We discuss trees as the most popular base learners in gradient boosting, with special emphasis on model structure and interaction depth.
