AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-26
Tracing the Cosmic Origins: Machine Learning Reconstruction of the Primordial Density Field from EoR Observations
Reconstructing the Universe's initial conditions from late-time tracers could recover cosmological information hidden by nonlinear structure formation. Using a 3D U-Net on simulated z=8 21-cm and CO(1-0) intensity maps, the joint reconstruction was the most accurate, with a cross-correlation coefficient of about 0.90 or higher for k ≲ 0.75 Mpc⁻¹ in the noiseless case. Gains of about 1.17 bits (1.29 with observational effects) in cosmological information were largest for σ8 and n_s.