Machine Learning
Supervised Learning
Learning a mapping from labelled examples. The workhorse of applied ML.
Grasp
You have inputs and you have the answers. Supervised learning fits a function between them and hopes it generalises to inputs it has never seen.
Two shapes cover nearly everything: classification predicts a category, regression predicts a number. The distinction matters less than it seems, because the same models usually handle both with a different output layer and loss function.
What actually determines success is rarely the model. It is the quality of the labels, whether your training distribution resembles production, and whether your evaluation split is honest. A logistic regression on clean, well-split data beats a transformer on leaky data every time, and the second failure is invisible until it reaches users.