Mean and standard deviation of estimate error, as well as the mean of the absolute estimate error, for each combination of method, true period length, signal amplitude, and waveform shape shown in S3 Fig.
The negative effect size indicates that filter bias leads to high inequality of relevance prediction. Both random selection and active learning significant decrease on the inequality.
Comparisons of the annotation accuracy rates for each category of the PASCAL VOC 2012 dataset based on different algorithms after 100 training iterations.