Non-Sequential Neural Network for Simultaneous, Consistent Classification and Photometric Redshifts of OTELO Galaxies
Authors:
José A. de Diego,
Jakub Nadolny,
Ángel Bongiovanni,
Jordi Cepa,
Maritza A. Lara-López,
Jesús Gallego,
Miguel Cerviño,
Miguel Sánchez-Porta,
J. Ignacio González-Serrano,
Emilio J. Alfaro,
Mirjana Pović,
Ana María Pérez García,
Ricardo Pérez Martínez,
Carmen P. Padilla Torres,
Bernabé Cedrés,
Diego García-Aguilar,
J. Jesús González,
Mauro González-Otero,
Rocío Navarro-Martínez,
Irene Pintos-Castro
Abstract:
Context. Computational techniques are essential for mining large databases produced in modern surveys with value-added products. Aims. This paper presents a machine learning procedure to carry out simultaneously galaxy morphological classification and photometric redshift estimates. Currently, only spectral energy distribution (SED) fitting has been used to obtain these results all at once. Method…
▽ More
Context. Computational techniques are essential for mining large databases produced in modern surveys with value-added products. Aims. This paper presents a machine learning procedure to carry out simultaneously galaxy morphological classification and photometric redshift estimates. Currently, only spectral energy distribution (SED) fitting has been used to obtain these results all at once. Methods. We used the ancillary data gathered in the OTELO catalog and designed a non-sequential neural network that accepts optical and near-infrared photometry as input. The network transfers the results of the morphological classification task to the redshift fitting process to ensure consistency between both procedures. Results. The results successfully recover the morphological classification and the redshifts of the test sample, reducing catastrophic redshift outliers produced by SED fitting and avoiding possible discrepancies between independent classification and redshift estimates. Our technique may be adapted to include galaxy images to improve the classification.
△ Less
Submitted 20 August, 2021;
originally announced August 2021.
