Scaling Up Crowd-Sourcing to Very Large Datasets: A Case for Active Learning-笔记