Abstract

This work shows a rigorous technique for efficiently selecting small subsets of the labelers so that a majority vote from such subsets has a provably low error rate.