Semi-Supervised Learning
Training on a small labeled set plus a much larger unlabeled pool, typically by having the model label the unlabeled data itself and keeping only confident guesses.
A model trains first on whatever labeled examples exist, then predicts labels for a much larger unlabeled pool and keeps only the predictions it's confident about, called pseudo-labels. Those pseudo-labels join the training set for another round, letting a small amount of real supervision extend across data that was never manually labeled.
The risk is confirmation bias: a wrong pseudo-label accepted early can make the next wrong prediction in that same region easier to accept, compounding across rounds if left unchecked.