In this paper we propose Softboost, a novel Boosting al-gorithm which combines the merits of transductive and inductive learning approaches to attack the problem of learning from very few labeled training examples. In the ...