Beronx86 / draw

Reimplementation of DRAW

Reimplementation of the DRAW network architecture

This repository contains a reimplementation of the Deep Recurrent Attentive Writer (DRAW) network architecture introduced by K. Gregor, I. Danihelka, A. Graves and D. Wierstra. The original paper can be found at

http://arxiv.org/pdf/1502.04623

Dependencies

• Theano
• Blocks
• Fuel

With attention

To train a model with a 2x2 read and a 5x5 write attention window run

./train-draw --attention=2,5 --niter=64 --lr=3e-4 --epochs=100

It will take about 1 day to train and will create samples similar to

with the KL divergence plottet over inference iterations and epochs

Note

Work in progress