There is nothing that prevents the backend propositional learner from employing either bagging [Breiman, 1996] or boosting [Schapire, 1999]. In addition, if the random search is used, multiple runs of the synthetic feature construction stage can be voted as well, since the results are non-deterministic. In fact, the two approaches can be combined and we found that generally the most effective technique was voting AdaBoost learners.