Go to DBLP homepageA hypothetical defenses-based training framework for generating transferable adversarial examples
Abstract: Highlights•A novel training framework TA-HD for transfer-based attacks.•A pair of novel adversarial loss functions for optimizing the parameters of hypothetical defenses.•Extensive experiments demonstrate that TA-HD markedly improves transfer-based attacks by up to 19.9%.
External IDs:dblp:journals/kbs/HaoHJZ24
Loading