Speaker
Description
Modern cosmology often requires repeated evaluations of expensive theoretical predictions, from inference and model comparison to forecasts and exploratory studies beyond ΛCDM. Symbolic regression offers a fast and accurate alternative for building emulators, competitive with traditional neural-network approaches while retaining compact analytical expressions that are transparent, interpretable, and easy to integrate.
We present CMBolic, a suite of symbolic emulators for CMB lensing, temperature, and polarization anisotropies in ΛCDM extensions including massive neutrinos and the now-standard CPL parametrization of evolving dark energy. These emulators reach the accuracy required for Stage-IV cosmological analyses while remaining exceptionally lightweight and easy to implement. We further show how symbolic regression can be applied to total matter power spectra in the Generalized Dark Matter framework, enabling fast and scalable constraints on broad classes of non-standard dark matter models.