Wireless Edge Machine Learning in 5G/6G Networks

With the advent of beyond 5G and 6G systems, wireless communication networks will evolve from a pure communication perspective to service enablers in different vertical sectors, incorporating machine learning (ML) mechanisms to build an effective complex system able to learn and dynamically adapt to the evolving network landscape. This book chapter describes some recent solutions for wireless edge machine learning, i.e., a novel class of distributed and reliable ML services that can be accessed by end-users via wireless communications. Differently from cloud-based ML, the edge machine learning process requires not only high learning accuracy and reliability, but also a very low-latency for autonomous decision making, while coping with communication and on-device resource constraints. In this context, considering both stand-alone and cooperative/federated learning scenarios, the goal of this chapter is to introduce a dynamic resource allocation framework to enable wireless edge machine learning, jointly optimizing radio parameters (e.g., set of transmitting devices, transmit powers, rates, etc.) and computation resources (e.g., CPU cycles at devices and at edge server) in order to strike the best trade-off between energy, latency, and performance of the inference and/or training task. The framework hinges on stochastic Lyapunov optimization, which enables adaptation to time-varying network conditions, without the need of a priori knowledge of the statistics of random context parameters (e.g., radio channels, data arrivals, etc.). Numerical results on both synthetic and real data assess the performance of the proposed resource allocation framework enabling machine learning at the wireless network edge