Combining Active Learning and Semi-Supervised Learning by Using Selective Label Spreading

In the literature, a number of methods have been proposed for semi-supervised learning. Recently, graph-based methods of semi-supervised learning have become popular because of their capability of handling large amounts of unlabeled data. However, the existing graph based semi-supervised learning algorithms do not optimize the process of selecting better labeled data. We have developed a new selective semi-supervised learning algorithm, called selective label spreading (SLS) by integrating the active learning model into the label spreading framework. SLS optimizes the process of selecting better labeled data in order to improve classification performance. We applied SLS to the well-known hand-written digits recognition data set and demonstrated that SLS can improve the classification performance. The selective label spreading scheme requires a much smaller number of queries to achieve high accuracy compared with random query selection.