7 ms·
It would be amazing if this could become the future, but as someone working in the adversarial robustness space I find it difficult to believe a system like thi
by trojstan 6y ago
It would be amazing if this could become the future, but as someone working in the adversarial robustness space I find it difficult to believe a system like this wouldn't be susceptible to adversarial patching...especially with whitebox access. It's been shown by Nicolas Carlini that nearly all adversarial defenses can be broken or are inherently flawed.
Very curious to see what makes this anti-spoofing algo different, will read the paper with interest.
- emilxix 6y agoright, adversarial attacks are tricky, we wrote about it a while ago - https://blog.ycombinator.com/how-adversarial-attacks-work/ https://blog.ycombinator.com/how-adversarial-attacks-work/ In reality, the production-grade security comes from a compound effect of three components: face-recognition, antispoofing for face recognition, and traditional controls of industry-standard protocols like SAML 2.0, OIDC, etc. Taking one of three out of the equation renders security nonexistent.
- JohnOffenhartz 6y agoAlso worth mentioning that since we're cloud-based, we can shore up our threat assessment models to include browser fingerprinting, device details, IP address history, what network the user is on, etc. to better detect threats/spoofing.
- 177tcca 6y agoThese seem like the main added-values.