In the first three articles of our section dedicated to the world of Artificial Intelligence, we introduced the main concepts that have marked the beginnings of this discipline, characterizing their study and the theoretical disputes. Thanks to these bases, it’s now possible to deepen several aspects that will allow us to better understand the real functioning of Artificial Intelligence and its applications.
In this article, therefore, we will face closely one of the most interesting areas to explore: the machine learning. We’ll try to understand together what it is, how it works and how, above all, it is becoming a central element in everyday life, finding several applications, like in the case of our A.I. developed for the management of the Smart Home: Momo OS.
What is machine learning
What characterizes Artificial Intelligence, both from a technological and a methodological point of view, is the learning model through which intelligence becomes skilled in a specific task or action. On the base of the different learning models used in the development of a machine learning system, it is possible to distinguish several algorithms classifications related to the type of algorithm used but also to the purposes the machines themselves are made. So, it is possible to identify three systems of machine learning: supervised, unsupervised or reinforcing.
The supervised learning provides to the machine specific and codified notions, which allow to build a real database of information and experiences. In this way, when the machine faces a problem, it will be able obtain the information contained in its database, analyze it, and decide which answer to give based on previously codified experiences. The machine, therefore, in this case must “simply” choose which is the best answer as a response to the stimulus identified. This kind of learning is used in several sectors, from the medical to the vocal identification, through inductive hypotheses, that are hypotheses obtained by scanning a series of specific problems in order to obtain a solution suitable for a general problem.
In this learning methodology the information put inside the machine haven’t been previously codified, giving the machine the possibility to identify the information without any prior example of their use and without any knowledge of the expected results based on the choice made. Therefore, it will be the machine to catalogue all the information available, organize them, learn their meaning, their possible use and make the prediction of the expected result. As is evident, the unsupervised learning guarantees greater freedom of choice to the machine, which will have the task of organizing information intelligently and learning which are the best results on the basis of the different operational contexts in which it operates.
Learning for reinforcement
Reinforcement learning is the most complex method, because expects the machine to be equipped with systems and tools that can improve its learning and, above all, understand the characteristics of the environment in which it operates. In this case, a number of support tools are supplied to the machine, such as sensors, cameras, etc., allowing the system to detect what is happening in the surrounding environment to make the best choices for a better adaptation to the environment. This is the most used learning method today. The more important example is the pilotless car that, thanks to a complex system of support sensors, is able to cover city and non-urban roads, recognizing possible obstacles, following the road signs and much more. It is an “adaptive” intelligence that in Morpheos we used in the development of the Artificial Intelligence of our Momo OS platform, able to learn the habits of the users, adapting their behavior on the environmental context and the needs of the users.
See you soon, with a new post dedicated to the fascinating world of artificial intelligence.