What is Machine Learning?

Machine Learning (ML) is the branch of Artificial Intelligence that makes it possible to extract structured information from large quantities of usually unstructured data.
The terms Machine Learning and Artificial Intelligence are often used as synonyms, yet they do not have the same meaning. In fact, although everything that has to do with ML belong to the area of Artificial Intelligence, AI is not limited to Machine Learning but includes several other disciplines such as Mathematical Optimization, Statistical Analysis, Prediction based on time series, modeling, and simulation. Machine Learning is method for analysis that automates the creation of analytical models. The system can therefore learn from data, identify models, and take decisions with limited human intervention. Through experience, the system can improve its own abilities and performances over time. The base of automated learning is a series of algorithms that, starting from initial concepts that are fed to the machine, will take a certain decision based on what has been learned.

The definition of Machine Learning

The term Machine Learning was first used in 1959 by the American scientist Arthur Lee Samuel, meaning “a field of study that gives computers the ability to learn without being explicitly programmed”. This was the primordial definition of a concept that greatly evolved over time. For a more modern definition of the term, a jump in time of more than 40 years is needed. In 1997, with the publication of the book Machine Learning by Tom Michael Mitchell, Machine learning was defined as follows: “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”
To simplify such definition based on our experience, we can affirm that Machine Learning is a branch of Artificial Intelligence that entails the creation of software applications able to learn to perform certain activities without being explicitly programmed to do so. The Algorithms of Machine Learning can extract information from large amounts of data and use it to solve a vast array of problems that would otherwise be difficult, if not impossible, to solve with conventional algorithms. A more detailed description of Machine Learning can be found in the dedicated area on our website.

ML: practical examples

In our daily lives, we often encounter technologies that employ Machine Learning. A very famous example is Google Translate, which uses deep learning techniques to translate multiple sentences at the same time from one language to another.
The spam filter, too, is based on automated learning models. Emails are in fact classified as spam or no spam thanks to an algorithm of classification that learns from provided examples.
Another example is the services from Expedia, a website for researching and booking hotels and flights. Through the command search for best rate, the system provides research results by continuously learning and adapting to variations in itinerary, schedules, and flight prices.
Identifying faces in pictures, now something that every smartphone and camera can do, is an example of unsupervised learning. Facial recognition makes it possible to recognize all faces in a given image and understand the distinguishing features of each face.
Playing chess with Machine Learning? Also possible. Stockfish, a chess engine that earned the world title 11 times, gave a hard time to very famous chess grand master, Magnus Carlesen, during the last world championship. A chess engine is based on reinforcement learning. In fact, the number of possible moves is very high, and it is impossible to calculate all the possible alternatives. The strength of such technology is therefore connected with calculating variables and defining the expected value of actions in uncertain conditions. In other words, it is a system that automatically learns from its own mistakes.

Machine Learning in the business world

In aHead Research, we model and implement supervised and unsupervised ML algorithms to extract useful information for business from data.
In the area od GDO, in fact, we help our partners and clients in the prediction of product sale thanks to unsupervised learning models that, based on their analysis of time series, try to define a relation between future and past sales.
Another example of industrial application of ML related to the manufacturing sector. Through our ML models we are able to perform anomaly detection, through the identification of elements, event and observations that raise concern in that they greatly differ from the average data.
In production plants, such operations allow us to identify machines that are close to suffering breakdowns. In fact, based on data collected about the plant, we develop algorithms that can classify the several kinds of malfunction, reducing costs and avoiding unforeseen interruptions of production.