Imagine teaching a computer to recognise cats in photographs. Instead of programming exact rules such as “it has pointed ears and whiskers”, you show it thousands of pictures of cats. From these examples the computer works out for itself which patterns and traits are typical of a cat. Machine learning is exactly this – the process by which a computer learns from data and experience without being explicitly programmed for every individual task. That is what lets it recommend films, translate texts or forecast the weather.
Machine learning is a subdiscipline of artificial intelligence concerned with developing algorithms and statistical models that enable computer systems to perform tasks without explicit instructions. Instead, the systems learn from data, identify patterns in it, and make predictions or decisions on the basis of those patterns. The learning process rests on optimising performance criteria using training data, which leads to the ability to generalise findings to new, previously unseen data.
There are three main categories of machine learning. Supervised learning uses labelled data, where the correct output is known for every input; an example is classifying e-mails as spam and non-spam. Unsupervised learning works with unlabelled data and searches for hidden structures in it, such as segmenting customers by purchasing behaviour. The third category is reinforcement learning, where a system (an agent) learns by trial and error through interaction with its environment, receiving rewards or penalties, typically applied in robotics or autonomous driving.