Partial Multi-label Learning with Instance Correlations

Partial multi-label learning aims to induce a multi-label classifier from partial multi-label data in which each instance is annotated with a number of candidate labels, but only a subset of them are valid. Although much progress has been made in this field, most of the existing methods fail to make the most of the instance correlations in partial multi-label data. In this paper, we propose a novel partial multi-label learning method that exploits instance correlations to eliminate noisy labels and induces a multi-label classifier by learning a linear mapping from the feature space to the label space. Experiments on the real-world partial multi-label dataset verify the effectiveness of the proposed method.