Physically Constrained Machine Learning for Non-linear Spatio-temporal Partial Differential Equations

This project was an EPSRC feasibility study, looking at applying probabilistic machine learning approaches to non-linear partial differential equations, particularly the Cahn-Hilliard model for polymer demixing. The work involved reviewing methods for physics-constrained learning, and developing a Bayesian optimisation framework to infer system parameters from differentiable simulation-based solvers.

Wil O. C. Ward
Research Associate