An artificial intelligence tool enabled the detection and confirmation of over a hundred new exoplanets. This harvest of worlds brings new information concerning the diversity and distribution of planetary systems.
TESS, NASA's "Transiting Exoplanet Survey Satellite," scans the sky to capture the tiny dips in brightness caused by a planet passing in front of its star. The data accumulated over four years was processed by specialized software, RAVEN, designed to identify signals with great accuracy.
Artist's impression of the Kepler-11 system, an example of a multi-planet system with close orbits. Around this Sun-like star, six planets orbit, sometimes with several simultaneous transits, as illustrated here for three of them observed by NASA's Kepler mission in August 2010.
Credit: NASA/Tim Pyle This software uses machine learning models trained on realistic simulations to distinguish genuine exoplanets from ambiguous signals, such as those generated by binary stars. This integrated method handles both detection and validation of information in a single step, providing greater consistency and objectivity than classical approaches.
Among the now-confirmed worlds are ultra-short-period planets, completing an orbit in less than 24 hours, as well as unusual specimens located in the 'Neptunian desert,' a region where bodies of this size are rare. Systems hosting multiple planets in tight orbits have also been identified.
Researchers were able to estimate that nearly 10% of Sun-like stars possess a short-orbit planet, a result compatible with previous work but with a reduced margin of error. The Neptunian desert, on the other hand, appears to exist around only 0.08% of these stars, providing a quantified measure for this zone.
The robustness of this new catalog opens doors for in-depth studies, especially since tools have been published to help astronomers select the most interesting systems. Future missions, like the European Space Agency's PLATO, will be able to rely on these resources.