My research focuses on the construction of representations of high-dimensional data that respect known symmetries through the appropriate operators, and that capture variability and uncertainty. Symmetries manifest differently across signals, graphs, and time series, and the appropriate representation depends on the problem. The probabilistic modelling of these representations is grounded in my longer-standing interest in inference and learning in latent variable models.

These ideas find application in domains where physical symmetries are known, data is scarce, and interpretability matters as much as predictive accuracy, such as quantum chemistry, astrophysics, and medical imaging. The underlying methods extend more broadly to machine learning, neuromorphic engineering, and signal processing.

Topics:

• Invariant Representations of High-Dimensional Data
• Probabilistic Modelling
• Latent Variable Models and Inference
• Interpretable AI for the Physical Sciences