right, but the computationally inefficent part, as far as I'm seeing, is the training. once it's trained, a compression algorithm could run on orders of magnitude less computation.

I mean, I'm just spitballin here, i don't have any high level experience in the arena.