Optimal, fast, and robust inference of reionization-era cosmology with the 21cmPIE-INN

TL;DR

This paper introduces a novel machine learning framework combining a convolutional summary network and a conditional invertible network, enabling fast, accurate, and unbiased inference of cosmological parameters from 21cm observations of the reionization era, offering a promising alternative to traditional power spectrum analysis.

Contribution

The paper presents a physics-inspired, simulation-based inference method that efficiently estimates posterior distributions of astrophysical and cosmological parameters from 21cm data.

Findings

Fast and unbiased parameter inference achieved

Sensitivity to non-Gaussian information demonstrated

Potential to outperform traditional power spectrum methods

Abstract

Modern machine learning will allow for simulation-based inference from reionization-era 21cm observations at the Square Kilometre Array. Our framework combines a convolutional summary network and a conditional invertible network through a physics-inspired latent representation. It allows for an efficient and extremely fast determination of the posteriors of astrophysical and cosmological parameters, jointly with well-calibrated and on average unbiased summaries. The sensitivity to non-Gaussian information makes our method a promising alternative to the established power spectra.