Variational autoencoders trained to minimize the reconstruction error are sensitive to the posterior collapse problem, that is the proposal posterior distribution is always equal to the prior. We propose a novel regularization method based on fraternal dropout to prevent posterior collapse. We evaluate our approach using several metrics and observe improvements in all the tested configurations.
Cite
@article{arxiv.2110.14945,
title = {Preventing posterior collapse in variational autoencoders for text generation via decoder regularization},
author = {Alban Petit and Caio Corro},
journal= {arXiv preprint arXiv:2110.14945},
year = {2021}
}
Comments
Accepted at NeurIPS 2021 Workshop DGMs Applications