Speaker
Description
Upcoming cosmological surveys, such as the Roman Space Telescope, will deliver unprecedented data for studies of the large-scale structure of the Universe. These observations will shed new light on cosmic evolution and the nature of its fundamental components, including dark matter and dark energy. Fully exploiting these datasets requires theoretical predictions that are both accurate and computationally efficient across an extended cosmological parameter space.
To address this challenge, we constructed the Goku simulation suite—the first N-body simulation suite spanning 10 cosmological parameters, including the five standard ΛCDM parameters and extensions that account for dynamical dark energy, massive neutrinos, the effective number of neutrino species, and the running of the primordial spectral index. Based on these simulations, we trained GokuNEmu, a neural-network emulator for the nonlinear matter power spectrum, using advanced multifidelity machine learning techniques.
GokuNEmu provides fast and accurate predictions for next-generation cosmological analyses. We are currently applying the emulator to weak-lensing and galaxy-clustering data from the Dark Energy Survey (DES). In parallel, we are extending the framework to additional summary statistics, including the halo mass function and non-Gaussian weak-lensing statistics, with the goal of further improving constraining power.