Boosting the Transferability of Adversarial Examples with Translation Transformation

Although the adversarial examples have achieved an incredible white-box attack rate, they tend to show poor transferability in black-box attacks. Date augmentation is considered be effective means of enhancing transferability. To this end, based on translation transformation we propose a new method generate more mobile advanced defense model. By optimizing original image, input diversity improved and generated by further training enhanced. This can also combined with gradient, make success rate higher. Experiments ImageNet data sets that proposed superior gradient methods such as MI-FGSM black box attack, while maintaining high white attack. We hope our serve benchmark for assessing robustness networks opponents effectiveness different defence.