Correlating Variational Autoencoders Natively For Multi-View Imputation

TL;DR

This paper introduces a multi-view VAE framework that models and leverages correlations between latent spaces to improve the imputation of missing data views and facilitate downstream analysis.

Contribution

It proposes a novel multi-view VAE with a correlated joint prior, enabling end-to-end learning of latent space correlations for better data imputation.

Findings

Uncovers stronger correlations in latent spaces of multi-view VAEs.

Enables effective imputation of missing views using learned correlations.

Maintains valid prior distribution during end-to-end training.

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.