How are optical illusions fundamentally different from adversarial examples? In both cases, the systems are hardwired to recognize something that is not there. Our conscious minds are unable to "unsee" what is prevented before it, just like DL nets are unable to ignore AE.
this beautiful illusion (described here http://journals.sagepub.com/doi/10.1177/2041669517742178 …) is about conflict between top-down cues (re surfaces) and bottom-cues. adversarial spoofing is different: a consequence of LACKING top-down cues.https://twitter.com/intuitmachine/status/952701525159301120 …
-
-
-
Many optical illusions can be explained as approximate inference from sparse clues. Lee&Mumford paper is a good one to read on this https://www.ncbi.nlm.nih.gov/pubmed/12868647 . Very different from the AE's in NN's.
End of conversation
New conversation -
-
-
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.