If you want to know what the "state of the art" generalization bounds for neural networks are, well, I don't even know, because they all suck so bad it's hardly worth a horse race. However, I know damn well sure which bounds aren't "SOTA".
-
-
I'll take a closer look, but I'm slightly skeptical seeing this plot.pic.twitter.com/oL7Fo5V9dx
-
This is the effective number of parameters (proportional to the generalization bound) compared with the one derived by the current state-of-the-art (Arora et al., 2018) for VGG-16.
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.