My Honours project was an investigation into how gradient quantization can improve the scalability, efficiency, and security of federated learning in resource-constrained IoT environments. The project evaluates the performance of a shallow 1D Convolutional Neural Network (1D-CNN) trained on non-IID partitions of the CICIDS2017 dataset, simulating real-world intrusion detection tasks across multiple IoT clients. Techniques like stochastic quantization are applied to compress gradient updates, aiming to reduce communication overhead without significantly impacting model accuracy. The system is implemented using TensorFlow and Flower, with metrics such as accuracy, training time, and communication size logged for analysis. The goal is to identify quantization strategies that enable practical, secure federated learning deployments in low-power IoT devices.

This project provided a security assessment of a simulated chemical plant's industrial control system. I aimed to identify vulnerabilities within the ICS and assess the system's resilience to a range of cyberattacks. The evaluation process includes building an asset inventory through network scanning and traffic analysis, identifying and profiling ICS devices, and performing targeted vulnerability assessments and attacks, such as a perception layer compromise and a man-in-the-middle attack.

This project involved installing and securing a CentOS 7 Linux server in a virtualized environment. The goal was to configure a secure and functional server setup with a static IP, hardened SSH access, and firewall rules to prevent unauthorized access. Key steps included installing the OS using VirtualBox, setting up user accounts and directory permissions, configuring a static IP address, securing SSH by disabling root login and changing the default port, and implementing firewall rules using firewall-cmd. The project also involved creating bash scripts and using SELinux tools to ensure secure remote access and user management.