A technique used to increase the size and diversity of training data by creating new variations of existing data.