cc @math_rachel @jeremyphoward, would love to see if your brilliant http://fast.ai students can make a dent in this one!
Show this thread
"Birds vs bicycles" is easy for ML in the average case, but *totally unsolved* in the worst case. For safety-critical applications, we *need* to fix this. We're launching the Unrestricted Adversarial Examples Challenge - *any* image of a bird or bike is a valid attack.https://twitter.com/nottombrown/status/1040286926857629696 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Is this equivalent to saying "find me an image of a bird that looks like a bicycle, or vice-versa"?
-
..."looks like a bicycle" to a neural net, that is. An image that all humans would say "yes, this is a bird, there is nothing like a bicycle about this, it is not ambiguous or confusing in any way", but the neural net labels it "bicycle" with high confidence.
-
Right. I was thinking of the fact that it should be easy for a network to pick up the difference, given that there are no similarities. But then I remembered the gibbon-panda example
-
Yeah, it's trivial in the average case - basically solved. But in those edge cases (that you can easily find with gradient descent!) it's laughably, terribly wrong.
End of conversation
New conversation -
-
-
What about a bird riding a bicycle...? :Dhttps://www.youtube.com/watch?v=PIM75j53k4g …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Indeed awesome Catherine
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.