Press conference results: Thanks to Google's AI, NASA confirmed the record of the Solar System was flattened

We found a mini version of the Solar system - Kepler-90 star system, with the most planetary in the same system ever.

Our solar system is really special, and there are two reasons that make that special. The first is Earth - the only planet in human knowledge that can nourish life. And second, this is the stellar system - the planet has the most planets, with eight spheres around the flaming Sun.

However, it seems that the Solar system has become a little less special after NASA's announcement in the press conference that just happened. Accordingly, the Kepler space telescope has found a new star planet - Kepler-90 planet, 2545 light-years from Earth.

This is the star system - the planet has the most planets, with 8 spheres revolving around the blazing Sun.

This star system has been confirmed to have a total of 7 planets. And this new discovery, it officially raises the number of spheres that rotate a star with our solar system.

The new planet - named Kepler-90i - is a hot planet with a solid surface. It has an orbit of 14.4 days around Kepler-90.

And the special point here is that Kepler has teamed up with Google 's AI (artificial intelligence) system to discover the planet.

Specifically, this AI system can "learn by itself" (machine learning). It analyzed the information in Kepler's data system during the reception of signals from exoplanets (exoplanets), and then confirmed the signal of a new planet.

Star system - planet Kepler-90.

"As expected, this discovery is actually lurking in Kepler's data, just waiting for us to have a powerful tool to dig it up" - Paul Hertz, director of the Department of Physics NASA astronomy at Washington headquarters.

"This proves our available data has been a treasure for the next generation" excavated "for years to come."

Discovery inspired by the human brain

This discovery stems from "Christopher Shallue and Andrew Vanderburg" researchers "training" a computer to find exoplanets , based on the light data recorded from Kepler. These are very small changes in brightness when a planet moves across a star.

Inspired by neurons in the human brain, the "neural networks" in AI will dig up Kepler's data, look for micro-changing signals, and eventually find the eighth planet Kepler-90's long-running hiding place in the constellation of Draco.

In fact, the use of machine learning technology to study Kepler's data has been done before, but is not expected too much. But this time, it showed great potential to find the smallest signals in extremely distant galaxies that we could not reach.

The planet across the star's star will create a very small change in the light signal.

Kepler-90i is quite impressive, about 30% larger than the Earth. However, the planet's surface is very hot - estimated at over 800 degrees Fahrenheit (over 420 degrees Celsius), equivalent to Mercury in the solar system. The furthest planet of Kepler-90 is Kepler-90h, which has the same distance as the Earth to the Sun.

"Star system - Kepler-90 planet is like the miniature version of the Solar system. The small inner planets, large outer planets, but with much smaller distances than us" - Vanderburg, research NASA student and astronomer from University of Texas said.

Shallue, Google's senior software engineer, was excited about finding a new exoplanet, after realizing that astronomy - like any other science - was always "flooded" in a tank. Data collected from the universe.

"In my free time, I tried to understand and realize Kepler's mission has a huge source of data," Shallue said. "The technology of self-study machines is really useful now, because such data is too much for people to study by themselves."

Within four years, Kepler's database has collected 35,000 "likely" signals as a new planet. In the past, the way to test them was with automatic software and . human eyes, but weak signals were often ignored. As for Shallue and Vanderburg, those data may contain unprecedented findings.

How did they train AI?

First, they teach the AI ​​system of neuron to confirm signals from exoplanet, based on 15,000 previously recognized signals.

In the first tests, this system achieved an accuracy of up to 96%. The system continues to self-study to achieve higher accuracy. The experts then directed it to the search for weaker signals out of a total of 670 star systems, which had many confirmed planets. This action is based on the assumption that star systems with more than 1 planet are the best place to search for more hidden planets.

Kepler-90i simulation.

"We found a lot of" fake "planets, but at the same time the ability to find a real planet is also larger," Vanderburg said. "Just like kicking stones to find jewels, you have a big sieve, you will pick up more stones, but also get more gems."

Kepler-90i is not the only "jewel" AI has found. In the star system Kepler-80, which has five planets, it also found the sixth planet. It is Kepler-80g , a planet the size of Earth. Together with neighboring planets, they form a resonant chain, making the whole system stable, as found in February 2017.

According to Shallue and Vanderburg, the duo plans to apply this AI system to Kepler's entire data system, including 150,000 stars. This huge number promises many planets to be found, and it is clear that the chance of encountering another civilization is also greater.

The study will be published in The Astronomical magazine in the near future.