핵심 개념
Deep neural networks are vulnerable to adversarial noise, and pre-processing methods can enhance white-box robustness by utilizing full adversarial examples.
초록
Deep neural networks face vulnerabilities from adversarial noise.
Pre-processing methods aim to improve white-box robustness.
Full adversarial examples positively impact defense robustness.
Joint Adversarial Training based Pre-processing (JATP) defense proposed.
JATP minimizes the robustness degradation effect across different target models.
통계
A potential cause of the negative effect is that adversarial training examples are static and independent to the pre-processing model.
Using full adversarial examples improves the robustness of defenses compared to oblivious ones.
인용구
"Using full adversarial examples could improve the white-box robustness of the pre-processing defense."