Bayesian active learning with abstention feedbacks

We study pool-based active learning with abstention feedbacks where a labeler can abstain from labeling queried example some unknown rate. This is an important problem many useful applications. take Bayesian approach to the and develop two new greedy algorithms that learn both classification rate at same time. These are achieved by simply incorporating estimated average into criteria. prove have near-optimality guarantees: they respectively achieve (1-1e) constant factor approximation of optimal expected or worst-case value utility function. Our experiments show perform well in various practical scenarios.