ExoMiner++: planet spotter
ExoMiner++ is a deeplearning artificial intelligence (AI) model developed by NASA researchers to spot exoplanets in telescope data. It’s the successor to ExoMiner, which examined Kepler telescope data. ExoMiner++ will comb through the Transiting Exoplanet Survey Satellite (TESS) mission as well.
ExoMiner++ analyses graphs showing the brightness of a star over time. When a planet crosses in front of its star, it causes a dip in brightness. The challenge is to separate true planetary signals from false positives, such as two stars orbiting each other or background objects that mimic planetary transits.
Unlike many blackbox AI models, ExoMiner++ is designed to be explainable. It provides astronomers with a score indicating the likelihood of a signal being a planet and offers insights into why it made that classification.
ExoMiner is credited with validating 370 new exoplanets from Kepler data. These planets were from a shortlist stuck in limbo: scientists had listed them as candidates but their signals were too ambiguous for standard validation techniques.
Compared to ExoMiner, ExoMiner++ is trained on both Kepler and TESS data and can compare a much greater number of stars at a time. As a result, thus far, ExoMiner++ has identified around 7,000 potential exoplanet candidates in TESS data.
NASA has released ExoMiner++ as opensource software on GitHub and has invited researchers to replicate NASA’s results, apply the model to their own datasets, and refine the algorithm for future missions, such as the Nancy Grace Roman Space Telescope.