SEL-BALD: Deep Bayesian Active Learning with Selective Labels

While modern deep learning algorithms' strong performance across applications makes them particularly promising for supporting experts, they are often data-hungry and require a large amount of labeled data to perform well.