Identifying Hα Emitters in GAIA Spectra Using Neural Networks

Observation of Hα emission line is a strong indicator of sources withintense ultraviolet radiation, capable of ionizing hydrogen (at an energygreater than 13.6 eV). These include numerous active astrophysicalsources and environments, such as planetary nebulae, symbiotic starsyoung stellar objects, cataclysmic variables, star-forming galaxies andAGNs, among others. Identifying such emitters in all-sky surveysalways provides a useful value-added catalogue and helps us to enrichour knowledge on the nature of the emitters. In this work, we aimto classify Hα-emitters in GAIA spectra. Particularly, we focus onthe 1.2 million variable sources, identified in the GAIA Data Release3 (DR3) catalogue. Our approach leverages modern, state-of-the-artconvolutional neural networks (CNNs), as well as fully connectedarchitectures. We also explore the potential of transformer-basedmodels to capture complex temporal patterns in the light curves.Thousands of GAIA spectra have been used in order to train our neuralnetworks. Preliminary results demonstrate promising performance indistinguishing Hα emitters from non-emitters, providing a pathwaytoward efficient, large-scale identification of emission-line objects inGAIA.