Leah Parry

Department of Mathematical Sciences Alumna

What do you do?

I have just finished the final year of my integrated master’s degree in Mathematics and Statistics. Throughout the 4 years I’ve spent in Durham, I’ve completed modules in a wide range of topics, but I have found myself specialising in Bayesian Statistics. I’ve also been involved in projects promoting accessibility and diversity within mathematics education. My final year project was an uncertainty quantification internship project – meaning it was completed alongside a non-expert third party.

How are you involved in this area of science? 

My final year project was an internship project within the area of uncertainty quantification. Since it was an internship project, the work was done in collaboration with an external company, DNV. The project was titled “Predicting Damage Done To Subsea Pipelines Using Bayesian Emulation”. We used both theoretical knowledge and coding skills to take known runs of a complex computer simulation (provided to us by DNV) and make predictions about the output of the simulation under different inputs. Crucially, this method allows us to quantify how uncertain we are about the predictions we are making - hence uncertainty quantification.

What do you love about this topic?

To me, uncertainty quantification just makes sense. Predictions are meaningless if we don’t establish how confident we are in those predictions. We should be using all the information at hand to inform not just our estimates but also the uncertainty surrounding these estimates. It has been a welcome challenge to adjust my view of theoretical statistics to one that coincides with the idea of uncertainty quantification.

How does this work deliver real-world impact?

The use of uncertainty quantification improves the robustness of predictive models by providing more detailed information on how confident we are in those predictions. Any future work in uncertainty quantification would further improve our modelling techniques and hence enhance mathematical decision making. Predictive modelling is used extensively across many disciplines, thus, introducing and developing uncertainty quantification techniques could have a far-reaching real-world impact. The real-world impact can already been seen through uncertainty quantification being used within galaxy formation models, climate change models, systems biology models, and more.