3 ms·
Very patronizing of you. You clearly never learned that both classical (like your "sample size" example) and advanced statistical models are susceptible to firs
by zekrioca 5y ago
Very patronizing of you. You clearly never learned that both classical (like your "sample size" example) and advanced statistical models are susceptible to first-order evasion attacks (or “Adversarial Examples”) that fool models at run-time. But go trust Apple's 1/1.000.000.000.000 claims.
- zepto 5y agoObviously such an attack has nothing to do with the false positive rate. But, setting that aside, can you explain how a first order evasion attack can be used against Apple’s mechanism? They are a real kind of attack in the lab, but it’s not obvious how they could be used to exploit Apple’s CSAM detection. If you have reason to think they are a real threat, I’m sure you can explain.
- zekrioca 5y ago> Obviously such an attack has nothing to do with the false positive rate. At face value, they indeed don’t. But the premise from Apple’s white paper that their mechanism can’t be tricked is not realistic. > If you have reason to think they are a real threat, I’m sure you can explain. Yes, I can.
- zepto 5y ago> At face value, they indeed don’t. So this wasn’t relevant. > But the premise from Apple’s white paper that their mechanism can’t be tricked is not realistic. They don’t say it can’t be tricked. You are misrepresenting them. >> If you have reason to think they are a real threat, I’m sure you can explain. > Yes, I can. No you can’t, because they are not a real threat.