Multi-scale simulation with physics-informed neural networks
Physics-informed neural networks have emerged as a promising tool for solving differential equations, and they have been applied to many scientific problems. However, they often struggle when solving equations that describe realistic, highly multi-scale physical systems, such the propagation of earthquakes, or the formation of galaxies. We are developing state-of-the-art methods to help PINNs scale in this regime.
Scientific machine learning-enhanced planetary exploration
Scientists and engineers planning planetary exploration missions and researching the history of the solar system heavily rely on high-resolution satellite imaging. However, the cameras capturing these images operate in extreme environments, and challenges like extreme noise and instrument malfunctions hinder the acquisition of clear images. We are developing scientific machine learning-based image processing tools to improve image quality and allow researchers to more effectively carry out tasks like traverse planning and surface characterisation.
Extending quantum theories with AI
Quantum mechanics is hugely successful at predicting the outcomes of quantum experiments. Yet, it cannot tell us the certain outcome of an experiment, only its probability. We are taking a fresh look at alternative deterministic quantum theories, in particular local hidden variable models, which attempt to explain this inherent randomness, and using SciML to provide key insights into these theories. Such theories could shed new light on the nature of quantum systems.
AI and scientific machine learning for model discovery
Building mathematical models that accurately describe physical systems is essential for science, but the scientific method – iterating between deriving a model and experimentally validating it – can be painstakingly slow. We are investigating whether AI and scientific machine learning can automate this process and learn mathematical models from observed experimental data, to accelerate scientific discovery.
See our lab website for more information on our research.