What is the difference between weak learners and strong learners in ensembles?

Weak learners perform slightly better than random guessing, while strong learners achieve high predictive accuracy individually.

Weak learners are simple models (like shallow decision trees) that perform only marginally better than chance. Ensemble methods like boosting combine many weak learners to create a strong learner. The idea is that while each weak learner is limited, their combined weighted predictions can approximate complex decision boundaries. Strong learners, by contrast, may already be complex models like deep trees or neural networks, but ensembles of strong learners focus more on variance reduction.