Intelligent Edge Caching based on Federated Deep Learning
Date: 7 June 2021
University of Exeter
Doctor of Philosophy in Computer Science
Caching contents at the edge of network is considered to be a cost-effective solution to cope with ongoing traffic growth and address the backhaul bottleneck problem in wireless networks. However, the inherent characteristics of wireless networks, including the high mobility of users and restricted storage capability of edge nodes, ...
Caching contents at the edge of network is considered to be a cost-effective solution to cope with ongoing traffic growth and address the backhaul bottleneck problem in wireless networks. However, the inherent characteristics of wireless networks, including the high mobility of users and restricted storage capability of edge nodes, cause many difficulties in the design of caching schemes. Driven by the recent advancements in Machine Learning (ML), learning-based proactive caching schemes are able to accurately predict content popularity and improve cache efficiency, but they need gather and analyse users’ content retrieval history and personal data, leading to privacy concerns. To address these challenges, this research mainly focuses on the design of learning-based caching schemes to improve caching efficiency and protect user privacy in various modern networks, such as Fifth Generation Mobile Networks (5G), Internet-of-Vehicles (IoV), and Fog Radio Access Networks (F- RANs). In modern networks, mobile phones, wearable devices, and autonomous vehicles provide growing computational power and storage capability. Coupled with the increasing concern about data privacy protection, the emerging framework of federated learning has been recognised as a promising framework to efficiently build ML models while protecting user privacy by keeping data at local devices and fitting ML techniques into the network edges. In 5G, a communication Efficient Federated learning based Proactive content Caching scheme (EFPC) is proposed to mitigate the privacy risks and reduce communication consumption. Based upon the federated learning framework, each user locally trains a shared model for content popularity prediction by using their own data, and only uploads the parameters of the model to the edge server for aggregation. To further reduce communication costs, the 3LC data compression scheme is used in EFPC to compress the upload parameters of the model. In F-RANs, a Federated Learning based Cooperative Hierarchical Caching scheme (FLCH) is designed to maximise the utilisation of available caches with edge node. FLCH exploits horizontal cooperation between neighbour F-APs and vertical cooperation between the baseband unit pool and fog access points to cache contents with different degrees of popularity. In IoV, a Mobility-aware Proactive edge Caching scheme based on Federated learning (MPCF) is developed to support mobility of vehicles. This new scheme enables multiple vehicles to collaboratively learn a global model for predicting content popularity with the private training data distributed on local vehicles. MPCF also integrates a mobility-aware cache replacement policy, which allows the network edges to add/evict contents in response to the real-time mobility patterns and dynamic preferences of vehicles. To ease reliance on the fixed central server, eliminate the issue of hand-over between RSUs, a peer-to-peer federated deep learning based proactive caching scheme (PPFC) is proposed. A vehicle rather than a fixed edge node, acts as a central server to aggregate ML models from nearby vehicles. A dual-weighted model aggregation scheme is designed to reduce the effect of straggler vehicles and further improve the global model accuracy. The proposed caching schemes in this thesis can greatly improve cache performance, effectively protect users’ privacy and significantly reduce communication costs. The simulation experiments are conducted to evaluate the performance of these caching schemes and the accuracy of the designed prediction models using real-world datasets.
Item views 0
Full item downloads 0