Hacker News Reader

Kind of agree, I think if they have a good way to create the underlying dataset, a NN to go in the opposite direction might be more interesting.

This would be very similar to the "super-resolution" approach.

E.g.,

https://medium.com/beyondminds/an-introduction-to-super-reso...

Sorta, it breaks down the images (from anime?) into three representations - surface, structure, and details, and also extracts each of those representations from generated images. Those representations are then cross-checked by the adverserial network, which improves the GAN's anime-esque generation ability.

I know in the case of physics simulations the neural net can perform better than the classical algorithm. Not sure if that's the case here but just thought it was worth mentioning.

By "perform better" I assume you mean time/memory performance and not accuracy