Abstract
Machine learning algorithms are prone to attacks: An attackers can use the malicious nodes to attack the training dataset to manipulate the process of learning and reduce the efficiency of the algorithm working performance. Optimal poisoning attacks have already been proposed to evaluate worst case scenarios, modelling attacks as a bilevel optimization problem. Solving these problems is computationally demanding and has limited applicability for some models such as deep networks. In this paper we introduce a novel generative model to craft systematic poisoning attacks against machine learning classifiers generating adversarial training examples, i.e. samples that look like genuine data points but that reduce the accuracy of the classifier in the process of training process. The proposed system have 3 components of Generative Adverserial networks (GAN) generator, discriminator, and the target classifier. The proposed system allows to detect the vulnerability easy and it can be found as similar as realistic attacks to detect the area where the underlying data distribution have more possibility of poising attack which cause vulnerability to the network. Our experimentation, proves the claim our that the proposed model is effective on compromising the classifiers uses the machine learning algorithms and also deep learning networks.