LatentGAN Autoencoder: Learning Disentangled Latent Distribution

TL;DR

This paper introduces LatentGAN Autoencoder, which learns a disentangled latent distribution to improve control over generated outputs, achieving better classification error rates on MNIST.

Contribution

It proposes using LatentGAN to directly learn the autoencoder's latent distribution with an information-theoretic constraint for better control and performance.

Findings

Achieves 2.38% error rate on MNIST classification

Demonstrates meaningful latent control on MNIST, 3D Chair, and CelebA

Outperforms InfoGAN and AAE in classification accuracy

Abstract

In autoencoder, the encoder generally approximates the latent distribution over the dataset, and the decoder generates samples using this learned latent distribution. There is very little control over the latent vector as using the random latent vector for generation will lead to trivial outputs. This work tries to address this issue by using the LatentGAN generator to directly learn to approximate the latent distribution of the autoencoder and show meaningful results on MNIST, 3D Chair, and CelebA datasets, an additional information-theoretic constrain is used which successfully learns to control autoencoder latent distribution. With this, our model also achieves an error rate of 2.38 on MNIST unsupervised image classification, which is better as compared to InfoGAN and AAE.