Discriminative Multi-label Model Reuse for Multi-label Learning.

Traditional Chinese Medicine (TCM) with diagnosis scales is a holistic way for diagnosing Parkinson’s Disease, where symptoms can be represented as multiple labels. To solve this problem, multi-label learning provides a framework for handling such task and has exhibited excellent performance. Besides, it is a challenging issue of how to effectively utilize label correlations in multi-label learning. In this paper, we propose a novel algorithm named Discriminative Multi-label Model Reuse (DMLMR) for multi-label learning, which exploits label correlations with model reuse, instance distribution adaptation and label distribution adaptation. Experiments on real-world dataset of Parkinson’s disease demonstrate the superiority of DMLMR for diagnosing PD. To prove the effectiveness of the proposed DMLMR, extensive experiments on four benchmark multi-label datasets show that DMLMR significantly outperforms other state-of-the-art multi-label learning algorithms.