DanielFilan comments on The other paper that killed deep learning theory

[DanielFilan]

Oh I currently think the thing that’s going on is that it’s a hypothesis-dependent bound that you then apply to the hypothesis learned from the data.

[LawrenceC]

Yep