Dynamic label propagation for semi-supervised multi-class multi-label classification

• We proposes a semi-supervised method on multi-class/multi-label classification.
• We extend our method to handle multiple modality.
• Our method is particularly effective with few number of labeled examples.
• Our method can output a useful graph as data representation for other applications.

Existing semi-supervised methods often have difficulty in dealing with multi-class/multi-label problems due to insufficient consideration of label correlation, and lack an unified framework for multi-modality data. Also, the classification rate is highly dependent on the size of the available labeled data, as well as the accuracy of the similarity measures. To overcome these disadvantages, we propose a semi-supervised multi-class/multi-label classification scheme, dynamic label propagation (DLP), which performs transductive learning through propagation in a dynamic process. Our algorithm emphasizes dynamic metric fusion with label information. A multi-modality extension of the proposed method has been demonstrated to be capable to deal with multiple data types. Significant improvement over the state-of-the-art methods is observed on benchmark datasets for both multi-class and multi-label tasks. The proposed method is proved to be particularly advantageous with very few labeled data.