Talk Title: Challenges in Learning from Behaviour

Abstract: Artificial Intelligence and Machine Learning are promoted as solutions for cyber security which can adapt to changes in attacker behaviour, learning from data what is normal and what may constitute a threat. Despite much research, there are still fundamental challenges in each stage of the process which prevent us having good benchmarks to understand the robustness of state-of-the-art models. The challenges span: collecting and sharing data safely, ensuring a sufficient variety of data, and then understanding model performance. I will consider these challenges in turn and present some recent research results from my group which address each aspect.

Bio: David Aspinall is Professor of Software Safety and Security at the University of Edinburgh. He is Director of the University's Academic Centre of Excellence in Cyber Security Research. His research interests range from foundations for formal methods in software security to more applied areas including anomaly detection in network security.