HOME Board
Notice

Notice

Hit 515
Subject [IEEE TNNLS] Advancing Adversarial Training by Injecting Booster Signal (by Hong Joo Lee and Youngjoon Yu) is accepted in IEEE Transactions on Neural Networks and Learning Systems
Name °ü¸®ÀÚ
Date 2023-03-27
Title: Advancing Adversarial Training by Injecting Booster Signal


Authors: Hong Joo Lee and Youngjoon Yu, and Yong Man Ro


Recent works have demonstrated that deep neural networks (DNNs) are highly vulnerable to adversarial attacks. To defend against adversarial attacks, many defense strategies have been proposed, among which adversarial training has  been demonstrated to be the most effective strategy. However,it has been known that adversarial training sometimes hurtsnatural accuracy. Then, many works focus on optimizing modelparameters to handle the problem. Different from the previousapproaches, in this paper, we propose a new approach to improvethe adversarial robustness by using an external signal rather thanmodel parameters. In the proposed method, a well-optimizeduniversal external signal called a booster signal is injected to theoutside of the image which does not overlap with the originalcontent. Then, it boosts both adversarial robustness and naturalaccuracy. The booster signal is optimized in parallel to modelparameters step by step collaboratively. Experimental resultsshow that the booster signal can improve both the natural androbust accuracies over the recent state-of-the-art adversarialtraining methods. Also, optimizing the booster signal is generaland flexible enough to be adopted on any existing adversarialtraining methods.


IMAGE VIDEO SYSTEM (IVY.) KOREA ADVANCED INSTITUTE OF SCIENCE AND TECHNOLOGY (KAIST), IEEE TNNLS