NASA’s Kepler data adds 301 planets thanks to machine learning


Machine learning (ML) methods not only complement existing technology, but also advance scientific research. Now a new deep learning method has added a whopping 301 exoplanets to the total. These planets have been added to the already validated 4,569 planets orbiting several distant stars. The additions were made using a deep neural method called ExoMiner, which works for NASA’s Pleiades supercomputer to discover new planets. Once fed with enough data, ExoMiner learns the task of differentiating between real planets and “false positives”. It was developed based on various tests and properties that human experts use to detect exoplanets. It is also fed a database of confirmed planets and false positive cases.

In an article published in the Astrophysical Journal, the team at the Ames Research Center in California’s Silicon Valley shows how ExoMiner discovered the 301 planets using data from NASA’s Kepler Archives.

Jon Jenkins, an exoplanet scientist at Ames Research Center, said, “Unlike other exoplanet detection machine learning programs, ExoMiner is not a black box – there is no secret why it decides something is a planet or not.” ExoMiner is transparent about the dates confirming or rejecting a planet. A planet is confirmed using identifiable features and then validated using statistics. None of the newly discovered 301 planets have Earth-like living conditions.

Hamed Valizadegan, ExoMiner project leader and machine learning manager, said, “ExoMiner is very accurate and, in some ways, more reliable than both existing machine classifiers and the human experts who are supposed to mimic it because of the bias associated with human labeling . “

Researchers believe that ExoMiner has enough “room to grow”.


Please enter your comment!
Please enter your name here