Abstract: Deep neural networks (DNNs) are vulnerable to adversarial examples, which subtly alter benign images to mislead predictions. While white-box attacks are highly effective, black-box attacks ...