Semi-Supervised Multi-Label Dimensionality Reduction Learning by Instance and Label Correlations

The label learning mechanism is challenging to integrate into the training model of multi-label feature space dimensionality reduction problem, making current methods primarily supervision modes. Many only focus attention on correlations and ignore instance interrelations between original low dimensional space. Additionally, very few techniques consider how constrain projection matrix identify specific common features in In this paper, we propose a new approach semi-supervised by (SMDR-IC, short). Firstly, reformulate MDDM which incorporates as least-squares problem so that propagation can be effectively embedded model. Secondly, investigate using k-nearest neighbor technique, then present l1-norm l2,1-norm regularization terms Experiments massive public data sets show SMDR-IC has better performance than other related methods.