ELI5: evasion attacks
// explanation
What are evasion attacks?
Evasion attacks are tricks that try to fool AI systems into making wrong decisions [2][3]. It's like slightly changing a picture so a computer can't recognize it correctly, even though a human would still see the same thing [2].
Why do people use them?
Evasion attacks work by making small changes to the information you send to an AI model [2]. The changes are so tiny that they might not be noticeable, but they can confuse the AI and make it give the wrong answer [3].
What happens when they work?
When an evasion attack succeeds, the AI model outputs incorrect results even though it's the same trained system [2]. It's like tricking a security guard by wearing a clever disguise that doesn't look obviously fake [3].
How can we stop them?
People can test their AI systems to find weaknesses before attackers exploit them [1]. By understanding how evasion attacks work, companies can build better defenses to protect their AI models [1].
// sources
This paper presents an overview of evasion attacks on machine learning models, its variants and conducts an adaptive white-box evasion attack.
Evasion attacks attempt to make a model output incorrect results by slightly perturbing the input data sent to the trained model.
Evasion attacks are a type of cyber attack that involves manipulating input data to evade detection or classification by a machine learning model.
In this paper, we first explore the evadable physical event fingerprints and show that an adversary can realize them to bypass the EVS given the same threatย ...
Aug 21, 2017 ... In this work, we present a simple but effective gradient-based approach that can be exploited to systematically assess the security of several, widely-usedย ...
Video by TechX Academy
Video by SOCDemystified
Video by Ryan Wiebe
