Welcome to Shenzhen V-Vision Technology Co., Ltd

News

Contact Us

You are here:Home >> News >> Industry information...

Industry information

Artificial intelligence into the astronomy?

Time:2018-01-09 Views:1436
Illustration of NASA Kepler Detector. The probe was launched in 2009 to look for extrasolar planets. Illustration: WENDY STENZEL, AMES RESEARCH CENTER / NASA
Author: Nadia Drake
   For the first time in astronomy, scientists trained artificial intelligence to screen huge amounts of data collected by telescopes, and the result really uncovered an entirely new planet.
The newly discovered planet, codenamed "Kepler-90i," has been hidden in data collected by NASA‘s Kepler detectors. The planet, about 2,500 light years away from Earth, revolves around a star with seven other planets. Therefore, the Kepler-90 system has many similarities with our solar system.
   "Kepler has proven like us that most stars have planets," NASA‘s Paul Hertz said at a press briefing announcing the discovery. Today, Kepler has confirmed that stars, like our solar system, have a huge family of planets.
   A few days prior to the press conference, media fanaticism may have detected extraterrestrial life. Not surprisingly, the news is completely unreliable, but it proves to the side that machine learning can help us learn more about the likely exciting planet in the entire galaxy.
Search in the sea of ​​stars
   The Kepler probe, launched in 2009, has stared at a small piece of 150,000 stars in the sky for a full four years. Its mission is to look for tiny obstructions to stars when the planet passes in front of the star. When scientists find such tiny signals in their data, they can figure out the size of a planet and how far it is from its parent star.
   Up to now, Kepler detectors have confirmed 2525 planets and more planets to be discovered in their data. However, it is not easy to confirm a planet. For humans, manual combing of large amounts of Kepler data is an insurmountable task, as these data contain 10 or so 8 potential planetary orbits. In addition, the stellar light weakened, not necessarily all the planets are: stellar sunspots, binary stars and other celestial bodies are likely to have the same effect as the planets to cover stars.
   Because of this, Chris Shallue of Google‘s artificial intelligence department decided to use neural networks to solve this problem. Previously, the machine learning approach had been used to screen and classify Doppler data, however, Shallue‘s neural network was able to provide more robust algorithms.
   Shallue said: "When I learned that Kepler detectors collected so much data that scientists could not rely solely on manual reviews, I wanted to use neural networks in astronomy. Our idea was to use this technology In the stars, teach the machine learning system how to distinguish the planets around distant stars. "


Open up a new perspective of observation
   
   As the name implies, neural networks are constructed on the basis of the workings of the human brain. Humans can train neural networks to identify and classify things, such as what distinguishes dogs from cats. Eventually, after having looked through enough samples, the computer can sort the cats and dogs by themselves.
   Shallue trained a neural network to recognize the planet‘s unique "fingerprint." He extracted 15,000 real planetary features from the Kepler database and allowed the neural network system to discern the difference between a real planet‘s signal and a signal disguised as a planet.
   After that is the actual verification stage. Shallue and Andrew Vanderburg of the University of Texas let the system scrutinize 670 stars known to own planets because there may be more planets around these stars.
   Then, they input to the system signals that are not strong enough and can not be handled by humans. In these signals, the neural network system identified two new planets. The results were published in the Journal of Astronomy.
   "The signals of these two stars are weak and all previous searches have missed them," Shallue said.
Still need to explore new areas
   One of the planets is the "Kepler-80g", the sixth known planet in the galaxy‘s home. Kepler-80g is about the size of Earth and takes 14.6 days to revolve around its parent star, while its parent star is smaller and redder than our own sun.
   The neural network also found out "Kepler-90i". The planet, slightly larger than Earth, takes two weeks to complete a revolution. It is the third rocky planet found in its host galaxy, while its parent star is slightly larger and hotter than our own sun. Inside Kepler-90i, there are two smaller planets, while the planets that revolve outside are much larger.
   These planets are large, but are all "clanking" together: the distance between eight planets and their parent star is the same as that of the Earth.
   Vanderburg said: "I do not want to go to a place such as the Kepler-90i, where the surface is likely to be very hot and we calculated that it has an average temperature of about 427 degrees Celsius."
   He also added that there may be more planets to be discovered on the Kepler-90. He and Shallue plan to enter all of Kepler‘s data into the neural network system and see what happens.
   However, there is no need to worry about computers instead of human astronomers.
   NASA‘s Jessie Dotson says: "It‘s never to be ruled out that this job must be done with astronomers, and you first need to have the initial classification to train machine learning before it can handle more than human beings signal."