🤖 AI Summary
To address the limited accuracy and poorly quantified uncertainties in photometric redshift (photo-z) estimation for wide-field imaging surveys, this work introduces conditional generative adversarial networks (CGANs) to photo-z estimation for the first time, shifting the paradigm from deterministic point estimates to full posterior probability density function (PDF) modeling. The proposed method takes multi-band photometric measurements as input and leverages CGANs to directly generate well-calibrated redshift posterior distributions under explicit conditioning. Experiments on the Dark Energy Survey (DES) Year 1 dataset demonstrate that our approach significantly improves PDF calibration compared to conventional methods such as random forests. It further exhibits superior robustness to outlier objects and more accurate uncertainty quantification. By delivering statistically well-calibrated, probabilistic distance inference, the framework provides a reliable foundation for large-scale cosmological surveys.
📝 Abstract
Accurate and reliable photometric redshifts determination is one of the key aspects for wide-field photometric surveys. Determination of photometric redshift for galaxies, has been traditionally solved by use of machine-learning and artificial intelligence techniques trained on a calibration sample of galaxies, where both photometry and spectrometry are determined. On this paper, we present a new algorithmic approach for determining photometric redshifts of galaxies using Conditional Generative Adversarial Networks (CGANs). Proposed CGAN implementation, approaches photometric redshift determination as a probabilistic regression, where instead of determining a single value for the estimated redshift of the galaxy, a full probability density is computed. The methodology proposed, is tested with data from Dark Energy Survey (DES) Y1 data and compared with other existing algorithm such as a Random Forest regressor.