Toward optimal probabilistic active learning using a Bayesian approach

Abstract Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active aims at reducing labeling costs by an efficient and effective allocation costly resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes gain misclassification error, (2) uses Bayesian approach introducing conjugate prior distribution determine class posterior deal with uncertainties. By reformulating existing strategies within our proposed model, can explain which aspects are not covered current state-of-the-art why leads superior performance approach. Extensive experiments on large variety datasets different kernels validate claims.