VAE

Auto-Encoding Variational Bayes

Diederik P. Kingma, Max Welling · ICLR 2014

Read paper · arXiv:1312.6114

Implement differentiable Gaussian sampling, analytic KL to a standard normal and a stable Bernoulli negative ELBO.

What you will implement

Complete three functions; the supplied encoder/decoder train for 120 steps on synthetic binary patterns. No MNIST or image-generation benchmark is reproduced.

• 99. VAE: Reparameterization and the Bernoulli ELBO

  Implement differentiable Gaussian sampling, analytic KL to a standard normal and a stable Bernoulli negative ELBO.

  Complete three functions; the supplied encoder/decoder train for 120 steps on synthetic binary patterns. No MNIST or image-generation benchmark is reproduced.