Federated Machine Learning in Edge Computing

Recent years have witnessed the rapid proliferation of mobile devices such as smart phones and wearable devices, which are equipped with various built-in sensors and possess powerful computing and communications capabilities. The large number and diverse advanced functionalities of mobile devices empower ordinary citizens to contribute heterogeneous and cross-space data including machine-sensed data from mobile devices (physical space) and user-generated data from mobile social networking (cyber space), which are aggregated and fused in the cloud are aggregated and fused in the cloud of Cyber-Physical- System (CPS) for knowledge discovery and service. The explosive growing of data traffic in CPS poses imminent challenges while opening up new opportunities to all the aspects of the wireless and mobile system design, such as bandwidth efficiency, computing performance and network capacity. In this project, we will investigate two emerging open problems facing networking system and protocol design: 1) How to design a scalable CPS architecture for efficient handling and robust delivery of ever-growing crowd-sensing data? 2) How to effectively learn from cross-space data to improve the Quality-of-Service (QoS) performance of heterogeneous mobile networking system? To answer these problems, this project will propose novel data-driven networking architectures and protocols for delivering the cross-space data from distributed participants to the backend CPS server over heterogeneous mobile networks with low bandwidth, high mobility, and energy-constrained devices.

In outline, the proposed project intends to

Design an adaptive network resource management framework where the packet routing and content caching decisions are jointly optimized.
Develop a learning-based proactive caching scheme to improve the caching efficiency by learning insights on the characteristics and popularity of contents, the preference and QoS demand of users, and the traffic and link conditions.
Develop and validate a comprehensive analytical model for evaluating the performance of the proposed resource management and caching techniques with multimedia applications and heterogeneous network conditions.
The project seeks to harness the complementary features of information-centric networking and network function virtualisation to fully unlock their potential capabilities and remedy the intrinsic deficiencies of each other. To improve the network performance, a machine learning based proactive caching strategy will be proposed to use the insights on the characteristics and popularity of contents, the preference and QoS demand, and the traffic and link conditions. In order to investigate the QoS of the proposed schemes, an analytical model will be developed using Stochastic process/Markov Chain etc., and validated by computer simulation/emulation with synthetic/real-world data.