This Monday, @ferrine presented Density estimation using Real NVP. It’s a set of invertible transformation: you can think of it as a more expressive version of normalizing flow.
You can find an implementation in Tensorflow on Github (https://github.com/taesung89/real-nvp), and also a short comment from Hugo Larochelle.
Related read:
Variational Inference with Normalizing Flows
One step further than RealNVP:
Multiplicative Normalizing Flows for Variational Bayesian Neural Networks
Improving Variational Inference with Inverse Autoregressive Flow, see Code here.