This is my final year project submission.

This presents the design and implementation of a distributed IoT testbed for split-federated learning. The testbed is built on a lightweight Kubernetes architecture which enables each cluster to represent a distinct device with its own isolated services and communication pipeline. EdgeX Foundry as the IoT framework, manages data flow from simulated edge sensors and facilitates integration with machine learning models for processing and analysis.

The testbed supports split learning, where deep neural networks are divided between edge clusters and a central control cluster, reducing computational overhead on resource-constrained devices. Federated learning is incorporated to aggregate model updates across clusters, enhancing generalisation and reducing over-fitting. The testbed also features load balancing between ranked clusters in order to support mission critical scenarios.

The system supports various configurations, including pure split computing, federated learning, and hybrid approaches, enabling flexible experimentation. This work demonstrates the feasibility of combining split and federated learning in IoT contexts, providing a robust platform for future research into distributed machine learning at the edge and rapid prototyping for model testing.

To begin running project, navigate to FYP/deployment and run bash EdgeClusterCreation