ABSTRACT Using Graph Pointer Neural Networks with edge computing (GPNN‐EC) to optimize 6G heterogeneous network data migration policies by artificial intelligence is significantly enhanced. This architecture integrates the decentralized intelligence of edge computing with the relational learning power of GPNNs to enable adaptive, efficient data management. The given GPNN architecture consists of multilayer graph attention networks that encode topological dependencies among network nodes and a decision module that uses a pointer to identify the most appropriate migration targets. The nodes in the graph represent network entities (edge devices, base stations, service nodes) that may have specific features such as workload, latency, bandwidth, and node availability, and the edges indicate relationships between nodes. The model is trained using federated learning, in which distributed edge nodes collaboratively update the model without exchanging raw data, thereby maintaining privacy and minimizing communication costs. Each edge node will run a small local instance of GPNN during operation, which autonomously predicts data migration behavior based on the current network state and enables decentralized decision‐making. Using the Wireless Network Slicing Dataset, experimental results indicate that GPNN‐EC reduces energy consumption by 20%, increases network lifetime by 98%, increases routing efficiency by 99%, and increases throughput by 98.43%. These findings verify that GPNN‐EC performs well, balances workload, and is more scalable for real‐time data migration in next‐generation 6G networks.
K et al. (Thu,) studied this question.