Machine Learning for Edge Computing and Task Placement in 5G/6G Architectures

Abstract

The emergence of 5G and the vision of 6G have shifted computation from centralized clouds to the network edge, enabling low-latency services for applications such as autonomous driving, augmented reality, and industrial IoT. However, the dynamic, heterogeneous, and resource-constrained nature of edge environments renders classical scheduling policies and static optimization methods inadequate for joint optimization of latency, energy, and fairness. This paper reviews machine learning approaches for task offloading and placement in multi-access edge computing architectures for 5G/6G networks. The review synthesizes work on optimization, queueing, heuristic, supervised learning, reinforcement learning, and federated learning methods, with attention to scalability, adaptability, fairness, and deployment practicality. It concludes that reinforcement learning schedulers often outperform classical baselines in controlled simulation, but the field still faces a generalization gap between benchmark performance and production deployments. Future work should prioritize standardized benchmarks, shared traces, reproducible code, fairness-aware scheduling, and integration of non-terrestrial networks into 6G edge orchestration.