Unveiling the lowest surface brightness regions in dwarf galaxies: a machine learning approach with Gaia early DR3

Recent observations suggest that dwarf galaxies appear to host extended tidal structures or debris from disrupted systems that contribute to their stellar halos. These features are extremely challenging to identify, yet they are relics of past interactions and provide crucial constraints on the nature of dark matter and the evolutionary history of galaxies. In this thesis, I investigated different methods to calculate membership probabilities for individual stars in dwarf galaxies of the Local Group, with a particular focus on identifying members in external structures such as tidal tails and outer halos. The analysis was performed on four mock catalogs constructed from Gaia eDR3, designed to reproduce different foreground conditions and dwarf galaxies, including cases with tidal tails. Three approaches were explored, all implemented within a machine learning framework but differing in the way the data were modeled: (i) an updated, machine learning based version of the probabilistic method used in Battaglia et al. (2022), (ii) a dimensionality-reduction approach, and (iii) a normalizing flow model capable of mapping complex distributions into simpler ones. All methods demonstrated a high capacity for detecting true members; however, the first two primarily identified stars in the central regions of the galaxies, whereas the normalizing flow method was the only one able to consistently recover the external structures, highlighting its potential as a powerful tool for probing the outskirts of dwarf galaxies.