Abstract
From the background information supplied by the inventors, news correspondents o btained the followingquote: “Many machine learning (ML) models employ labeled d atasets to “learn” a mapping from a setof input features to a desired output, s uch as a class label or regression value. The “desired output”for a particular set of input features is generally specified by a human annotator, and may be re ferred toas the ground truth. The performance of a ML model may be constrained by the quality of the groundtruth data used during a training process, as an up per limit to ML model accuracy is the accuracy of theground truth labels themse lves. Therefore, it is generally desirable to explore approaches for increasingthe accuracy/consistency of ground truth labels provided by human annotators. Fu rther, as availability oflabeled data is a recognized bottleneck in the field o f ML, it is further desirable to explore approaches forincreasing speed of grou nd truth data generation.
NETL
NSTL