On March 29th, foreign media reported on March 27 that scientists used AI technology to discover two alien worlds in their archives. In China, the use of AI technology to study aerospace astronomy in the field of less academic, AI technology gradually mature and development, will promote this application field to explore the process of civilization, on AI + aerospace astronomical application technology, foreign scientists have developed planetary machine learning Code.

Machine learning algorithm empowers astronomical recognition

Machine learning algorithms have been used to reveal two previously invisible exoplanets in the data archive of the NASA retired Kepler Space Telescope. Kepler was launched in 2009 and was sent to investigate the dark section of the Milky Way. Its job is to look for alien worlds by scrutinizing the light from distant stars. Armed with a photometer, Kepler looked for a drop in the brightness characteristic of the planet in front of its parent star. It retired last October.

The spacecraft helped scientists discover more than 2,000 distant planets, many of which have yet to be discovered. A team of astronomers and engineers led by the University of Texas at Austin and Google collaborated to use the convolutional neural network to sniff out potential candidate exoplanets. The software uses Kessel's observed stellar and planetary datasets to train, so when looking at other star brightness readings, it predicts the alien planets that exist for each star.

The neural network discovered two previously unknown worlds because it looked for signs of orbiting planets through Kepler data and confirmed their presence using telescopes in Arizona and Hawaii. Christians K2-293b and K2-294b are close to each other, 1300 light years and 1,230 light years away in Aquarius, both of which are larger and hotter than Earth.

Neural network + deep learning training

Revealing the data for this pair of planets comes from the Kepler K2 phase of the mission. In 2013, the four reaction wheels of the two spacecraft failed, and they could no longer stay on a particular star, so NASA reconfigured it to keep its propellers and other remaining wheels stable. .

Anne Dattilo, an undergraduate physics student at the University of Texas at Austin, explains that dealing with K2 data is even more tricky. The team must consider the slight swing and system noise in the spacecraft readings. “I trained my modified algorithm using data from more than 27,000 stars from K2,” she said. “My laptop only takes 40 minutes to successfully train, but we need a few months to figure out how to use it successfully. K2 data."

The convolutional neural network was trained to look for a periodic decrease in the brightness of the stellar light over time, indicating the passage of the planet. Because it trains only with the example of a nearby planet's star, it does not capture all the different types of exoplanets, such as the solar system with a distant world.

“The algorithm missed the 'special' planets: those planets with different signal shapes than ordinary planets,” Dattilo said. “Disintegrating planets is an example. The shape transformation of these planets is different from the 'typical' planets. This means that human astronomers still need to find more interesting planets,” Dattilo said.

However, Dattilo believes that neural networks can still be used to find new planets discovered by NASA's TESS space telescope. The spacecraft was launched last year and is expected to discover thousands of exoplanets within two years. "The same approach should apply to the future TESS, because TESS works the same way as Kepler and K2 - they can measure changes in stellar brightness," Dattilo said. "But I suspect that some changes are needed. I am on Kepler." The K2 code has been changed, and the TESS data is very different from Kepler or K2 data because it can view stars in less time."

This is not the first time AI has helped scientists find a new world. A similar research team using machine learning last year stumbled upon the eighth planet that bypassed the previously missed Kepler 90 star. This discovery made the Kepler-90 the only planetary system with eight planets, just like our own solar system, we have already seen.