Graph Robustness Benchmark: Benchmarking the Adversarial Robustness of Graph Machine Learning

Adversarial attacks on graphs have posed a major threat to the robustness of\ngraph machine learning (GML) models. Naturally, there is an ever-escalating\narms race between attackers and defenders. However, the strategies behind both\nsides are often not fairly compared under the same and realistic conditions. To\nbridge this gap, we present the Graph Robustness Benchmark (GRB) with the goal\nof providing a scalable, unified, modular, and reproducible evaluation for the\nadversarial robustness of GML models. GRB standardizes the process of attacks\nand defenses by 1) developing scalable and diverse datasets, 2) modularizing\nthe attack and defense implementations, and 3) unifying the evaluation protocol\nin refined scenarios. By leveraging the GRB pipeline, the end-users can focus\non the development of robust GML models with automated data processing and\nexperimental evaluations. To support open and reproducible research on graph\nadversarial learning, GRB also hosts public leaderboards across different\nscenarios. As a starting point, we conduct extensive experiments to benchmark\nbaseline techniques. GRB is open-source and welcomes contributions from the\ncommunity. Datasets, codes, leaderboards are available at\nhttps://cogdl.ai/grb/home.\n