Accelerating cosmological modelling with symbolic regression
Date:
Upcoming cosmological surveys will map the Universe with unprecedented precision, demanding theoretical tools that are not only accurate enough to match the data, but also flexible enough to enable discovery of new fundamental physics. Machine learning has shown remarkable success in accelerating cosmological inference, but many approaches trade transparency for predictive power. In this talk, I will discuss how symbolic regression - a machine learning technique that discovers analytic expressions directly from data - offers a complementary path. Unlike black-box models, symbolic regression produces human-readable equations that can both emulate costly simulations and reveal functional relationships hidden in high-dimensional parameter spaces. I will present recent applications to the matter power spectrum, where symbolic regression enables accurate and efficient emulation across wide cosmological parameter ranges, including extensions with neutrino mass and evolving dark energy. Beyond emulation, I will highlight how this framework may provide new ways to confront theory with data, suggesting a route toward interpretable machine-assisted discovery in cosmology.
- Given at:
- Oxford Dalitz Seminar in Fundamental Physics, University of Oxford (December 2025)
- University of Hull (Oct. 2025)
- The Royal Society Discussion Meeting on Symbolic Regression in the Physical Sciences, London (April 2025)
- The Johns Hopkins University, Baltimore (March 2025)
