A stunning 301 freshly verified exoplanets have just been added to the overall number of exoplanets. The swarm of planets is the most recent addition to the 4,569 already certified planets circling many distant stars. How did scientists find such a large number of planets so quickly? The solution is ExoMiner, a novel deep neural network.
Deep neural networks are machine learning systems that, when given enough data, learn a task autonomously. ExoMiner is said to be a new deep neural network that uses NASA’s Pleiades supercomputer to identify between actual exoplanets and other forms of imposters, or “false positives.” Its design is influenced by the many tests and qualities used by humans to confirm new exoplanets. It also learns from previously verified exoplanets as well as false positives.
ExoMiner augments those skilled at going through data and determining what is and isn’t a planet. Data obtained by NASA’s Kepler spacecraft and its successor mission, K2, in particular. Poring over vast datasets is a tremendously time-consuming operation for tasks like Kepler, which has thousands of stars in its vision, each of which might contain numerous possible exoplanets. ExoMiner provides a solution to this problem.
“Unlike previous exoplanet-detecting machine learning systems, ExoMiner isn’t a black box – there’s no mystery as to why it determines something is a planet or not,” said Jon Jenkins, an exoplanet scientist at Ames Research Center of NASA in Silicon Valley. “We can simply explain which data factors cause ExoMiner to reject or affirm a planet.”
What is the distinction between an exoplanet that has been confirmed and one that has been validated? When several observation techniques reveal traits that a planet can only explain, the Earth is said to be “confirmed.” Statistics are used to “validate” a world, indicating how probable or unlikely it is to be a planet based on the facts.
The Ames team demonstrates how ExoMiner identified the 301 planets by utilizing data from the Kepler Archive’s remaining potential worlds – or candidates. The Kepler Science Office upgraded all 301 machine-validated planets to planet candidate status after discovering them via the Kepler Science Operations Center pipeline. However, it was not confirmed that they were planets until ExoMiner.
The article also shows how ExoMiner is more exact and reliable in ruling out false positives and better equipped to disclose the actual signs of planets circling their parent stars — all while allowing scientists to view in detail what drove ExoMiner to its conclusion. When ExoMiner confirms it is a planet, you can be confident it’s a planet,” said Hamed Valizadegan, ExoMiner project lead at Ames. “Because of the biases inherent in human labeling, ExoMiner is very accurate and, in some respects, more dependable than both current machine classifiers and the human experts it is supposed to replicate.”
None of the newly discovered planets are considered to look like Earth or in their parent stars’ habitable zones. They do, however, share similarities with the general population of verified exoplanets in our galactic vicinity. “These 301 findings help us better comprehend planets and solar systems other than our own, as well as what distinguishes ours,” Jenkins added.
ExoMiner will have additional opportunities to prove its worth as the search for exoplanets continues, with missions utilizing transit photometry such as NASA’s Transiting Exoplanet Survey Satellite, or TESS, and the European Space Agency’s planned PLAnetary Transits and Oscillations of Stars, or PLATO.