The vulnerability of Deep Neural Network (DNN) to adversarial attacks has become an issue in recent years. Attackers can degrade the accuracy of DNN models by introducing computational noise in the input data. Graph Convolutional Network (GCN), a derivative of DNN, have also been found to have similar vulnerabilities. In this study, we propose a method for adversarial attack detection in GCN. More concretely, we build a model for the attack detection from latent information of each vertex

more »

... ed by the application of autoencoders to hidden layers. The effectiveness of the proposed method was evaluated using real world graph datasets.