Sample from the above paper: "give me a picture that looks like a tiger cat, but is classified as a dalmatian dog". Imagine the implications on e.g. autonomous driving!pic.twitter.com/TO7SYDT16V
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Sample from the above paper: "give me a picture that looks like a tiger cat, but is classified as a dalmatian dog". Imagine the implications on e.g. autonomous driving!pic.twitter.com/TO7SYDT16V
Da gabs auch nen Talk zu für Leute die trockene Papers nicht so mögenhttps://media.ccc.de/v/34c3-8860-deep_learning_blindspots …
That talk is about a different attack method that works only under very specific circumstances (e.g. the recognition of very simple objects). It wouldn't work in real-life applications such as autonomous cars or speech-to-text systems like Siri.
Speaker here. I actually cited your library as well as a variety of other methods than simply FGSM, so I would say the talk overall does not focus on any one specific method -- but an overview of available open-source options.
Thanks for the clarification! Btw, your "Facebook adversarial" is a nice example. It made me wonder whether I should try our attack in the same scenario. Haven't had time yet but lets see.
Cool! Lmk if you test it out. I got a tip that FB exposes classification via alt text as well in case you want to test target classes (also on my TODO list to try with Foolbox) 
You are right, FB is really using alt-text to expose classification (e.g. "2 people, 1 person that is sitting and 1 child"...). Amazing! Here is a good article about it: https://www.wired.com/2016/04/facebook-using-ai-write-photo-captions-blind-users/ …. Sounds like an interesting target to test the Boundary Attack.
This is super important stuff. When talking about AI dangers people think of Skynet, but attacking the networks themselves *and* the learning (remember MS' Tay bot?) are a tiny bit closer to reality. And can hurt real life applications pretty damn badly.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.