Several labeling functions may vote on an example and a model estimates their reliability or correlations. The combined labels can train a downstream model, but systematic errors in the sources can persist.
Weak supervision creates approximate labels from rules, heuristics, distant signals, or noisy models instead of labeling every example manually.
Several labeling functions may vote on an example and a model estimates their reliability or correlations. The combined labels can train a downstream model, but systematic errors in the sources can persist.