AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-23
Optimal cosmological cross-correlation analysis with U-Net-based processing
Standard cosmological cross-correlation analyses use two-point statistics, which miss non-Gaussian information in late-time cosmic fields. The authors trained a 1D U-Net on paired simulated fields to learn a transformed tracer that correlates maximally with a target field, testing Gaussian and log-normal cases. For Gaussian fields the network matched the linear solution, while for log-normal fields the mean Pearson correlation rose from 0.857 for the linear estimator to 0.953, indicating use of information beyond the power spectrum.