Assurance of Machine Learning Enabled Systems
This project is to develop techniques for the assurance of learning-enabled systems, whose components may be obtained by training with data-driven machine learning techniques. Such systems include many of the next generation autonomous systems, for example self-driving cars. The machine learning components can be e.g., image classifiers based on convolutional neural networks, or control unit based on deep reinforcement learning, etc.
Where such systems are to be deployed in safety critical situations it is important that they have been subject to robust, ideally formal, validation. The object of this project would be to develop techniques that can be used for the assurance of such systems with a view to developing an understanding of best practice in the development of machine learning tools, their use in generating subsystems for integration into autonomous systems, and validation of the resulting systems. These techniques should ideally be useable for the development of safety cases which can be presented to relevant regulatory bodies charged with oversight and certification of autonomous systems.
The work would be motivated by the study of practical systems, in particular the potential use of an intelligent robot arm for nuclear decommissioning which includes an image classification system developed using machine learning. There would also be the possibility to explore the use of image classifiers in automotive vehicles.
Please include the project title and supervisors in your formal application.
Informal enquiries can be made to [Email Address Removed]
This PhD is funded by the School of EEE&CS in support of the EPSRC "Robotics and AI for Nuclear" Hub and the successful student will be expected to liaise with, and attend meetings of, researchers on this and other relevant Hubs.
Jonathan M. Aitken, Affan Shaukat, Elisa Cucco, Louise A. Dennis, Sandor M. Veres, Yang Gao, Michael Fisher, Jeffrey A. Kuo, Thomas Robinson, Paul E. Mort. Autonomous Nuclear Waste Management. IEEE Intelligent Systems, 2017 (in press).
Xiaowei Huang, Marta Kwiatkowska, Sen Wang, Min Wu:
Safety Verification of Deep Neural Networks. CAV (1) 2017: 3-29