Accelerating Gradient Boosting Machines

Gradient Boosting Machine (GBM) introduced by \cite{friedman2001greedy} is a widely popular ensembling technique and is routinely used in competitions such as Kaggle and the KDDCup \citep{chen2016xgboost}. In this work, we propose an Accelerated Gradient Boosting Machine (AGBM) by incorporating Nesterov’s acceleration techniques into the design of GBM. The difficulty in accelerating GBM lies in the fact that weak (inexact) learners are commonly used, and therefore, with naive application, the errors can accumulate in the momentum term. To overcome it, we design a “corrected pseudo residual” that serves as a new target for fitting a weak learner, in order to perform the z-update. Thus, we are able to derive novel computational guarantees for AGBM. This is the first GBM type of algorithm with a theoretically-justified accelerated convergence rate.