Gradient Boosting Forest: a Two-Stage Ensemble Method Enabling Federated Learning of GBDTs

Graph neural network interatomic potential ensembles with calibrated aleatoric and epistemic uncertainty on energy and forces

“…However, it is important to note that most of this computation can be performed in parallel and thus only leads to a small overhead of computing the ensemble approximation and recalibration in real time. Some works have proposed methods for speeding up the training of ensembles, such as snapshot ensembles, 35,36 which could also be applied in this case. Another widely accepted advantage of ensembles is that they often improve prediction accuracy (see Fig.…”