Building a generative model

We will use a Variational AutoEncoder (VAE) as a generator factory. The VAE is based on an example in the Tensorflow documentation, kept simple, for speed: the generator and encoder are 5-layer convolutional neural nets.

_images/DCVAE_structure.jpg

The structure of the VAE used to train the generator

We train the VAE on 40-years of daily temperature anomalies from ERA5.