Pull to the manifold autoencoder compressive sensing

I hope this works out well when I code it:
https://groups.google.com/forum/#!topic/artificial-general-intelligence/NLqBlanCSHw
It actually should be possible to use it with a selection of autoencoders operating at different scales somewhat like a convolutional network. Let’s see.

It looks like it works out okay-ish. You should try with a better autoencoder though.
https://drive.google.com/open?id=0BwsgMLjV0BnhaFNLTEt1TmN5YTA