NASA Uses Deep Learning to Add 301 Entries to Kepler's List of Exoplanets

For years, the Kepler mission scanned the skies and sent us data to determine which of these stars could possibly host or support life on their orbiting planets. As per recent updates, NASA currently has 3,392 planetary systems that it is analyzing and close to 8,000 exoplanets candidates. Experts at NASA sift through the available data to determine if a candidate is actually an exoplanet. They have had the assistance of the Pleiades supercomputer so far, but a new deep neural network called ExoMiner has now begun validating them. 

Deep neural networks learn how to perform a task when you provide them with sufficient data. ExoMiner uses the same tests that human experts use to confirm exoplanets and has also been trained on the "false positive" dataset to increase its accuracy. Details of the workings of the neural network will soon be published in a paper in The Astrophysical Journal, NASA said. 

ExoMiner was tasked with scanning data from the Kepler archives. The Kepler Science Operations Center pipeline had already shortlisted the new entries as candidates which needed to be further validated. ExoMiner used multiple observation techniques and statistics to validate the exoplanets. The neural network can pinpoint the signatures that reveal that the orbiting body is indeed a planet, something that the scientists can also verify in detail, the press release said. 

Unfortunately, none of the 301 newly confirmed exoplanets fit into the category of Earth-like exoplanets or are orbiting in the habitable zones of their stars. However, NASA is confident that with a little bit of fine-tuning, it will be able to repurpose ExoMiner to look into the data generated by its Transiting Exoplanet Survey Satellite (TESS) or the European Space Agency's PLAnetary Transits and Oscillations of stars, or PLATO missions, which are both aimed at capturing exoplanets as their transit across their stars, the press release said.