Supervised Learning
Learning from labeled examples — the model sees inputs paired with correct answers and learns to map one to the other.
Supervised learning trains a model on labeled examples — inputs each paired with the correct output — so it learns a mapping from input to output that generalizes to new data. It is the most common ML setting: classification (discrete labels) and regression (continuous values).
Worked example: to build a spam filter you feed thousands of emails each labeled spam or not-spam; the model learns the patterns that separate them and then labels new emails. Gotcha: supervised learning is bottlenecked by labels — they are expensive, slow, and often noisy or biased, and the model inherits any bias in them — which is exactly why self-supervised pretraining (learn structure from unlabeled data first, then fine-tune on a few labels) became the dominant recipe.