Disentangling by Factorising

FactorVAE：通过分解特征表示的分布进行解耦.

Sliced-Wasserstein Autoencoder: An Embarrassingly Simple Generative Model

SWAE：引入Sliced-Wasserstein距离构造VAE.

Log Hyperbolic Cosine Loss Improves Variational Auto-Encoder

使用对数双曲余弦损失改进变分自编码器.

Learning to Generate Images with Perceptual Similarity Metrics

使用多尺度结构相似性度量MS-SSIM学习图像生成.

Learning Disentangled Joint Continuous and Discrete Representations

Joint VAE：学习解耦的联合连续和离散表示.

Categorical Reparameterization with Gumbel-Softmax

使用Gumble-Softmax实现离散类别隐变量的重参数化.