“Universal adversarial perturbations” seems most dramatic ML result in years; if so, not getting deserved attention https://arxiv.org/pdf/1610.08401v1.pdf …
-
-
I have long suspected that DL image classifiers depend mainly on texture and maybe color, making much less use of shape than vertebrates do…
-
The form of the universal adversarial perturbation is consistent with this hypothesis. It subtly screws up texture/color info:pic.twitter.com/tv3tRPp1UF
- 6 more replies
New conversation -
-
-
I looked for, but couldn't find, a lazy way to run code to generate those patterns (e.g.Matlab, Jupyter). Do you have one?
-
No… you could contact the authors of the paper I guess?
- 2 more replies
New conversation -
-
-
@logodaedalus it might be jpeg artifacts but "human invisible" doesn't describe those images. I see a wood grain type pattern -
Yes, “invisible” is a bit of an exaggeration—clearest in the “parrot.” 140 chars makes precision difficult smtimes
End of conversation
New conversation -
-
-
@fuzzleonard pitch: stenographically encoded hoodies/sportswear to deliberately confuse the upcoming TensorFlow drone swarm
-
Drone 1: Is that Jeff Hobbes? Drone 2: Nah, it’s a philodendron walking its pet African elephant.
End of conversation
New conversation -
-
-
@bklimt cool, but not surprising from earlier rslts. Basic issue is dot products on randomish vectors + Central limit theorem. -
causes real data to have small dot products, and adversarial data to have big dot products as dimensionality increases.
End of conversation
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.