Fig. 7

Accuracy comparison over AL methods with clean and noisy datasets. a Comparison between the proposed and uncertainty-based methods. The results on a clean dataset are presented on the left, while the results on a noisy dataset are displayed on the right. In each experimental trial, there were 10 sampling iterations, and this process was repeated five times for each method to calculate the accuracy. The figure shows the average and standard deviation of each iteration from the five trials. b Comparison of random sampling, coreset, and proposed methods, using the same experimental setup as in (a). Abbreviations AL active learning, LC least confidence, BALD, Bayesian active learning by disagreement, LL: learning loss; std: standard deviation