Multi-target Category Adversarial Example Generating Algorithm Based on GAN

Deep neural networks perform well in many fields, but research shows that they are easily attacked by adversarial samples. At present, there are many algorithms for attacking neural networks, but the attack speed of most attack algorithms is slow, so the rapid generation of adversarial samples has gradually become a The focus of research in the field of adversarial samples. AdvGAN is an algorithm that uses a network to attack the network. The speed of generating adversarial samples is extremely fast, but when a targeted attack is performed, it needs to train a network for each target, which makes the attack less efficient. Aiming at the above problems, a multi-target attack network MTA based on generative adversarial network is proposed. MTA only needs to be trained once to complete multi-target attacks and generate adversarial samples quickly. The success rate of targeted attacks on the dataset is higher than that of Adv GAN. The paper also conducts adversarial sample migration experiments and attack experim