Boosting Transferability of Targeted Adversarial Examples via Hierarchical Generative Networks

Transfer-based adversarial attacks can evaluate model robustness in the black-box setting. Several methods have demonstrated impressive untargeted transferability, however, it is still challenging to efficiently produce targeted transferability. To this end, we develop a simple yet effective framework craft transfer-based examples, applying hierarchical generative network. In particular, contribute amortized designs that well adapt multi-class attacks. Extensive experiments on ImageNet show our method improves success rates of by significant margin over existing methods—it reaches an average rate 29.1% against six diverse models based only one substitute white-box model, which significantly outperforms state-of-the-art gradient-based attack methods. Moreover, proposed also more efficient beyond order magnitude than