reference paper
pribadihcr opened this issue · comments
Pribadi HCR commented
Xiao Lin commented
Vanilla active learning, that is selecting an example with maximum entropy after softmax works the best, resulting in 60%~70% reduction in the amount of training examples. Fancy approaches don't seem to work as well as vanilla active learning. These are empirical results, no formal paper.
Xiao Lin commented
Algorithm published as arxiv.org/abs/1711.01732