Harry graduated with an MMath from the University of Bath in 2024, with an masters thesis entitled “The Deep Ritz Method for Solving PDEs in the Variational Form Using Neural Networks”. His mathematical interests include Scientific Computing, Mathematical Biology, Machine Learning, Juggling, and Numerical Analysis in general.

Outside of maths, he enjoys archery, cycling, origami, and learning German.

Project title:
Numerical Analysis of Neural Stochastic Differential Equations 

Supervisor(s):
James Foster, Eike Muller

Project description:
Neural Stochastic Differential Equations (Neural SDEs) are a highly prominent model class in machine learning, underpinning applications ranging from weather forecasting to generative modelling. Neural SDEs are defined by declaring drift and/or diffusion terms of an SDE as neural networks, generalising Neural ODEs in accounting for random processes. Through training, such a model can conform to complex, noisy time-series data representing a random process with signatures we want to quantify. 

 This project aims to investigate several avenues in the alternative construction and implementation of Neural SDEs. This includes considering a manufactured driving processes for higher-order convergence – implemented on the Deep BSDE model for solving PDEs – as well as learning the noise distribution from raw data.