Abstract: Transfer-based adversarial attacks are key for evaluating the robustness of deep neural networks (DNNs) in black-box settings, yet their effectiveness is often constrained by limited ...