Training Complex Models with Multi-Task Weak Supervision

5 Oct 2018 · Alexander Ratner, Braden Hancock, Jared Dunnmon, Frederic Sala, Shreyash Pandey, Christopher Ré ·

Snorkel MeTaL: A framework for training models with multi-task weak supervision

Code

Add Remove Mark official

HazyResearch/metal official

420

Tasks

Add Remove

Matrix Completion

Natural Language Inference

Paraphrase Identification

Semantic Textual Similarity

Sentiment Analysis

Datasets

Quora Question Pairs

Results from the Paper

Edit

Ranked #1 on Semantic Textual Similarity on SentEval

Get a GitHub badge

Task	Dataset	Model	Metric Name	Metric Value	Global Rank	Result	Benchmark
Natural Language Inference	MultiNLI	Snorkel MeTaL (ensemble)	Matched	87.6	# 19		Compare
Natural Language Inference	MultiNLI	Snorkel MeTaL (ensemble)	Mismatched	87.2	# 12		Compare
Paraphrase Identification	Quora Question Pairs	Snorkel MeTaL(ensemble)	Accuracy	89.9	# 7		Compare
Paraphrase Identification	Quora Question Pairs	Snorkel MeTaL(ensemble)	F1	73.1	# 9		Compare
Semantic Textual Similarity	SentEval	Snorkel MeTaL(ensemble)	MRPC	91.5/88.5	# 1		Compare
			SICK-R	-	# 3		Compare
			SICK-E	-	# 3		Compare
			STS	90.1/89.7*	# 1		Compare
Sentiment Analysis	SST-2 Binary classification	Snorkel MeTaL(ensemble)	Accuracy	96.2	# 19		Compare

Methods

Add Remove

No methods listed for this paper. Add relevant methods here