Title: A machine learning-based tool for open cluster membership determination in Gaia DR3

Abstract: Membership studies characterising open clusters with Gaia data, most using DR2, are so far limited at magnitude G = 18 due to astrometric uncertainties at the faint end. Our goal is to extend current open cluster membership lists with faint members and to characterise the low-mass end, which members are important for many applications, in particular for ground-based spectroscopic surveys. We use a deep neural network architecture to learn the distribution of highly reliable open cluster member stars around known clusters. After that, we use the trained network to estimate new open cluster members based on their similarities in a high-dimensional space, five-dimensional astrometry plus the three photometric bands. Due to the improved astrometric precisions of Gaia DR3 with respect to DR2, we are able to homogeneously detect new faint member stars (G > 18) for the known open cluster population. Our methodology can provide extended membership lists for open clusters down to the limiting magnitude of Gaia, which will enable further studies to characterise the open cluster population, e.g. estimation of their masses, or their dynamics. These extended membership lists are also ideal target lists for forthcoming ground-based spectroscopic surveys.