Keywords: variational autoencoder, vae, missing data, incomplete data, mixtures, mixture distributions, posterior
TL;DR: Incomplete data increases posterior complexity compared to the fully-observed case; we propose that variational mixtures are a natural approach to handle this increased complexity.
Abstract: We consider the task of estimating variational autoencoders (VAEs) when the training data is sparse due to missing values. We show that missing data increases the complexity of the posterior distribution over the latent variables compared to the fully-observed case. This increased complexity may adversely affect the fit of the model due to variational posterior mismatch. To address this increased posterior complexity, we introduce two strategies: using (i) finite variational-mixture and (ii) imputation-based variational-mixture distributions. Through a comprehensive evaluation of the proposed approaches, we verify that the use of variational mixtures proves effective in enhancing the accuracy of VAE estimation from incomplete data.
Primary Subject Area: Other
Paper Type: Research paper: up to 8 pages
DMLR For Good Track: Participate in DMLR for Good Track
Participation Mode: In-person
Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.
Submission Number: 26
Loading