How to optimize speed or setup

[root2]

And there's those approaches that base on some feedback. But feedback, that make no sense for slow hashes. For example NN. I can't understand how such an paper gets accepted to BH. The NN approach needs a fast hash to be useful. But then, there's no way to implement NN fast enough for a fast hash. So we can't use it for slow hashes and we can't use it for fast hashes. There's nothing left makes this approach a typical theoretical attack which is useless in a real-life scenario.

Never heard of "NN" before. Can you provide a link or a title of the talk? BH => BlackHat, is a hacker conference and quite often does not reach the academic audience. Thx.

Edit:
Do you mean NN => Neural Networks?
The aforementioned paper was accepted at USENIX Security conference in Austin, Texas, USA this year. It even won the best paper award.