Description
Emulator and Field Level Inference
In this talk I will introduce Kun, a recently finished high resolution simulation suite for precision cosmology and Chinese Space Station Survey Telescope cosmology science. The Kun suite consists of 129 simulations covering 8 dimensional cosmological parameter space, including dynamic dark energy and massive neutrinos. The CSST Emulator constructed on Kun suite has competitive performance in...
The nonlinear matter power spectrum is the basic statistic in cosmological analysis. However, the theoretical prediction with percent-level accuracy is only from the expansive numerical simulation, which is feasible in the likelihood analysis. The solution is to construct an emulator using a finite number of simulations in a given parameter space. However, most emulators cannot cover the...
Persistent homology is a powerful tool from the field of topological data analysis, which has shown promise as a novel statistic for cosmological parameter estimation. Compared with traditional two-point statistics, topological measurement presents information on a wide variety of scales and demonstrates a higher sensitivity to distinguish neutrino mass. We build a FLAMINGO-based topology...
Tentatively, the first probabilistic and differentiable forward model for galaxies at the field level.
Many inverse problems are only implicitly defined through a forward simulator.
In such cases, no closed-form likelihood function is known.
In cosmology, the canonical example is parameter inference from higher-order and field-level statistics.
By training neural networks on samples generated by the simulator, one can obtain an approximate likelihood and thus perform parameter inference.
I...
