return to the ECR seminar schedule

Emulating Clustering statistic templates for RSD parameter estimation

Speaker

Miguel Icaza
KASI (Daejeon, Korea)

Abstract

New cosmological surveys like DESI are producing the most complete maps of the universe up to this date. One of the main scientific goals of these surveys is the measurement of percent accuracy clustering statistics that can be compared with theoretical templates of specific cosmological models. With the increasing resolution of the measured statistics, there is also a need for templates of increasing complexity, which usually translates into increasing the computational cost of each evaluation. However, millions of evaluations of models are needed to explore the parameter space with e.g. MCMC chains. Therefore there is a strong incentive to build efficient evaluation methods. Recently Machine Learning accelerators have been used to emulate these templates, these techniques can reproduce models with high accuracy, and once trained they require virtually no evaluation time. During this talk, I present the methodology that we follow to emulate our Hybrid methodology presented in Song et. all (2021), which can generate templates of the galaxy power spectrum with sub-percent accuracy at k < 0.18h Mpc−1

Date and Time

April 20 2023
4pm KST (= 7am UTC)

Zoom link (active once the seminar starts)

Recording

Link to the recording on YouTube