Resource Allocation based on Federated Learning for Next Generation Wireless Communication

Abstract

Federated Learning (FL) stands out as a decentralized Machine Learning (ML) method that enables model training using distributed data while safeguarding data privacy. Its application in the next-generation wireless communication, especially the Internet of Things (IoT), realm has the potential to provide more intelligent and efficient solutions for addressing the challenges posed by massive data and security concerns. However, the performance of wireless FL is often hampered by constraints related to wireless communication resources and participant mobility. To address this issue, this thesis proposes a scheduling strategy based on model quality and communication quality. The strategy employs interpretable ML to assess the contribution of each local model to the convergence of the global model and adjusts the weights of model quality and training participant communication quality dynamically, called dynamical balance quality, to achieve a more efficient and fair resource allocation strategy. Finally, the thesis compares the proposed strategy with traditional ones, and simulation results demonstrate that analyzing the value of local models using Interpretable Machine Learning techniques can help maximize the overall learning efficiency of FL systems. Furthermore, this thesis delves into the practical applications of wireless IoT and federated systems. I have designed a distributed federated IoT system framework tailored to the healthcare sector, encompassing a contactless health self-check-in web application, thermal imaging cameras, and physical barriers. This integrated system streamlines the COVID-19 health screening and data recording process. Through collaboration with a major urban hospital in Australia, we implemented a pilot electronic gate solution.