Abstract
We investigate how an adversary can optimally use its query budget for targeted evasion attacks against deep neural networks in a black-box setting. We formalize the problem setting and systematically evaluate what benefits the adversary can gain by using substitute models. We show that there is an exploration-exploitation tradeoff in that query efficiency comes at the cost of effectiveness. We present two new attack strategies for using substitute models and show that they are as effective as previous query-only techniques but require significantly fewer queries, by up to three orders of magnitude. We also show that an agile adversary capable of switching through different attack techniques can achieve pareto-optimal efficiency. We demonstrate our attack against Google Cloud Vision showing that the difficulty of black-box attacks against real-world prediction APIs is significantly easier than previously thought (requiring approximately 500 queries instead of approximately 20,000 as in previous works).
Original language | English |
---|---|
Title of host publication | AISec'19: Proceedings of the 12th ACM Workshop on Artificial Intelligence and Security |
Publisher | ACM |
Pages | 83–94 |
Number of pages | 12 |
ISBN (Print) | 978-1-4503-6833-9 |
DOIs | |
Publication status | Published - 2019 |
MoE publication type | A4 Article in a conference publication |
Event | ACM Workshop on Artificial Intelligence and Security - London, United Kingdom Duration: 15 Nov 2019 → 15 Nov 2019 Conference number: 12 https://aisec.cc/ |
Workshop
Workshop | ACM Workshop on Artificial Intelligence and Security |
---|---|
Abbreviated title | AISec |
Country | United Kingdom |
City | London |
Period | 15/11/2019 → 15/11/2019 |
Internet address |
Keywords
- adversarial examples
- Neural Networks