Contrastive estimation reveals topic posterior information to linear models Journal Article

  


Authors: Tosh, C.; Krishnamurthy, A.; Hsu, D.
Article Title: Contrastive estimation reveals topic posterior information to linear models
Abstract: Contrastive learning is an approach to representation learning that utilizes naturally occurring similar and dissimilar pairs of data points to find useful embeddings of data. In the context of document classification under topic modeling assumptions, we prove that contrastive learning is capable of recovering a representation of documents that reveals their underlying topic posterior information to linear models. We apply this procedure in a semi-supervised setup and demonstrate empirically that linear classifiers trained on these representations perform well in document classification tasks with very few training examples. ©2021 Christopher Tosh, Akshay Krishnamurthy, Daniel Hsu.
Keywords: statistics; naturally occurring; classification (of information); learning systems; topic modeling; embeddings; latent dirichlet allocation; linear modeling; contrastive estimation; representation learning; information retrieval systems; datapoints; document classification; model assumptions
Journal Title: Journal of Machine Learning Research
Volume: 22
ISSN: 1532-4435
Publisher: Microtome Publishing  
Date Published: 2021-01-01
Start Page: 281
Language: English
PROVIDER: scopus
DOI/URL:

https://www.scopus.com/inward/record.uri?eid=2-s2.0-85121143204&partnerID=40&md5=2bf883e1122c7e38b52ab7b2a65dfc0d
Notes: Article -- Export Date: 3 January 2022 -- Source: Scopus