Abstract
We have been developing a Bayesian parameter estimator which is very competitive compared with other machine learning methods, as evidenced by several experiments performed by our group (e.g., on photometric redshifts and galaxy spectral synthesis). Our approach relies on a training set, i.e., a (empirical, theoretical or mixed) data set with known parameters, and outputs the probability distribution function of a certain parameter, as well as other statistical summaries of this distribution, for all galaxies in the survey. We propose to build a large training set using theoretical libraries and use them to derive galaxy parameters from S-PLUS, J-PLUS and J-PAS observations.
