Hacker News new | past | comments | ask | show | jobs | submit login
Disrupting Deepfakes: Adversarial Attacks on Image Translation Networks (Code) (github.com)
46 points by natanielruiz 13 days ago | hide | past | web | favorite | 8 comments

Doesn't this make it worse because you can just make another GAN or something that undoes this?

Yes, it's a never ending race.

For now, we show that if StarGAN is trained on images augmented by adversarial attack it does become more resistant but not completely resistant to attacks.


If you are going to shill comment with a two-hour old account you should make the comment a little bit less like the comment above it!

Please, you can do better.

I was going to take issue with your comment, but I noticed the two other older commenters ('MITpmt, 'lauvidi) were created on the same day, and one of them had posted only one other comment, on an item submitted by the same original poster.

That is pretty strange, huh?

I mean.. maybe they're colleagues, and he posted the link to his post in like a Discord, or a Slack. Why be so cynical?

Why be cynical? Because the web is frought with fake views and click/comment farms and "growth hacking" and "native advertising".

I posted this on a small Facebook group with some of my friends. Those people commenting were some of my best friends.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact