QUBRICS: machine learning for searching bright, high-redshift quasars

The size and complexity of current astronomical datasets has grown to the point of making human analysis, in many cases, impractical or even impossible. With the advent of future observatories (e.g. LSST) and the generation of an unprecedented amount of data, automatic tools to extract information have become mandatory.
A practical example of this situation is the search for bright, high-z QSOs in wide area surveys. These targets, among the rarest sources in the sky, enable a wide range of scientific applications both in cosmology and fundamental physics, but their number in the southern sky is relatively scarce. To fill the gap, we developed QUBRICS (QUasars as BRIght beacons for Cosmology in the Southern hemisphere) that is based on machine learning techniques applied to current and future photometric databases. Since 2019 over 450 new, bright (i < 18 and/or Y<18.5) and high-redshift (z > 2.5) QSOs have been identified using different, complementary methods (e.g. CCA, PRF, XGB). This talk will describe the QSO selection algorithms, their performances and the current state of QUBRICS, highlighting peculiarities, lessons learned, future prospects and scientific advancements enabled by QSOs discovered by QUBRICS. As an example, the gain produced by QUBRICS to carry out the Sandage Redshift-Drift Test at the ELT will be shown.