Coming from the computer networking space this is somewhere I've been hoping to see a big machine learning splash like language translation has seen since large network graphs are either a big ball of mess or out of date because it takes too much human time to keep up to date.

It seems no matter how complicated you make the layout heuristics or what kind of dynamic iterating system you use when it comes to large connected graphs they just look nothing like a human would make. I'm not sure how would would get a large enough quality training set but it just feels like something that could be better done by machine learning than further manual definition.