Learners trained serially so that instances on which the preceding base learners are not accurate are given more emphasis in training later base-learners; actively tries to generate complementary learners, instead of leaving this to chance.