In the 17th century, when Galileo first saw the mountains on the moon, the rings around Saturn, and the satellites orbiting Jupiter, an early telescope of his own making aided his eye. Now astronomers are adding a new tool to their arsenal.
A team of researchers led by David Armstrong at the University of Warwick in the UK recently trained a machine learning algorithm to identify “exoplanets” – that is, planets outside our solar system – from NASA data. The team used the tool to confirm 50 potential new planets, a first for artificial intelligence applied to astronomy.
“Our models can validate thousands of invisible candidates in seconds,” the study authors wrote in the abstract of their paper, which appears in Monthly Notices of the Royal Astronomical Society. Given the gigantic size of many astronomical data sets, the method could greatly increase the speed of discovery for world searches.
Scientists, including those employed by Google, previously used machine learning to identify possible exoplanets. However, the new experiment represents the first time scientists have applied machine learning to “validation,” a further step toward confirming results that involves additional statistical calculations.
The planets mined by the scientists’ program range from the size of Neptune to smaller than Earth, the researchers said. The duration of their stays around their respective stars lasts from 200 days to just one day.
Astronomers search for potential exoplanets by looking for fluctuations in the brightness of distant stars. Periodic dimming of starlight could indicate the presence of orbiting bystanders, such as planets. But aberrations and other celestial objects, like asteroids, can be misleading.
The researchers trained their algorithm on already analyzed data sets collected by NASA’s retired Kepler mission that searches for planets. Once the algorithm got used to distinguishing confirmed planets from false positives, the scientists fed it with data containing as yet unconfirmed planetary candidates. The result: 50 new planets were approved.
The A.I. The technique could be applied to data collected by other space probes. That includes NASA’s Transiting Exoplanets Study Satellite, launched in April 2018. The team behind that telescope concluded its main mission, a two-year study, this summer after finding 66 new exoplanets and 2,100 candidate exoplanets.
The scientists say their tool can work with TESS, as well as the European Space Agency’s upcoming PLATO mission, which stands for “planetary transits and oscillations of stars.
Theo Damoulas of the University of Warwick, a computer science professor and another author on the paper, said A.I. The techniques “are especially well suited for an exciting problem like this in astrophysics.”
