R. Carvajal, I. Matute, J. Afonso, S. Amarantidis, D. D. Barbosa

Abstract
The study of Active Galactic Nuclei (AGN) is fundamental to comprehend their evolution and connection with star-formation history and galaxy evolution. Powerful radio emission from AGN traces the largest structures in the Universe and, given the radiative and kinetic energy associated with the phenomena, is a prime feedback candidate to understand the correlations between the properties of the Super-massive Black Hole (SMBH) and the host galaxy. The Epoch of Reionisation (EoR) witnessed the birth of the first luminous sources and the initial steps of the SMBH-host connection. But few AGN have been identified in the EoR of which only a small fraction have radio detections. Recent and future large-scale surveys render the use of regular AGN detection and redshift estimation techniques inefficient. On the other hand, Machine Learning (ML) methodologies can help overcome the computational bottleneck to predict the presence of an ever-increasing number of AGN up to the highest redshifts.

We have developed a series of ML models that, using multi-band photometry, select a list of candidates with their predicted redshift. Models were trained and tested on the The Hobby-Eberly Telescope Dark Energy Experiment (HETDEX) and Stripe 82 fields. We find that AGN selection and redshift estimation metrics are similar to traditional techniques but with a fraction of the computational cost. The pipeline recovers 50%–60% of the radio population, 2x - 5x better than chance selection, allowing to shed some light on the origin and duty cycle of radio emission.

Machine Learning for Astrophysics
Filomena Bufano, Simone Riggi, Eva Sciacca, Francesco Schilliro

Springer
2023 October

>> DOI