We present a novel framework for multi-label learning that explicitly addresses the challenge
arising from the large number of classes and a small size of training data. The key …