Improving Variational Autoencoder Estimation from Incomplete Data with Mixture Variational Families
This addresses a domain-specific issue for researchers and practitioners using VAEs with incomplete data, but it is incremental as it builds on existing VAE methods.
The paper tackled the problem of estimating variational autoencoders (VAEs) with incomplete training data, showing that missing data increases posterior complexity and can degrade model fit. They introduced mixture variational families and imputation-based strategies, which improved VAE estimation accuracy, though no concrete numbers were provided.
We consider the task of estimating variational autoencoders (VAEs) when the training data is incomplete. We show that missing data increases the complexity of the model's posterior distribution over the latent variables compared to the fully-observed case. The increased complexity may adversely affect the fit of the model due to a mismatch between the variational and model posterior distributions. We introduce two strategies based on (i) finite variational-mixture and (ii) imputation-based variational-mixture distributions to address the increased posterior complexity. Through a comprehensive evaluation of the proposed approaches, we show that variational mixtures are effective at improving the accuracy of VAE estimation from incomplete data.