Re:Optical illusions that flummox computers
by verroq on Saturday April 15, 2017 @12:13PM
(#14121361)
Note that to generate perturbations you need access to the underlying model. The attack is just optimising towards a specific class with added cost for visual differences.
[ Parent
| on HN ]
Re:Optical illusions that flummox computers
by mholt on Saturday April 15, 2017 @12:19PM
(#14121390)
That's not entirely true:
> If we transfer adversarial examples from one model to another model trained with one of these defenses, the attack often succeeds, even when a direct attack on the second model would fail [PMG16].
http://www.cleverhans.io/security/privacy/ml/2017/02/15/why-...