Supervised and unsupervised: learning with and without answers
Supervised learning is training on labeled examples, where every input comes with the correct answer attached, like an email marked 'spam' or 'not spam,' or a house's features paired with its actual sale price. The model's job is to learn the mapping from input to output so it can predict labels for new, unseen inputs. This is the most common form of ML in production today, powering things like credit scoring, medical image diagnosis, and voice-to-text transcription.
Unsupervised learning has no labels at all, the model is just given raw data and asked to find structure in it. A classic example is customer segmentation: a retailer feeds in purchase histories with no predefined categories, and a clustering algorithm groups customers into segments based on similarity, like 'bulk weekend shoppers' versus 'frequent small purchasers.' Nobody told the model what the groups should be, it discovered them from patterns in the data itself.