A Self-Adaptive Approximated-Gradient-Simulation Method for Black-Box Adversarial Sample Generation

Deep neural networks (DNNs) have famously been applied in various ordinary duties. However, DNNs are sensitive to adversarial attacks which, by adding imperceptible perturbation samples an original image, can easily alter the output. In state-of-the-art white-box attack methods, successfully fool through network gradient. addition, they generate only considering sign information of gradient and dropping magnitude. Accordingly, gradients different magnitudes may adopt same construct samples, resulting inefficiency. Unfortunately, it is often impractical acquire real-world scenarios. Consequently, we propose a self-adaptive approximated-gradient-simulation method for black-box (SAGM) efficient samples. Our proposed uses knowledge-based differential evolution simulate momentum To estimate efficiency SAGM, series experiments were carried out on two datasets, namely MNIST CIFAR-10. Compared techniques, our quickly efficiently search misclassify The results reveal that SAGM effective technique generating