【论文阅读】Efficient Model-Stealing Attacks Against Inductive Graph Neural Networks（2024）

摘要

Graph Neural Networks（图神经网络） (GNNs) are recognized as（被认为） potent tools（有力工具） for processing（处理） real-world data organized in graph structures（以图结构组织的现实世界数据）. Especially inductive GNNs（归纳GNNs）, which enable the processing of graph-structured data（图结构数据） without（不需要） relying on（依赖） predefined graph structures（预定义的图结构）, are gaining importance in an increasingly wide variety of applications（越来越广泛的应用）. As these networks demonstrate proficiency（熟练） across a range of tasks（一系列任务）, they become lucrative targets（成为有利可图的目标） for model-stealing attacks（模型窃取攻击） where an adversary seeks to（寻求） replicate the functionality of the targeted network（复制目标网络的功能）.
Large efforts（巨大努力） have been made to develop model-stealin