Deep Recurrent Attentive Writer

This post implements the Deep Recurrent Attentive Writer

Many people preceded me with fascinating implementations

• Am implementation in Torch
• An implementation in Theano
• An implementation in Tensorflow with Pretty Tensor
• A clean implementation in Tensorflow

This implementation goes end to end in 250 lines of code. We start at import MNIST and end up with these lovely visualizations.

Read

Write

This series of posts, implementing an Variational autoencoder, followed by a variational recurrent autoencoder and now the DRAW was inspired after Karol Gregor's lecture at Oxford. More recently, he ties his work together at ICML (or use this link).

Moreover if you'd like a better introduction to variational inference, go for this talk by Diederik Kingma at ICLR or go for his paper.

As always, I am curious to any comments and questions. Reach me at romijndersrob@gmail.com