Learning-Priors-for-Adversarial-Autoencoders

Experimental results for learning priors for adversarial autoencoders

Content

• Quantitative Analysis
  • Inception Scores Evaluation
• Image Generation
• Learning Disentangled Representations (Supervised)
• Learning Disentangled Representations (Unsupervised)
• Latent Space Visualization

Quantitative Analysis

Inception Scores Evaluation on CIFAR10

	Proposed	Baseline
Image generation	7.31	6.37
Supervised	6.52	6.2
Unsupervised	6.02	5.73

Image Generation

MNIST

Proposed	Baseline

CIFAR-10

Proposed	Baseline

Learning Disentangled Representations (Supervised)

MNIST

Proposed	Baseline

SVHN

Proposed	Baseline

CIFAR-10

Proposed	Baseline

Learning Disentangled Representations (Unsupervised)

MNIST

Proposed	Baseline

SVHN

Proposed	Baseline

CIFAR-10

Proposed	Baseline

Latent Space Visualization

MNIST - Supervised Settings

Proposed - Encoder	Proposed - Learned Prior	Baseline - Fixed Prior

CIFAR-10 - Supervised Settings

Proposed - Encoder	Proposed - Learned Prior	Baseline - Fixed Prior

MNIST - Unsupervised Settings

Proposed - Encoder	Proposed - Learned Prior	Baseline - Fixed Prior

CIFAR-10 - Unsupervised Settings

Proposed - Encoder	Proposed - Learned Prior	Baseline - Fixed Prior