Correlating Variational Autoencoders Natively For Multi-View Imputation

Ella S C Orme, Marina Evangelou, Ulrich Paquet
Proceedings of UniReps: the Second Edition of the Workshop on Unifying Representations in Neural Models, PMLR 285:181-193, 2024.

Abstract

Multi-view data from the same source often exhibit correlation. This is mirrored in correlation between the latent spaces of separate variational autoencoders (VAEs) trained on each data-view. A multi-view VAE approach is proposed that incorporates a joint prior with a non-zero correlation structure between the latent spaces of the VAEs. By enforcing such correlation structure, more strongly correlated latent spaces are uncovered. Using conditional distributions to move between these latent spaces, missing views can be imputed and used for downstream analysis. Learning this correlation structure involves maintaining validity of the prior distribution, as well as a successful parameterization that allows end-to-end learning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v285-orme24a, title = {Correlating Variational Autoencoders Natively For Multi-View Imputation}, author = {Orme, Ella S C and Evangelou, Marina and Paquet, Ulrich}, booktitle = {Proceedings of UniReps: the Second Edition of the Workshop on Unifying Representations in Neural Models}, pages = {181--193}, year = {2024}, editor = {Fumero, Marco and Domine, Clementine and Lähner, Zorah and Crisostomi, Donato and Moschella, Luca and Stachenfeld, Kimberly}, volume = {285}, series = {Proceedings of Machine Learning Research}, month = {14 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v285/main/assets/orme24a/orme24a.pdf}, url = {https://proceedings.mlr.press/v285/orme24a.html}, abstract = {Multi-view data from the same source often exhibit correlation. This is mirrored in correlation between the latent spaces of separate variational autoencoders (VAEs) trained on each data-view. A multi-view VAE approach is proposed that incorporates a joint prior with a non-zero correlation structure between the latent spaces of the VAEs. By enforcing such correlation structure, more strongly correlated latent spaces are uncovered. Using conditional distributions to move between these latent spaces, missing views can be imputed and used for downstream analysis. Learning this correlation structure involves maintaining validity of the prior distribution, as well as a successful parameterization that allows end-to-end learning.} }
Endnote
%0 Conference Paper %T Correlating Variational Autoencoders Natively For Multi-View Imputation %A Ella S C Orme %A Marina Evangelou %A Ulrich Paquet %B Proceedings of UniReps: the Second Edition of the Workshop on Unifying Representations in Neural Models %C Proceedings of Machine Learning Research %D 2024 %E Marco Fumero %E Clementine Domine %E Zorah Lähner %E Donato Crisostomi %E Luca Moschella %E Kimberly Stachenfeld %F pmlr-v285-orme24a %I PMLR %P 181--193 %U https://proceedings.mlr.press/v285/orme24a.html %V 285 %X Multi-view data from the same source often exhibit correlation. This is mirrored in correlation between the latent spaces of separate variational autoencoders (VAEs) trained on each data-view. A multi-view VAE approach is proposed that incorporates a joint prior with a non-zero correlation structure between the latent spaces of the VAEs. By enforcing such correlation structure, more strongly correlated latent spaces are uncovered. Using conditional distributions to move between these latent spaces, missing views can be imputed and used for downstream analysis. Learning this correlation structure involves maintaining validity of the prior distribution, as well as a successful parameterization that allows end-to-end learning.
APA
Orme, E.S.C., Evangelou, M. & Paquet, U.. (2024). Correlating Variational Autoencoders Natively For Multi-View Imputation. Proceedings of UniReps: the Second Edition of the Workshop on Unifying Representations in Neural Models, in Proceedings of Machine Learning Research 285:181-193 Available from https://proceedings.mlr.press/v285/orme24a.html.

Related Material