Possible topics for postgraduate research

Probability bounding to support decision making under severe uncertainty

To support decision making, modern data science and machine learning methods are able to large data sets for ever more complex models. However, at the same time, despite the abundance of data, severe uncertainty may still be present in certain parts of our models. This can happen for instance when future decisions are made under circumstances about which we have little data or experience, for instance in industries that are affected by long term climate change.

I am particularly interested in situations where experts cannot identify the probabilities themselves, and where they are more comfortable to specify bounds on probabilities instead. When probabilities are bounded rather than exactly specified, for complex models, we need to rely on methods that combine sampling with optimization. Existing methods for propagating probability bounds unfortunately do not scale well to complex systems. A question is thus: how can we improve sampling and optimisation methods to better support complex decision making under severe uncertainty? Besides this theoretical question, a potential student could also investigate the use of such methods in specific applications, including engineering, ecology, and other fields.

Contact: Matthias Troffaes.