Research

My research is organised around four connected themes.

Scientific machine learning

Combining data-driven learning with mechanistic or physical knowledge, particularly for systems in which purely black-box prediction is insufficient. Current interests include physics-informed models, digital twins, graph neural networks, neural operators, inverse modelling and interpretable physical latent states.

Learning and decision-making under uncertainty

Methods for allocating limited observations or actions intelligently. This includes sequential decision-making, bandits, partial monitoring, online change detection and adaptive allocation.

Uncertainty and reliability

Understanding not only what a model predicts, but how strongly the evidence supports that prediction, what information produced it, and when the model is being asked to extrapolate beyond what has been established.

Autonomous science

Research into systems that decide which experiment, simulation or observation should happen next, particularly when scientific computation is expensive and the aim is defensible knowledge rather than predictive performance alone.