Graph Attention Networks and Transformers in Edge Computing: A Survey of Applications, Mechanisms, and Integrations

Graph Attention Networks (GAT) and Transformer-based architectures have become cornerstone technologies for modeling irregular spatial relationships and long-range temporal dependencies in resource-constrained edge environments. This survey systematically reviews the evolution, core mechanisms, edge-specific adaptations, and integration strategies of GAT and Transformer models. Through a comprehensive analysis of ۳۸ recent studies, the paper classifies applications across anomaly detection, resource allocation, QoE prediction, and intelligent transportation while highlighting performance gains, computational trade-offs, and scalability challenges. Special emphasis is placed on emerging hybrid GAT-Transformer frameworks that simultaneously capture dynamic network topology and temporal evolution, offering a promising pathway toward real-time edge intelligence in ۶G, MEC, and beyond.

کلیدواژه ها:

Graph Attention Networks ، Transformers ، Edge Computing ، QoE prediction ، GAT-Transformer frameworks