An active learning algorithm for multi-class classification

Since the number of instances in the training set is very large, data annotating task consumes plenty of time and energy. Active learning algorithms can efficiently reduce the number of instances that need to be annotated. In this paper, authors propose a new active learning algorithm. The algorithm is mainly proposed for multi-class classification model based on support vector machine (SVM). In the algorithm, the unlabeled instances that can promote several SVM classifiers in the multi-class classification model will be selected firstly. So when one newly selected instance is added into training set, more than one classification hyper-planes in the multi-class classification model will be promoted. During the process of instance selection, the algorithm also tries to choose the instance that is least similar with the instances that have already been annotated. In this way, the instances selected by the algorithm for annotating will perfectly represent the feature of the whole dataset.