Semi-Supervised Classification Based on Robust Path Regularization

In many classification problems, when labeled data are limited, the classification results may be very poor if only using labeled data. A semi-supervised classification method based on robust path regularization (SSCRPR) is proposed in this paper which can utilize both labeled and unlabeled data to construct a classifier in the semi-supervised setting. The method uses robust path based similarity to capture the manifold structure of both labeled and unlabeled data and then uses the obtained similarity to construct a regularization term which measures the distribution of the manifold to control the characteristic of the classifier. Experimental results on several data sets demonstrate the effectiveness of our method.