Conditional variational autoencoders for cosmological model discrimination and anomaly detection in cosmic microwave background power spectra

Research summary

A parameter-conditioned variational autoencoder learns compact representations of cosmic microwave background power spectra. The study examines reconstruction, cosmological model discrimination, and anomaly-sensitive diagnostics, including tests beyond the standard cosmological model. Its learned representation connects cosmological parameters with rapid exploratory inference while retaining diagnostics for assessing model behavior.