Cross-silo implementation of federated learning in healthcare for 6G: Leveraging VPN-based wireless backhaul networks / Atif Mahmood

Atif , Mahmood (2024) Cross-silo implementation of federated learning in healthcare for 6G: Leveraging VPN-based wireless backhaul networks / Atif Mahmood. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (222Kb)
	PDF (Thesis PhD) Restricted to Repository staff only until 31 December 2025. Download (3174Kb)

Abstract

The widespread use of smart devices and the Internet of Everything (IoE) strains current wireless systems. Sixth-generation (6G) research is underway, with Artificial Intelligence (AI) playing a picotal. Terahertz (THz) communication spectrum and Federated Learning (FL) are gaining traction in 6G as advanced AI. FL, a decentralized approach, faces challenges in resource management, statistical and system heterogeneity, and security. Communication bottlenecks occur during FL, and efficient resource management is crucial for faster global convergence. Recent research explores methods like local updates, compression, and decentralized training. Privacy concerns arise, but sharing model updates may expose sensitive information. Solutions address challenges like stragglers and fault tolerance, which are critical for 6G. The Terahertz spectrum, vital for 6G, offers wide bandwidth and potential commercial use, especially in high-speed data and video transmission. Despite limitations, Terahertz technology maximizes spectrum utilization and enhances transmission security. The study addresses the challenges of federated learning (FL) in three phases. Firstly, it provides a concise overview of the THz spectrum in fixed wireless communication, highlighting its suitability for fixed wireless links within 1 km. The study examines THz properties, estimates data rates, and suggests its use to enhance Federated Learning (FL) communication. In the healthcare sector, FL relies on THz-based wireless backhaul with a VPN for hospitals, laying the groundwork for THz utilization in 6G wireless backhaul. This introduces an innovative network architecture for secure cross-silo FL, focusing on healthcare enhancement. In the second phase, data from a new HFMD biosensor is classified using centralized machine learning to analyze the HFMD dataset and create a benchmark. Initial experiments with Support Vector Machine (SVM) yielded a 72% accuracy rate, and a neural network achieved 80%. Federated learning with four clients reached a maximum accuracy of 91%, addressing healthcare data security concerns. The third phase focuses on improving dataset convergence and distribution, introducing an efficient network architecture for Federated Learning in the subsequent section on the wireless backhaul-based federated network for healthcare dataset experiments. Here, an efficient network for healthcare records analysis is introduced. The framework is built upon VPN technology, and a cross-silo-based federated learning approach is applied over the wireless backhaul network. A comparison is made between the Terahertz-based wireless backhaul channel bandwidth, the E/V band (mmWave), and the microwave. A significant enhancement in convergence time is observed when utilizing the higher bandwidth wireless channel. In summary, this research indicates that optimal accuracy is attained when utilizing FedAvg and THz as the communication channel, with convergence times ranging from 55 seconds to 24 seconds for FedAvg and transitioning from THz to downgraded MW. This highlights the pivotal significance of a higher bandwidth communication link. Relying solely on the dataset for determining device rules is not advisable. A three-step strategy is employed to bolster security, encompassing a private network within the telecom network, a private network with licensed frequency channels, and a private network with a licensed band. This security is further strengthened through VPN-based measures.

Actions (For repository staff only : Login required)