Heterogeneous Graph Neural Network
Heterogeneous Graph Neural Networks (HGNNs) extend graph neural network capabilities to data with diverse node and edge types, aiming to learn more nuanced representations from complex relational data. Current research emphasizes efficient HGNN architectures, including those leveraging hierarchical structures, meta-paths, and hybrid aggregation methods, as well as addressing challenges like scalability, explainability, and robustness to adversarial attacks. HGNNs are proving valuable across diverse applications, from network intrusion detection and traffic flow prediction to recommendation systems and biological analysis, offering improved accuracy and interpretability compared to traditional methods.
Papers
SiHGNN: Leveraging Properties of Semantic Graphs for Efficient HGNN Acceleration
Runzhen Xue, Mingyu Yan, Dengke Han, Zhimin Tang, Xiaochun Ye, Dongrui Fan
XG-NID: Dual-Modality Network Intrusion Detection using a Heterogeneous Graph Neural Network and Large Language Model
Yasir Ali Farrukh, Syed Wali, Irfan Khan, Nathaniel D. Bastian