Sub-domain of AI where systems learn from data without being explicitly programmed. The more data the model sees, the more accurate it becomes.
🎯 Why it matters
This concept is the fundamental driving force behind almost all current AI technologies. Understanding that a system learns by example rather than by fixed rules allows us to understand why data quality is crucial. It also helps to accept that an AI system can make mistakes, just like a human in the learning phase.
💡 Concrete example
A spam filter learns to recognize unwanted emails by analyzing thousands of examples.
⚙️ In practice
In practice, the process begins with collecting and cleaning a data set, divided into training and testing data. An algorithm is then chosen that is trained to identify recurring patterns. Once the model is trained, its performance on the test data is assessed to ensure that it is able to generalize its predictions to new real situations.
Common mistake: To believe that machine learning produces infallible results. In fact, a model will never be smarter or more impartial than the data used to drive it.
❓ Frequently asked questions
In supervised learning, training data already contain the expected answers. In unsupervised learning, the algorithm itself must find logical structures or groups without prior assistance.
This depends on the complexity of the problem. For simple tasks, a few hundred well-targeted examples may suffice, while complex models of vision or language require millions of data.