Neural Architecture Search for Defence and Security

Objectives

1. Create a meta-dataset of defence and security tasks requiring specialist architectures
2. Evaluate and enhance evolutionary search algorithms in expressive search spaces on the meta-dataset
3. Assess the operational robustness and scalability of the newly optimised architectures in simulated scenarios

Description

This PhD project aims to refine search methods within einspace, an innovative, expressive search space developed for neural architecture search (NAS), with a focus on defence and security applications. These sectors present unique challenges, such as dealing with non-standard, heterogeneous data, necessitating specialised architectures that are robust to environmental changes. Initially, the research will explore evolutionary search strategies, with an emphasis on developing novel mutation strategies and selective pressure mechanisms to enhance the adaptability and performance of these architectures under unconventional conditions. Dr Crowley and Dr Gouk, who are experts in automated machine learning—including NAS—will provide ideal supervision for this project, leveraging their extensive knowledge to guide the development of advanced, application-specific neural networks.

Research Theme

• Sensor Signal Processing

Principal Supervisor

• Dr Elliot J. Crowley
• University of Edinburgh, School of Engineering
• elliot.j.crowley@ed.ac.uk

Assistant Supervisor

• Dr Henry Gouk
• University of Edinburgh, School of Informatics
• henry.gouk@ed.ac.uk