3 ms·
umm no!...I don't know too much about this sub-field..but there are machine methods designed explicitly for adversarial data. it even has a wikipedia page! I do
by mathperson 10y ago
umm no!...I don't know too much about this sub-field..but there are machine methods designed explicitly for adversarial data. it even has a wikipedia page! I don't know if facebook's methods implement any of this yet..but your statements about machine learning not modelling this just ain't so.
https://en.wikipedia.org/wiki/Adversarial_machine_learning https://en.wikipedia.org/wiki/Adversarial_machine_learning
http://www.cs.virginia.edu/~evans/ http://www.cs.virginia.edu/~evans/
- jwtadvice 10y agoGreat resources! It's important to understand that these approaches are on the leading edge of research, and represent unsolved problems!
- mathperson 10y agoyes precisely! i'm sure Facebook's methods are very susceptible to these types of attacks. Hopefully they still are in 6 years and we can get jobs fixing this for them lol. I also really recommend this book https://www.amazon.com/Trust-Me-Lying-Confessions-Manipulator/dp/1591846285/ref=sr_1_1?ie=UTF8&qid=1479519844&sr=8-1&keywords=trust+me+im+lying https://www.amazon.com/Trust-Me-Lying-Confessions-Manipulato... The type of attack is a lot more oldschool but nonetheless, pretty effective.