By applying machine learning to the huge volumes of data collected by the TESS satellite, researchers have created one of the most accurate catalogs of exoplanets in our cosmic neighborhood, reports SciTechDaily.
Astronomers at the University of Warwick have confirmed more than 100 exoplanets, including 31 newly identified worlds, using a new artificial intelligence system applied to data from NASA and its Transiting Exoplanet Survey Satellite (TESS). This mission scans the sky for slight dips in star brightness that occur when planets pass in front of their host stars.
The results, published in the journal Monthly Notices of the Royal Astronomical Society (MNRAS), come from a new artificial intelligence-based processing tool called RAVEN. The team used it to analyze observations of more than 2.2 million stars, collected during the first four years of the TESS mission. Their search focused on planets with very short orbits, which complete one rotation around the star in less than 16 days, to better understand how common these close worlds are.
“Using our new RAVEN processing flow, we were able to validate 118 new planets and more than 2,000 high-quality candidates, nearly 1,000 of which are completely new,” said Marina Lafarga Magro, postdoctoral researcher at Warvick University and lead author of the study.
Among the confirmed planets are some particularly important groups:
- ultra-short orbital period planets that orbit their stars in less than 24 hours;
- “Neptunian desert” planets, a type rarely found in a region with few planets due to solar system formation mechanisms;
- multiplanetary systems with close orbits, including newly discovered pairs around the same star.
RAVEN's asset in exoplanet discovery
Modern studies often identify thousands of possible planets, but verifying the actual signals remains difficult. Many false signals come from natural phenomena in space.
“The challenge is to determine whether the dimming is really caused by a planet orbiting the star or by something else, such as eclipsing binary stars – exactly what RAVEN is trying to determine. Its strength comes from our carefully constructed dataset, which includes hundreds of thousands of realistically simulated planets and other astrophysical events that can mimic planets,” said Andreas Hadjigeorghiou, the Warwick researcher who led the development of the instrument. AI processing.
“We've trained machine learning models to identify patterns in the data that can tell us what type of event we've detected, something that AI models excel at,” he added.
“Furthermore, RAVEN is designed to handle the entire process in one step, from signal detection, to its verification using machine learning and statistical validation. This gives the system an additional advantage over contemporary tools that only focus on certain parts of the workflow,” he explained.
Hunting for rare alien planets with an AI tool
In a related study also published in the journal MNRAS, the researchers showed how frequently planets in close orbits around Sun-like stars occur, mapping the results against the orbital period and size of the exoplanets at an unprecedented level of detail.
They found that about 9–10% of Sun-like stars host a nearby planet. This result is consistent with previous findings from NASA's Kepler mission, but RAVEN reduces the uncertainty by up to tenfold.
The study also provides the first direct measurement of planets in the “Neptunian desert”, showing that they exist around just 0.08% of Sun-like stars.
“For the first time, we can assign a precise value to how empty this 'desert' is,” said Kaiming Cui, a postdoctoral researcher at Warwick and lead author of the planetary populations study.
“These measurements show that TESS can now match, and in some cases even surpass, the Kepler mission in studying planet populations,” he pointed out.
