Class-based attack on graph convolution network

• Main Author: He, HeFei
• Other Authors: Anupam Chattopadhyay
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/156472
• Institution: Nanyang Technological University

Prevalent use of graph structure data for classification tasks has brought attention to the robustness of graph convolutional networks. Recent study has been shown that graph convolutional networks are vulnerable to adversary attacks, causing a severe threat to real world application. In this report, we have conducted two types of untargeted attack on the graph convolutional network by injecting the fake node. The perturbation of fake nodes was based on the corresponding feature of the class and connected to different classes(dConnClass) or the same class(sConnClass), aiming to minimise the classification accuracy of the graph convolutional network.