Non-Sequential Neural Network for Simultaneous, Consistent Classification and Photometric Redshifts of OTELO Galaxies

TL;DR

This paper introduces a non-sequential neural network that simultaneously classifies galaxy morphology and estimates photometric redshifts, improving consistency and reducing errors compared to traditional SED fitting methods.

Contribution

The novel non-sequential neural network architecture enables concurrent galaxy classification and redshift estimation, ensuring consistency and reducing catastrophic errors.

Findings

Successfully recovers galaxy morphology and redshifts

Reduces catastrophic redshift outliers

Ensures consistency between classification and redshift estimates

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. 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.