Diverse ensembles for active learning

摘要: Query by Committee is an effective approach to selective sampling in which disagreement amongst ensemble of hypotheses used select data for labeling. Bagging and Boosting are two practical implementations this that use Boosting, respectively, build the committees. For active learning, it critical committee be made up consistent very different from each other. DECORATE a recently developed method directly constructs such diverse committees using artificial training data. This paper introduces ACTIVE-DECORATE, uses good examples. Extensive experimental results demonstrate that, general, ACTIVE-DECORATE outperforms both Boosting.