Speaker
Description
Hot subdwarfs are compact, core helium-burning stars located on or near the extreme horizontal branch, with temperatures of 20,000–50,000 K and very thin hydrogen envelopes. They are widely regarded as products of binary interaction and are important contributors to the ultraviolet emission of old stellar populations. In this talk, we present our classification of 20,061 hot subdwarfs into single and binary systems. We analysed low-resolution Gaia XP spectra of our sample objecs with several machine learning methods. To explore structure in the high-dimensional spectral data, we applied Uniform Manifold Approximation and Projection for dimensionality reduction, and then used an ensemble of convolutional neural networks, trained on literature-labelled systems, to assign binary probabilities while accounting for class imbalance. The similarity mapping revealed distinct substructures associated with composite systems and enabled the identification of likely contaminants. Our classification showed a strong increase in the inferred binary fraction among stars with astrometric and photometric variability. These results demonstrated the potential of Gaia low-resolution spectroscopy combined with machine learning for population-scale studies of hot subdwarf evolution.