Symmetric Variational Autoencoder and Connections to Adversarial Learning

Liqun Chen; Shuyang Dai; Yunchen Pu; Erjin Zhou; Chunyuan Li; Qinliang Su; Changyou Chen; Lawrence Carin

Symmetric Variational Autoencoder and Connections to Adversarial Learning

Liqun Chen
, Shuyang Dai
, Yunchen Pu
, Erjin Zhou
, Chunyuan Li
, Qinliang Su
, Changyou Chen
, Lawrence Carin

Department of Computer Science and Engineering

Duke University
Megvii Technology Limited
Sun Yat-Sen University

Research output: Contribution to journal › Conference article › peer-review

Abstract

A new form of the variational autoencoder (VAE) is proposed, based on the symmetric KullbackLeibler divergence. It is demonstrated that learning of the resulting symmetric VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship helps unify the previously distinct techniques of VAE and adversarially learning, and provides insights that allow us to ameliorate shortcomings with some previously developed adversarial methods. In addition to an analysis that motivates and explains the sVAE, an extensive set of experiments validate the utility of the approach.

Original language	English
Journal	Proceedings of Machine Learning Research
Volume	84
State	Published - 2018
Event	21st International Conference on Artificial Intelligence and Statistics, AISTATS 2018 - Playa Blanca, Lanzarote, Canary Islands, Spain Duration: Apr 9 2018 → Apr 11 2018

Cite this

@article{d24989d4cc134d13b9971be331ff8697,

title = "Symmetric Variational Autoencoder and Connections to Adversarial Learning",

abstract = "A new form of the variational autoencoder (VAE) is proposed, based on the symmetric KullbackLeibler divergence. It is demonstrated that learning of the resulting symmetric VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship helps unify the previously distinct techniques of VAE and adversarially learning, and provides insights that allow us to ameliorate shortcomings with some previously developed adversarial methods. In addition to an analysis that motivates and explains the sVAE, an extensive set of experiments validate the utility of the approach.",

author = "Liqun Chen and Shuyang Dai and Yunchen Pu and Erjin Zhou and Chunyuan Li and Qinliang Su and Changyou Chen and Lawrence Carin",

note = "Publisher Copyright: {\textcopyright} 2018 by the author(s).; 21st International Conference on Artificial Intelligence and Statistics, AISTATS 2018 ; Conference date: 09-04-2018 Through 11-04-2018",

year = "2018",

language = "English",

volume = "84",

journal = "Proceedings of Machine Learning Research",

issn = "2640-3498",

publisher = "ML Research Press",

}

TY - JOUR

T1 - Symmetric Variational Autoencoder and Connections to Adversarial Learning

AU - Chen, Liqun

AU - Dai, Shuyang

AU - Pu, Yunchen

AU - Zhou, Erjin

AU - Li, Chunyuan

AU - Su, Qinliang

AU - Chen, Changyou

AU - Carin, Lawrence

PY - 2018

Y1 - 2018

N2 - A new form of the variational autoencoder (VAE) is proposed, based on the symmetric KullbackLeibler divergence. It is demonstrated that learning of the resulting symmetric VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship helps unify the previously distinct techniques of VAE and adversarially learning, and provides insights that allow us to ameliorate shortcomings with some previously developed adversarial methods. In addition to an analysis that motivates and explains the sVAE, an extensive set of experiments validate the utility of the approach.

AB - A new form of the variational autoencoder (VAE) is proposed, based on the symmetric KullbackLeibler divergence. It is demonstrated that learning of the resulting symmetric VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship helps unify the previously distinct techniques of VAE and adversarially learning, and provides insights that allow us to ameliorate shortcomings with some previously developed adversarial methods. In addition to an analysis that motivates and explains the sVAE, an extensive set of experiments validate the utility of the approach.

UR - https://www.scopus.com/pages/publications/105020018105

M3 - Conference article

AN - SCOPUS:105020018105

SN - 2640-3498

VL - 84

JO - Proceedings of Machine Learning Research

JF - Proceedings of Machine Learning Research

T2 - 21st International Conference on Artificial Intelligence and Statistics, AISTATS 2018

Y2 - 9 April 2018 through 11 April 2018

ER -

Symmetric Variational Autoencoder and Connections to Adversarial Learning

Abstract

Other files and links

Fingerprint

Cite this