Cyber-Security and Deep Learning for Autonomous Vehicles

The goal of this PhD is to develop algorithms, software tools and engineering methodologies to make computer vision systems more resilient to cyber-attack. The project specifically targets resilience in visual navigation systems used within autonomous vehicles such as ships and cars. The project will explore the emerging topic of Adversarial Images, exploring attacks on computer vision systems delivered using carefully crafted images and how these could be physically manifested in real world scenes using carefully crafted objects. This will in turn lead to new methods for quantifying the resilience of such systems to attacks of this kind, and ways to minimise that attack risk e.g. through engineering improved neural network architectures, through development of new software tools to test network resilience, or through new engineering methodologies e.g. to ensure unbiased or resilient training sets. The PhD is an iCase in collaboration with Thales.