Graph-based active learning based on label propagation

Abstract

By only selecting the most informative instances for labeling, active learning could reduce the labeling cost when labeled instances are hard to obtain. Facing the same situation, semi-supervised learning utilize unlabeled instances to strengthen classifiers trained on labeled instances under suitable assumptions. However, the current active learning methods often ignore such effect. Combining semi-supervised learning, we propose a graph-based active learning method, which can also handle multi-class problems, in the entropy reduction framework. The proposed method trains the base classifier using a popular graph-based semi-supervised label propagation method and samples the instance with the largest expected entropy reduction for labeling. The experiments show that the proposed method outperforms the traditional sampling methods on selected datasets. © 2008 Springer Berlin Heidelberg.