Skip to content

Latest commit

 

History

History

mnist_autoenc

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 

In this example, we train a MNIST autoencoder with a discrete channel as a Sender/Receiver game. The end-to-end training over the channel is done via Gumbel-Softmax relaxation.

The training can be started as:

python -m egg.zoo.mnist_autoenc.train --vocab_size=10 --n_epochs=50 --batch_size=16 --random_seed=7

Please refer to the MNIST autoencoder tutorial for more details.