Boosting is a family of ensemble methods based on the aggregation of weak learners trained sequentially. These weak learners are trained on the residuals from the previous steps. Generally these weak (or base) learners are trees and more specifically CART trees.