Matthew graduated from Imperial College London in 2020 with an MSci in Mathematics. While at Imperial he developed an interest in statistics and completed projects on a range of topics, including stochastic modelling of retail systems, MCMC for Bayesian inference in astrophysics, and classifying the qualitative behaviours of spherical orbits around Kerr black holes. In his Master’s thesis, he developed a machine learning model for survival analysis. Outside of Maths, Matthew enjoys playing futsal, watching sport, and reading philosophy.

Research project title:
Statistical learning methods for dimension reduction in multivariate extremes

Supervisor(s):
Christian Rohrbeck, Evangelos Evangelou

Project description:
In many application areas, such as climate science and finance, it is important to assess the risk of rare, extreme events. Mathematically, this involves modelling the tail behaviour of a multivariate random variable using techniques from a field of statistics called multivariate extreme value theory. However, the tail dependence between the components is characterised by a measure that must be estimated using only a small number of observations. Traditional methods for modelling this measure are limited to small or moderate dimensions. Recently, methods from unsupervised learning, such as clustering and principal components analysis, have been adapted to extreme; these methods are better-equipped to handle high-dimensional settings. Matthew’s project aims to advance existing techniques and develop new methodology in this area.