A new artificial intelligence algorithm has discovered over 300 previously unknown exoplanets in data gathered by a now-defunct exoplanet-hunting telescope.
The Kepler Space Telescope, NASA's first dedicated exoplanet hunter, has observed hundreds of thousands of stars in the search for potentially habitable worlds outside our solar system. The calatog of potential planets it had compiled continues generating new discoveries even after the telescope's demise. Human experts analyze the data for signs of exoplanets. But a new algorithm called ExoMiner can now mimic that procedure and scour the catalog faster and more efficiently.
The telescope, which stopped working in November 2018, looked for temporary decreases in the brightness of the stars that might be caused by a planet crossing in front of the star's disk as seen from Kepler's perspective. But not all such dimmings are caused by exoplanets, and scientists had to follow elaborate procedures to distinguish false positives from the real stuff, according to a NASA statement.
ExoMiner, is a so-called neural network, a type of artificial intelligence algorithm that can learn and improve its abilities when fed a sufficient amount of data. And Kepler generated plenty of data: In the less than 10 years of its service, the telescope discovered thousands of planet candidates, nearly 3,000 of which have since been confirmed. That is a vast majority of the overall 4,569 exoplanets currently known.
For each candidate exoplanet, scientists poring through the Kepler data would look at the light curve and calculate how large a portion of the star the planet seems to be covering. They would also analyze how long it appears to take the would-be planet to cross the star's disk. In some cases, the observed brightness changes are not likely to be explained by an orbiting exoplanet. The ExoMiner algorithm follows exactly the same process but more efficiently, which allowed the researchers to add a batch of 301 previously unknown exoplanets into the Kepler planet catalog at once.
"When ExoMiner says something is a planet, you can be sure it's a planet," Hamed Valizadegan, ExoMiner project lead and machine learning manager with the Universities Space Research Association at the NASA Ames Research Center, said in the statement. "ExoMiner is highly accurate and in some ways more reliable than both existing machine classifiers and the human experts it's meant to emulate because of the biases that come with human labeling."
Now that ExoMiner proved its skills, scientists are looking to use it to help sift through data from other existing and upcoming exoplanet-searching missions, such as NASA's current Transiting Exoplanet Survey Satellite (TESS) or the European Space Agency's Planetary Transits and Oscillations of Stars (PLATO) mission that will launch in 2026.
Unfortunately, none of the newly confirmed exoplanets are likely candidates to host life, as they are outside of the habitable zones of their parent stars.
The paper was accepted for publication in the Astrophysical Journal, NASA said in the statement; a draft of the paper is available to read on the preprint site arXiv.org.