weak supervision (WS) is a rich set of techniques that produce pseudolabels
by aggregating easily obtained but potentially noisy label estimates from a
variety of sources. WS is theoretically well understood for binary
classification, where simple approaches enable consistent estimatio