Black-Box Based Limited Query Membership Inference Attack

Conventional membership inference attacks usually require a large number of queries the target model when training shadow models, and this task becomes extremely difficult is limited. Aiming at problem insufficient data for models due to limited queries, we propose attack method based on generative adversarial networks (GAN). First, use augment samples obtained by small expand model; Secondly, improved CNN obtain that have higher degree fitting different structures; Finally, evaluate accuracy proposed algorithm XgBoost, Logistic, neural network using public datasets MNIST CIFAR10 in black-box setting, results show our has an average 62% 83%, respectively. It can be seen that, compared with existing research methods, better effects under condition significantly reducing which shows feasibility attacks.