Report that 50 kinds of new planets were discovered for the first time by verification with machine learning algorithm
A research team
Exoplanet Validation with Machine Learning: 50 new validated Kepler planets | Monthly Notices of the Royal Astronomical Society | Oxford Academic
50 new planets confirmed in machine learning first
A star that emits light like the sun is relatively easy to observe, but it is difficult to find a planet that does not emit light. A new planet can be discovered by examining the 'brightness change when the planet crosses the front of a star and blocks light,' but it may also be a false detection due to an error in the observation camera or gravity interference.
Therefore, the research team used an algorithm learned from huge observation data collected by the Kepler Space Telescope and the Transient Exoplanet Search Satellite (TESS) , and whether the brightness change seen in the data is due to a real planet, or just I verified whether it was a false positive.
Then, when the data that the algorithm judged as a planet was observed again with the telescope, it was confirmed that 50 kinds of planets were real. The discovered planets had an orbital period of 1 to 200 days and varied in size from the same size as the earth to Neptune with a diameter about 4 times smaller than the earth.
“There has been no probabilistic verification of data suspected to be a planet using machine learning methods,” says David Armstrong, a professor of physics at the University of Warwick.
'Although about 30% of the planets discovered so far have been validated in one way, it is not ideal. It is desirable to develop new methods for validation,' Armstrong said. Training takes a long time, but an algorithm can analyze tens of thousands of data at once, and the algorithmic verification method allows the system to automatically execute with few steps, accelerating the process. ].
in Science, Posted by log1i_yk