British astronomers have developed a machine learning algorithm that can analyze images from the TESS and Kepler telescopes and check if distant stars actually have exoplanets. In particular, he has already confirmed the existence of 50 exoplanets by analyzing Kepler's data. The results of their work were published in the scientific journal Monthly Notices of the Royal Astronomical Society.
“Thanks to this algorithm, we transferred 50 candidates to the category of confirmed exoplanets at once. Nobody has used machine learning systems for this before. Now we can not only say which of the candidates is most likely a planet, but we can also accurately calculate the probability of this "- explained one of the authors of the study, planetary scientist from the University of Warwick (UK) David Armstrong.
Over the past few years, astronomers have found more than a thousand exoplanets and several thousand candidates for this role. Most of them belong to the so-called hot Jupiters - planets the size of Jupiter, which are an order of magnitude closer to their star than Mercury is to the Sun. At the same time, among exoplanets, smaller planets are increasingly found, which are comparable in size to the Earth.
Most of the known exoplanets were discovered by the Kepler telescope. For almost four years, he continuously monitored hundreds of thousands of stars that are located on the border of the constellations Cygnus and Lyra. If his pictures showed that some star periodically decreases in brightness, then this could be a sign that from time to time it was "blocked" from the telescope by a planet revolving around the star. Astronomers call this phenomenon a passage or transit.
However, the reason for this may be other phenomena, including processes within the luminaries themselves. As a rule, long-term observations make it possible to separate one from the other, but this requires a very long and painstaking comparison of images and analysis of all available scientific data on the activity of a star.
Artificial intelligence clues
British scientists have developed a machine learning algorithm that can solve this problem faster and better than a human or classical statistical methods for analyzing information. It is a multi-layered neural network that can find hidden patterns in a series of images of stars.
To train this artificial intelligence, scientists used a dataset that Kepler collected from the discovery of already confirmed exoplanets, as well as objects whose existence was not later confirmed. In total, more than 30 thousand transits were driven through artificial intelligence for training.
Scientists have tested the operation of the algorithm on several hundred yet unconfirmed planets from the Kepler catalog. The algorithm has identified 50 objects that are more than 99% likely to be exoplanets. Astronomers subsequently confirmed this using other data analysis methods.
The researchers believe that their development can be used to automatically and very quickly search for new exoplanets. The algorithm can analyze data from TESS and other telescopes in real time. In particular, Armstrong and his colleagues hope that their methodology will be used in the work of the PLATO European space observatory under construction, which is scheduled to launch in 2026.