Improving Variational Autoencoder Estimation from Incomplete Data with Mixture Variational Families

Vaidotas Simkus, Michael U. Gutmann
3/5/2024
cs.LGstat.ML

No official code found. Best implementation: vsimkus/demiss-vae (1 star, last commit 2024-06-26).

Abstract

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.

View on arXivView PDF

Code Implementations(1)

vsimkus/demiss-vaeCommunity100%

[TMLR] Research code for the paper "Improving Variational Autoencoder Estimation from Incomplete Data with Mixture Variational Families".

11Shell, PythonJun 25, 20242 years ago

Similar Papers

Posterior Consistency for Missing Data in Variational Autoencoders

90% match2023

Towards Model-Agnostic Posterior Approximation for Fast and Accurate Variational Autoencoders

82% match2024

Matching aggregate posteriors in the variational autoencoder

81% match2023

Distributional Learning of Variational AutoEncoder: Application to Synthetic Data Generation

81% match2 repos72023

Statistical Guarantees for Variational Autoencoders using PAC-Bayesian Theory

81% match2026

Uniform Transformation: Refining Latent Representation in Variational Autoencoders

80% match1 repo32024

Cite this paper

@article{simkus2024improving,
  title  = {Improving Variational Autoencoder Estimation from Incomplete Data with Mixture Variational Families},
  author = {Vaidotas Simkus and Michael U. Gutmann},
  year   = {2024},
  eprint = {2403.03069},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2403.03069v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2403.03069.svg)](https://codeofpaper.com/paper/2403.03069)

Discussion