Learning novel representations of variable sources from multi-modal Gaia data via autoencoders | Publicación