Xiyue Fan MINDS CDT, 2023

Doctoral Researcher, University of Southampton

Xiyue Fan is a doctoral researcher using transformer models to tackle differential equations: the branch of mathematics that describes how quantities change over time. These equations appear throughout physics, engineering, biology and climate science.

Most current approaches are data-driven, such as Physics-Informed Neural Networks (PINNs) and Neural Operators (NOs), or use a mix of numerical and symbolic data through methods like symbolic regression. These approaches either lack interpretability and sufficient generalisation, or remain sensitive to noise and face challenges in efficient search.

Xiyue's research trains a transformer model from scratch on purely symbolic data, exploring deep representation techniques to learn better representations of symbolic expressions and produce more accurate approximations of these problems.

She is a contributing author on published work examining the temporal dynamics of TiOx memristors for reservoir computing applications. She is due to graduate in 2027.