Weak Supervision
A machine learning approach that rapidly generates vast amounts of noisy, lower-quality training labels using heuristics rather than human annotators.
Think of It Like This
Like a teacher quickly grading 1,000 math tests by just glancing to see if the final number is even, rather than checking the complex work.
Frameworks like Snorkel allow data scientists to write simple Python rules, keyword searches, or leverage cheaper models to programmatically label datasets. While individual rules are noisy and inaccurate, combining many weak rules mathematically yields a highly accurate probability label, bypassing the massive cost of manual data annotation.