9
$\begingroup$

I read that deep neural networks can be relatively easily fooled (link) to give high confidence in recognition of synthetic/artificial images that are completely (or at least mostly) out of the confidence subject.

Personally, I don't really see a big problem with DNN giving high confidence to those synthetic/artificial images but I think giving high confidence for white noise (link) may be a problem since this is a truly natural phenomenon that may the camera see in the real world.

How much of a problem is white noise for the real-world usage of a DNN? Can such false positives be detected from plain noise be prevented somehow?

$\endgroup$
7
$\begingroup$

The white noise that fools DNNs isn't really white noise. It has been altered in the same way as the synthetic misclassified pictures have been altered. You have to change many input pixels in exactly such a way, that these little changes aren't perceptible, but propagated through the network add up to a misclassification. This is not going to happen by chance.

$\endgroup$
  • 1
    $\begingroup$ Ok, so if I understand this correctly, such a misclassification can not happen with natural noise... that is a good news :). $\endgroup$ – Kozuch Feb 24 '17 at 11:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.