Adversarial Machine Learning
A subfield of machine learning that studies attack and defense methods targeting machine learning models. Its central object of study is the adversarial example—an input sample in which small perturbations cause a model to produce incorrect outputs.
Timeline (last 90 days)
In a lecture by Nicholas Carlini, one slide was titled '9,000 Papers and Nowhere to Go,' referring to the field of adversarial machine learning.
The core dilemma of adversarial machine learning lies in the asymmetry between attack and defense: an attacker only needs to find one effective perturbation, while a defender must remain robust against all possible attacks.
All Facts (2)
In a lecture by Nicholas Carlini, one slide was titled '9,000 Papers and Nowhere to Go,' referring to the field of adversarial machine learning.
50%UnverifiedThe core dilemma of adversarial machine learning lies in the asymmetry between attack and defense: an attacker only needs to find one effective perturbation, while a defender must remain robust against all possible attacks.
50%