Active learning for regression via density power divergence

The accuracy of active learning is crucially influenced by the existence of noisy labels given by a real-world noisy oracle. In this paper, we propose a novel pool-based active learning framework through density power divergence. It is known that density power divergence, such as β-divergence and γ-divergence, can be accurately estimated even under the existence of outliers (noisy labels) within data. In addition, we propose an evaluation scheme for these measures based on those asymptotic statistical analyses, which enables us to perform active learning by evaluating an estimation variance. Experiments on artificial and real-world datasets show that our active learning scheme performs better than state-of-the-art methods.