Overview of Graph Neural Networks Application on State-of-the-Art Cyber Security Network Threat Detection Techniques

Cyber security threats are becoming increasingly complex, demanding advanced solutions for effective anomaly and threat detection. Machine learning (ML) and deep learning techniques have proven to be excellent tools for these tasks. In the field of cybersecurity malicious activities have increased in sophistication and scale, requiring existing tools to constantly adapt. The data that is used for deep learning tasks contains rich information on relationships between elements which naturally conforms to graph data representation. For several years Graph Neural Networks (GNN) have gained traction demonstrating excellent performance in anomaly detection on many deep learning tasks due to their ability to learn from such data. This paper reviews current state-of-the-art research utilizing graph neural networks (GNNs) and their different architectures used in the domain of cyber and information security. This paper shows how they offer robust capabilities in analyzing network data for identification of anomalous traffic, their operational components, discuss their strengths and weaknesses, common datasets and implementations, and their performance. Finally, three open challenges are outlined for further research opportunities.