those used during its training and then optimizing the model by modifying the algorithm’s parameters.
