Semi-supervised and active learning with the probabilistic RBF classifier

The probabilistic RBF network (PRBF) is a special case of the RBF network and constitutes a generalization of the Gaussian mixture model. In this paper we propose a semi-supervised learning method for PRBF, using labeled and unlabeled observations concurrently, that is based on the expectation–maximization (EM) algorithm. Next we utilize this method in order to implement an incremental active learning method. At each iteration of active learning, we apply the semi-supervised method, and then we employ a criterion to select an unlabeled observation and query its label. This criterion identifies points near the decision boundary. In order to assess the effectiveness of our method, we propose an adaptation of the well-known Query by Committee (QBC) algorithm for the active learning of the PBFR, and present experimental comparisons on several data sets that indicate the effectiveness of the proposed method.