Hybrid AI-driven Approach to Context-Aware Inter-Slice Load Balancing for Cloud-Native Functions in 5G Networks
Alireza Yaghoobi · 2025
This thesis explores transformative technologies in academia and industry, focusing on cloud-native or container-based network functions (CNFs) that improve scalability and lifecycle management compared to virtual network functions (VNFs). CNFs simplify orchestration by leveraging reliable service meshes and eliminating proxies. The thesis emphasizes high-level connectivity and APIs within custom execution environments (EEs), enabling AI-agent interactions for intelligent capabilities. Graph neural networks (GNN) and spatio-temporal multi-head graph attention networks (SP-mGAT) are utilized to generate context-aware embeddings, clustering clients by traffic characteristics into priority labels. These labels feed multi-agent deep reinforcement learning (MADRL) models like deep Q-networks (DQN) and proximal policy optimization (PPO) for traffic load balancing across core network slices, enhancing service quality. Kubernetes-integrated deployments use Prometheus for real-time predictive analytics, expanding beyond network-centric metrics to system-centric metrics. The hybrid platform achieves over 50% error reduction compared to earlier AI-assisted systems, demonstrating superior performance for intelligent, real-time decision-making in network environments.