Adversarial Symmetric Variational Autoencoder

TL;DR

This paper introduces a symmetric variational autoencoder that uses a novel adversarial training approach to improve data reconstruction and generation, achieving state-of-the-art results on image benchmarks.

Contribution

It develops a symmetric VAE framework with a new adversarial training method and joint distribution modeling, advancing generative modeling techniques.

Findings

Achieves state-of-the-art image reconstruction results

Demonstrates improved data generation quality

Validates the effectiveness of symmetric joint distribution modeling

Abstract

A new form of variational autoencoder (VAE) is developed, in which the joint distribution of data and codes is considered in two (symmetric) forms: ($i$) from observed data fed through the encoder to yield codes, and ($ii$) from latent codes drawn from a simple prior and propagated through the decoder to manifest data. Lower bounds are learned for marginal log-likelihood fits observed data and latent codes. When learning with the variational bound, one seeks to minimize the symmetric Kullback-Leibler divergence of joint density functions from ($i$) and ($ii$), while simultaneously seeking to maximize the two marginal log-likelihoods. To facilitate learning, a new form of adversarial training is developed. An extensive set of experiments is performed, in which we demonstrate state-of-the-art data reconstruction and generation on several image benchmark datasets.