Implementation of a convolutional autoencoder in Julia using Flux.
All of the interesting info can be found in the Jupyter Notebook. In summary, a model learns how to represent a whole image of a digit in just two numbers (which is a 99.21% compression rate).