The Evaluation of a Machine Learning Emulator of Cosmological Boltzmann Codes

Abstract

In this work, we seek to investigate the accuracy of the machine learning emulator Cosmopower in the reconstruction of temperature power spectra using the parameters of the Early Dark Energy model of the universe. Cosmopower claims vast improvements in speed in comparison to traditional Boltzmann codes, while retaining ease of use and accuracy. We find in our investigation that Cosmopower is able to reconstruct the TT power spectra of various Early Dark Energy models of the universe with accuracy of approximately 0.02σ.