Click to rebuild a grid pattern after it squeezes through a narrow middle layer. An autoencoder learns by copying its own input this way.
The task is simply: reproduce your own input. It sounds pointless until you look at the narrow part in the middle.
