WP4 Network Edge Automation
Exploring effective network edge solutions supporting high-demand services in large numbers of locations
Lead: Prof. Toktam Mahmoodi, King’s College London
Overview
Within the context of edge computing becoming a key topic for a number of vertical applications in the era of emerging 6G networks, Work Package 4 (WP4) explores how to push the automation of the edge even further by leveraging the power of Artificial Intelligence (AI).
WP4 focuses on four main areas to deliver on this:
Monitoring and Tracking performance at (and with the) Edge
Fundamental to our objectives is the implementation of AI-based methods for flexible and modular edge design to achieve specific levels of automation in distributed edge systems. With the primary aim to improve network efficiency as well as effectively manage computational and network resources, we have designed and developed a Monitoring App (mApp) to track and collect a broad spectrum of the systems performance.
Offloading tasks to the edge
One of the traditional use of computational resources at the edge is to offload extra and untenable compute, e.g. large AI models, from devices to the edge. WP4 takes a novel spin on this and explores task-oriented, and semantic-aware offloading, with the aim of minimizing the data exchange required with the edge, and additionally thinning the communication overheard between device and the edge.
Joint communication and computing considerations
WP4 designs a joint communication and compute framework for the edge. The solutions to achieve this include Split computing (SC) and federated learning (FL) methods to optimize data processing and management for the constrained device. The focus is to reduce latency through SC at inference phase and deploy FL systems in the training phase. Together, these methods address the dynamic nature of edge networks and ensure resource utilization efficiency with a focus on scalability, flexibility and privacy-preservation.
AI at the Edge in Industry 5.0
The WP4 solutions are being tested in Industry automation setting, referred to as Industrial Metaverse, or Factory of the future. The deployed scenario explores the use of drones for warehouse monitoring, with deploying split computing, and semantic-aware communication through video question answering.
Note: Refer to WP1 use case description for further detail on Industrial Metaverse (UC1), and Factory of the future (UC5).
Partners
WP4 is led by King’s College London and contributions are made by:
- University of Bristol
- University of Cambridge
- Samsung
- Nokia Bell Labs
- Digital Catapult
- BT
- BBC R&D
- Weaver Labs