Bagging (bootstrap aggregation)

Each resample draws rows with replacement, so about a third of rows are left out of any one model. Averaging high-variance, low-bias learners such as deep trees reduces variance; it does little for bias. Random forests add a random subset of features at each split to decorrelate the trees. More models lower variance and then plateau; they do not by themselves cause overfitting.