What are adversarial attacks on AI models, and how do they pose a threat to AI-driven defenses?
- From Live Chat
- 4 replies
- 2 views
- Author: Thomson A.
- Category: General
- Date asked:
- Last update:
-
Anonymous
There are a lot of adversarial attacks! Do you have any in particular you want to chat about, or I can give a general overview of the ones we come across the most/I find most interesting?
-
Anonymous
Not sure if you're referring to adversarial inputs, which is an attack where the actual individual input data is slightly modified (imperceptible to humans) and exploits the model's ability to make decisions - I find this really cool as it could be something as simple as changing a few pixels in a picture of a giraffe which you upload to the AI model and it can break it!
-
TAThomson A. Candidate
In particular MITREE Attack
-
Anonymous
I think a lot fall into the "Defence Evasion" part of the framework. It is so tricky to keep up though when you have AI models attacking AI models that are meant to defend against other AI models... so definitely a change because they are all full of innovative attacks and defences!
Have a question of your own?
What makes EY different from the other firms in the Big Four?
Why us, Company comparison- 5 replies
- 31439 views
How will technology change the nature of work conducted in technology consulting and more broadly across EY?
General- 6 replies
- 2777 views
What is the difference between Transactions Support and Transactions Restructuring?
Role, Role description- 1 replies
- 1932 views
