Adversarial Attack and Defense on Graph Data: A Survey

Deep neural networks (DNNs) have been widely applied to various applications, including image classification, text generation, audio recognition, and graph data analysis. However, recent studies shown that DNNs are vulnerable adversarial attacks. Though there several works about attack defense strategies on domains such as images natural language processing, it is still difficult directly transfer the learned knowledge due its representation structure. Given importance of analysis, an increasing number over past few years attempted analyze robustness machine learning models data. Nevertheless, existing research considering behaviors often focuses specific types attacks with certain assumptions. In addition, each work proposes own mathematical formulation, which makes comparison among different methods difficult. Therefore, this review intended provide overall landscape more than 100 papers for data, establish a unified formulation encompassing most models. Moreover, we also compare defenses along their contributions limitations, well summarize evaluation metrics, datasets future trends. We hope survey can help fill gap in literature facilitate further development promising new field.