Training Complex Models with Multi-Task Weak Supervision

Benchmark Model Rank Results
paraphrase-identification-on-quora-question-pairsSnorkel MeTaL(ensemble)#9F1: 73.1Accuracy: 89.9
natural-language-inference-on-multinliSnorkel MeTaL (ensemble)#17Matched: 87.6Mismatched: 87.2
sentiment-analysis-on-sst-2-binary-classificationSnorkel MeTaL(ensemble)#19Accuracy: 96.2