Thanks to the use of the Google neural network, the aerospace agency NASA opened the eighth planet in the seemingly already studied system of the sun-like star Kepler-90, located 2545 light years from us. Find the “loss” was in the database of the space telescope “Kepler”. Now the Kepler-90 system has become more like Solar. The truth is only in the quantitative question, since its planets are most likely unsuitable for life.
Discovered exoplanet Kepler-90i is a very hot, rocky world with an average temperature of 426 degrees Celsius, which makes it look like our Venus. The planet makes a full revolution around its sun-like star for 14.4 terrestrial days. It is about 30 percent larger than the Earth.
“As we expected, the archival data of the Kepler telescope contain real treasures, waiting for a suitable tool or technology to unearth them,” commented Paul Hertz, director of astrophysics at NASA’s aerospace agency.
This discovery was made possible thanks to the Google AI engineer-programmer Christopher Shall and the astronomer from the University of Texas, Andrew Vanderburg, who trained their neural network for identifying weak transit data (the moments when the exoplant passes by its star) recorded by the Kepler telescope .
The system of the sun-like star Kepler-90 in the artist’s view
“The Kepler-90 star system is similar to the mini version of the Solar System. In it, too, like ours, small planets are located in internal borders, and large ones are on the outside, although they are packed much more densely. For example, the external Kepler-90h is removed from the star by 1 astronomical unit, which is equal to the distance from the Earth to the Sun, “explained Andrew Vanderburg.
Machine learning has already been used to analyze Kepler data, but a new study demonstrates that neural networks are a more powerful tool in the search for weak signals from distant worlds.
The data set collected over the last four years of the Kepler telescope contains information on 35,000 potential planetary signals. Automatic analysis checks the most promising of them, but the weakest signals often elude when searching by this method. Astronomers often have to analyze the data virtually manually, but their huge volume makes this work extremely inefficient and long.
To solve the problem, the researchers trained the neural network to identify the transit of planets, using a set of 15,000 previously analyzed signals. In the tests, the neural network correctly identified the true planets and false positives in 96 percent of the cases. After the engineers taught it to determine the time of transit, the researchers used the system to search for weaker signals in systems of 670 stars, which already had several known planets. Their assumption was that systems with several exoplanets are the best places to search for undiscovered worlds.
The planet Kepler-90i was not the only world that could detect a neural network. For example, in the system Kepler-80, she discovered a sixth planet Earth-sized, which received the designation Kepler-80g.
Next, scientists plan to use a neural network to analyze the entire Kepler telescope database, which contains information on more than 150,000 stars. Who knows what other secrets and treasures she can find. Perhaps thanks to a neural network that can more effectively analyze planetary signals, we will finally find that very Earth 2.0.