There are many approaches to combine multiple models without resorting to
ensemble learning schemes. Best known are simpler ensemble methods such
as Bagging (Breiman, 1996) and AdaBoost (F reund & Schapire, 1996), which
rely on training a set of diverse base classiers (typically via dierent subsam-ples of the training set), whose predictions are then combined by more-or-less
elaborate voting techniques, see e.g.