Dhanya Sridhar

Assistant Professor


Curriculum vitae


dhanya.sridhar <at> mila.quebec


DIRO

University of Montreal, Mila

F.04, 6666 Rue St. Urbain



Identifiable Deep Generative Models via Sparse Decoding


Journal article


Gemma E. Moran, Dhanya Sridhar, Yixin Wang, D. Blei
Transactions of Machine Learning Research, To Appear, 2022

Semantic Scholar ArXiv
Cite

Cite

APA   Click to copy
Moran, G. E., Sridhar, D., Wang, Y., & Blei, D. (2022). Identifiable Deep Generative Models via Sparse Decoding. Transactions of Machine Learning Research, To Appear.


Chicago/Turabian   Click to copy
Moran, Gemma E., Dhanya Sridhar, Yixin Wang, and D. Blei. “Identifiable Deep Generative Models via Sparse Decoding.” Transactions of Machine Learning Research To Appear (2022).


MLA   Click to copy
Moran, Gemma E., et al. “Identifiable Deep Generative Models via Sparse Decoding.” Transactions of Machine Learning Research, vol. To Appear, 2022.


BibTeX   Click to copy

@article{gemma2022a,
  title = {Identifiable Deep Generative Models via Sparse Decoding},
  year = {2022},
  journal = {Transactions of Machine Learning Research},
  volume = {To Appear},
  author = {Moran, Gemma E. and Sridhar, Dhanya and Wang, Yixin and Blei, D.}
}

Abstract

We develop the sparse VAE for unsupervised representation learning on high-dimensional data. The sparse VAE learns a set of latent factors (representations) which summarize the associations in the observed data features. The underlying model is sparse in that each observed feature (i.e. each dimension of the data) depends on a small subset of the latent factors. As examples, in ratings data each movie is only described by a few genres; in text data each word is only applicable to a few topics; in genomics, each gene is active in only a few biological processes. We prove such sparse deep generative models are identifiable: with infinite data, the true model parameters can be learned. (In contrast, most deep generative models are not identifiable.) We empirically study the sparse VAE with both simulated and real data. We find that it recovers meaningful latent factors and has smaller heldout reconstruction error than related methods.


Share



Follow this website


You need to create an Owlstown account to follow this website.


Sign up

Already an Owlstown member?

Log in