Statistical machine learning with application to sports modelling

The project involves the development and testing of contemporary machine learning methods with application to sports data, ranging from simple match summary statistics to high-frequency high-dimensional tracking data.

The student will work at the interfaces of statistics, machine leaning and decision theory - producing novel methodology in response to new data sources and technologies.

In initial phases of the work the student will use the 'transfer learning' framework to quantify dissimilarities between data sets, models and predictive distributions.

The sizes of the dissimilarities will inform the degree to which models can be repurposed for tasks they were not originally trained for.