Saved in:
Bibliographic Details
Main Authors: Dorta, Gara, Vicente, Sara, Agapito, Lourdes, Campbell, Neill D. F., Simpson, Ivor
Format: Preprint
Published: 2018
Subjects:
Online Access:https://arxiv.org/abs/1804.01050
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917491744178176
author Dorta, Gara
Vicente, Sara
Agapito, Lourdes
Campbell, Neill D. F.
Simpson, Ivor
author_facet Dorta, Gara
Vicente, Sara
Agapito, Lourdes
Campbell, Neill D. F.
Simpson, Ivor
contents Variational auto-encoders (VAEs) are a popular and powerful deep generative model. Previous works on VAEs have assumed a factorized likelihood model, whereby the output uncertainty of each pixel is assumed to be independent. This approximation is clearly limited as demonstrated by observing a residual image from a VAE reconstruction, which often possess a high level of structure. This paper demonstrates a novel scheme to incorporate a structured Gaussian likelihood prediction network within the VAE that allows the residual correlations to be modeled. Our novel architecture, with minimal increase in complexity, incorporates the covariance matrix prediction within the VAE. We also propose a new mechanism for allowing structured uncertainty on color images. Furthermore, we provide a scheme for effectively training this model, and include some suggestions for improving performance in terms of efficiency or modeling longer range correlations.
format Preprint
id arxiv_https___arxiv_org_abs_1804_01050
institution arXiv
publishDate 2018
record_format arxiv
spellingShingle Training VAEs Under Structured Residuals
Dorta, Gara
Vicente, Sara
Agapito, Lourdes
Campbell, Neill D. F.
Simpson, Ivor
Machine Learning
Computer Vision and Pattern Recognition
Variational auto-encoders (VAEs) are a popular and powerful deep generative model. Previous works on VAEs have assumed a factorized likelihood model, whereby the output uncertainty of each pixel is assumed to be independent. This approximation is clearly limited as demonstrated by observing a residual image from a VAE reconstruction, which often possess a high level of structure. This paper demonstrates a novel scheme to incorporate a structured Gaussian likelihood prediction network within the VAE that allows the residual correlations to be modeled. Our novel architecture, with minimal increase in complexity, incorporates the covariance matrix prediction within the VAE. We also propose a new mechanism for allowing structured uncertainty on color images. Furthermore, we provide a scheme for effectively training this model, and include some suggestions for improving performance in terms of efficiency or modeling longer range correlations.
title Training VAEs Under Structured Residuals
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/1804.01050