A Systematic Review of Contemporary Applications of Privacy-Aware Graph Neural Networks in Smart Cities
Jingyan Zhang, Irina Tal · 2024
In smart cities, graph embedding technologies, Graph Neural Networks (GNNs), and related variants are extensively employed to address predictive tasks within complex urban networks, such as traffic management, the Internet of Things (IoT), and public safety. These implementations frequently require processing substantial personal information and topological details in graph formats, thereby raising significant privacy concerns. Mitigating these concerns necessitates an in-depth analysis of existing privacy preservation techniques integrated with GNNs in the specific context of smart cities. To this end, this paper provides a comprehensive systematic review of current applications of privacy-aware GNNs in smart cities. Our research commenced with a methodical literature search that identified 14 pertinent papers and summarized prevalent privacy preservation mechanisms, including federated learning, differential privacy, homomorphic encryption, adversarial learning, and user-trust-based approaches. Subsequent analysis examined how the integration of these technologies with GNNs enhances privacy security and model utility in smart city applications. Further, we proposed an analytical framework for privacy-aware GNNs across the machine learning lifecycle, assessing the challenges of current integration from a practical viewpoint. The paper concluded by suggesting potential directions for future research.