An adaptive streaming active learning strategy based on instance weighting
Abstract
This paper addresses stream-based active learning for classification. We propose a new query strategy basedon instance weighting that improves the performance of the active learner compared to the commonly useduncertainty strategies. The proposed strategy computes the smallest weight that should be associated withnew instance, so that the classifier changes its prediction regarding this instance. If a small weight is suffi-cient to change the predicted label, then the classifier was uncertain about its prediction, and the true labelis queried from a labeller. In order to determine whether the sufficient weight is “small enough”, we proposean adaptive uncertainty threshold which is suitable for the streaming setting. The proposed adaptivethreshold allows the stream-based active learner to achieve an accuracy which is similar to that of a fullysupervised learner, while querying much less labels. Experiments on several public and real world data provethe effectiveness of the proposed method.