English

Embedding-reparameterization procedure for manifold-valued latent variables in generative models

Machine Learning 2018-12-10 v1 Machine Learning

Abstract

Conventional prior for Variational Auto-Encoder (VAE) is a Gaussian distribution. Recent works demonstrated that choice of prior distribution affects learning capacity of VAE models. We propose a general technique (embedding-reparameterization procedure, or ER) for introducing arbitrary manifold-valued variables in VAE model. We compare our technique with a conventional VAE on a toy benchmark problem. This is work in progress.

Keywords

Cite

@article{arxiv.1812.02769,
  title  = {Embedding-reparameterization procedure for manifold-valued latent variables in generative models},
  author = {Eugene Golikov and Maksim Kretov},
  journal= {arXiv preprint arXiv:1812.02769},
  year   = {2018}
}

Comments

Presented at Bayesian Deep Learning workshop (NeurIPS 2018)

R2 v1 2026-06-23T06:34:44.751Z