Data-driven Galaxy Population Prior for Photometric Redshifts
Time: 14:00–15:00 (Munich time; Europe/Berlin)
Location: European Southern Observatory (ESO), Phoenix (C.2.02).
Paper summary
Future cosmological surveys need precise estimates of galaxy redshift distributions. This paper explores how a generative model can learn the galaxy population prior needed for photometric redshift inference directly from noisy spectra. A probabilistic autoencoder represents spectra in a compact latent space, while a normalising flow models their distribution together with luminosity and redshift. The approach is trained and tested on simulated galaxy populations. Under these controlled conditions, its predictions for mean redshifts in colour-selected bins meet the precision targets of Stage-IV surveys. The results provide a proof of concept for learning accurate population priors with minimal physical assumptions.
Materials
Slides will be added here when available.