Batch-Mode Active Learning via Error Bound Minimization

Active learning has been proven to be quite effective in reducing the human labeling efforts by actively selecting the most informative examples to label. In this paper, we present a batch-mode active learning method based on logistic regression. Our key motivation is an out-of-sample bound on the estimation error of class distribution in logistic regression conditioned on any fixed training sample. It is different from a typical PACstyle passive learning error bound, that relies on the i.i.d. assumption of example-label pairs. In addition, it does not contain the class labels of the training sample. Therefore, it can be immediately used to design an active learning algorithm by minimizing this bound iteratively. We also discuss the connections between the proposed method and some existing active learning approaches. Experiments on benchmark UCI datasets and text datasets demonstrate that the proposed method outperforms the state-of-the-art active learning methods significantly.