boosting: subsequent samples have more weight of those observations which had relatively higher errors in previous weak learners