3 ms·
"The field of adversarial machine learning has experienced a near exponential growth in the amount of papers being produced since 2018. This massive informati
by belter 5y ago
"The field of adversarial machine learning has experienced a
near exponential growth in the amount of papers being produced
since 2018. This massive information output has yet to be
properly processed and categorized. In this paper, we seek
to help alleviate this problem by systematizing the recent
advances in adversarial machine learning black-box attacks
since 2019.
Our survey summarizes and categorizes 20 recent black-box attacks.
We also present a new analysis for understanding the attack success
rate with respect to the adversarial model used in each paper.
Overall, our paper surveys a wide body of literature to highlight
recent attack developments and organizes them into four attack
categories: score based attacks, decision based attacks,
transfer attacks and non-traditional attacks. Further,
we provide a new mathematical framework to show exactly
how attack results can fairly be compared."
https://arxiv.org/pdf/2109.15031.pdf https://arxiv.org/pdf/2109.15031.pdf