ABSTRACT
A key problem in cosmology is the estimation of the redshifts of galaxies using photometric data. For instance, the Sloan Digital Sky Survey (SDSS) has already collected photometric data of about one billion objects, and it is necessary to estimate their redshifts. Typically, this is done by using supervised learning methods. In this work we show that existing redshift prediction methods can be combined in order to obtain more accurate predictions. We tackle this problem under two perspectives: (i) estimation of the regression function of the redshift y on the photometric x, , and (ii) estimation of the conditional density .We apply the proposed techniques to data from the Sloan Digital Sky Survey, and show that the combined predictions are indeed more accurate.
Keywords:
machine learning; stacking; conditional densities; cosmology