Labeling Information Enhancement For Multi-Label Learning With Low-Rank Subspace

In multi-label learning, each training example is represented by an instance while associated with multiple class labels simultaneously. Most existing approaches make use of multi-label training examples by utilizing the logical labeling information, i.e., one class label is either fully relevant or irrelevant to the instance. In this paper, a novel multi-label learning approach is proposed which aims to enhance the labeling information by extending logical labels into numerical labels. Firstly, a stacked matrix is constructed where the feature and the logical label matrix are placed vertically. Secondly, the labeling information is enhanced by lever-aging the underlying low-rank structure in the stacked matrix. Thirdly, the multi-label predictive model is induced by the learning procedure from training examples with numerical labels. Extensive comparative studies clearly validate the advantage of the proposed method against the state-of-the-art multi-label learning approaches.