Shoring up the foundations: fusing model embeddings and weak supervision

Mayee F. Chen, Daniel Y. Fu, Dyah Adila, Michael Zhang, Frederic Sala, Kayvon Fatahalian, Christopher Ré
Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, PMLR 180:357-367, 2022.

Abstract

Foundation models offer an exciting new paradigm for constructing models with out-of-the-box embeddings and a few labeled examples. However, it is not clear how to best apply foundation models without labeled data. A potential approach is to fuse foundation models with weak supervision frameworks, which use weak label sources—pre-trained models, heuristics, crowd-workers—to construct pseudolabels. The challenge is building a combination that best exploits the signal available in both foundation models and weak sources. We propose LIGER, a combination that uses foundation model embeddings to improve two crucial elements of existing weak supervision techniques. First, we produce finer estimates of weak source quality by partitioning the embedding space and learning per-part source accuracies. Second, we improve source coverage by extending source votes in embedding space. Despite the black-box nature of foundation models, we prove results characterizing how our approach improves performance and show that lift scales with the smoothness of label distributions in embedding space. On six benchmark NLP and video tasks, LIGER outperforms vanilla weak supervision by 14.1 points, weakly-supervised kNN and adapters by 11.8 points, and kNN and adapters supervised by traditional hand labels by 7.2 points.

Cite this Paper

BibTeX
@InProceedings{pmlr-v180-chen22e, title = {Shoring up the foundations: fusing model embeddings and weak supervision}, author = {Chen, Mayee F. and Fu, Daniel Y. and Adila, Dyah and Zhang, Michael and Sala, Frederic and Fatahalian, Kayvon and R\'e, Christopher}, booktitle = {Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence}, pages = {357--367}, year = {2022}, editor = {Cussens, James and Zhang, Kun}, volume = {180}, series = {Proceedings of Machine Learning Research}, month = {01--05 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v180/chen22e/chen22e.pdf}, url = {https://proceedings.mlr.press/v180/chen22e.html}, abstract = {Foundation models offer an exciting new paradigm for constructing models with out-of-the-box embeddings and a few labeled examples. However, it is not clear how to best apply foundation models without labeled data. A potential approach is to fuse foundation models with weak supervision frameworks, which use weak label sources—pre-trained models, heuristics, crowd-workers—to construct pseudolabels. The challenge is building a combination that best exploits the signal available in both foundation models and weak sources. We propose LIGER, a combination that uses foundation model embeddings to improve two crucial elements of existing weak supervision techniques. First, we produce finer estimates of weak source quality by partitioning the embedding space and learning per-part source accuracies. Second, we improve source coverage by extending source votes in embedding space. Despite the black-box nature of foundation models, we prove results characterizing how our approach improves performance and show that lift scales with the smoothness of label distributions in embedding space. On six benchmark NLP and video tasks, LIGER outperforms vanilla weak supervision by 14.1 points, weakly-supervised kNN and adapters by 11.8 points, and kNN and adapters supervised by traditional hand labels by 7.2 points.} }
Endnote
%0 Conference Paper %T Shoring up the foundations: fusing model embeddings and weak supervision %A Mayee F. Chen %A Daniel Y. Fu %A Dyah Adila %A Michael Zhang %A Frederic Sala %A Kayvon Fatahalian %A Christopher Ré %B Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2022 %E James Cussens %E Kun Zhang %F pmlr-v180-chen22e %I PMLR %P 357--367 %U https://proceedings.mlr.press/v180/chen22e.html %V 180 %X Foundation models offer an exciting new paradigm for constructing models with out-of-the-box embeddings and a few labeled examples. However, it is not clear how to best apply foundation models without labeled data. A potential approach is to fuse foundation models with weak supervision frameworks, which use weak label sources—pre-trained models, heuristics, crowd-workers—to construct pseudolabels. The challenge is building a combination that best exploits the signal available in both foundation models and weak sources. We propose LIGER, a combination that uses foundation model embeddings to improve two crucial elements of existing weak supervision techniques. First, we produce finer estimates of weak source quality by partitioning the embedding space and learning per-part source accuracies. Second, we improve source coverage by extending source votes in embedding space. Despite the black-box nature of foundation models, we prove results characterizing how our approach improves performance and show that lift scales with the smoothness of label distributions in embedding space. On six benchmark NLP and video tasks, LIGER outperforms vanilla weak supervision by 14.1 points, weakly-supervised kNN and adapters by 11.8 points, and kNN and adapters supervised by traditional hand labels by 7.2 points.
APA
Chen, M.F., Fu, D.Y., Adila, D., Zhang, M., Sala, F., Fatahalian, K. & Ré, C.. (2022). Shoring up the foundations: fusing model embeddings and weak supervision. Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 180:357-367 Available from https://proceedings.mlr.press/v180/chen22e.html.

Related Material