Embedded Files
Machine learning (ML) techniques have emerged as powerful tools for emulating the 21-cm signal and circumventing the time-consuming nature of traditional simulators. The importance of machine learning in this context lies in its ability to significantly accelerate the parameter inference process. By training a machine learning model on a large set of pre-computed simulations, it becomes possible to generate fast and accurate emulators that can predict the 21-cm signal for a given set of cosmological and astrophysical parameters. Moreover, machine learning emulators can be trained on a diverse range of simulations, enabling the exploration of different cosmological scenarios and astrophysical processes.
Machine learning (ML) techniques have emerged as powerful tools for emulating the 21-cm signal and circumventing the time-consuming nature of traditional simulators. The importance of machine learning in this context lies in its ability to significantly accelerate the parameter inference process. By training a machine learning model on a large set of pre-computed simulations, it becomes possible to generate fast and accurate emulators that can predict the 21-cm signal for a given set of cosmological and astrophysical parameters. Moreover, machine learning emulators can be trained on a diverse range of simulations, enabling the exploration of different cosmological scenarios and astrophysical processes.
We have developed an ANN-based emulator of the 21-cm signal which was trained on data produced by the 21cmFAST semi-numerical code. We used this emulator in an MCMC pipeline to reproduce the parameter constraints based on the HERA phase-I upper limits on the 21-cm power spectrum.
We have developed an ANN-based emulator of the 21-cm signal which was trained on data produced by the 21cmFAST semi-numerical code. We used this emulator in an MCMC pipeline to reproduce the parameter constraints based on the HERA phase-I upper limits on the 21-cm power spectrum.
This figure illustrates the marginalized posterior distribution of the X-ray luminosity of high-redshift sources per unit Star Formation Rate (SFR). It compares results from various scenarios: using only Atomic Cooling Galaxies (ACGs), including both ACGs and Molecular Cooling Galaxies (MCGs), and analyzing datasets with and without constraints from HERA observations. This illustrates that the high preference for values greater than 10^40 goes away as soon as we introduce MCGs. [2307.15577]
This figure illustrates the marginalized posterior distribution of the X-ray luminosity of high-redshift sources per unit Star Formation Rate (SFR). It compares results from various scenarios: using only Atomic Cooling Galaxies (ACGs), including both ACGs and Molecular Cooling Galaxies (MCGs), and analyzing datasets with and without constraints from HERA observations. This illustrates that the high preference for values greater than 10^40 goes away as soon as we introduce MCGs. [2307.15577]
This shows the posterior distribution of astrophysical parameters with and without HERA data, including MCGs. [2307.15577]
This shows the posterior distribution of astrophysical parameters with and without HERA data, including MCGs. [2307.15577]
Page updated
Google Sites
Report abuse