Artificial intelligence first discovered the planet

Techno 15 December, 2017

2017-12-15 19:43

Artificial intelligence first discovered the planet
Experts taught artificial intelligence to find other worlds.

Sign up for news “UkrMedia” in Facebook, Twitter or Google+

14 December 2017 in the US at 13:00 et, started a NASA press conference, reports Rus.Media. Reason happened serious: experts have taught artificial intelligence to find other worlds.

The Kepler telescope, designed to search for exoplanets, and the method of transits, which is used for this. In a nutshell the essence of this. When the planet obscures the star from the observer, the apparent brightness of the lights falls in a characteristic way. Such periodic attenuation of light and allow to detect a planet.

However, to highlight them on the chart of star Shine. The reduction in brightness caused by the Eclipse, is very small. To the extent star can Shine and change itself (e.g., due to spots like the sun). Standard mathematical methods to search for periodic signals with the task of finding exoplanets cope poorly.

In this regard, the “Kepler” even attracted volunteers. Every Internet user can participate in the discovery of new planets, becoming a member of the project Planet hunters. The human brain is to cope with the recognition (including periodic drops on the chart) is much better than any software.

However, our natural computer is not performance. Besides a weak brightness changes caused by exposure to small or far from its sun worlds, even for the perfected evolution of human vision drowned out by the noise.

This prompted Christopher Frisky (Christopher Shallue), a specialist in artificial intelligence and employee of Google, to develop systems otakibashi new planet.

We talked about what neural networks are and how they learn. These programs to some extent, reproduce the principles of teaching are characteristic of the nervous system of a human or other animal. This allows them to solve problems that are difficult to algorithmization.

An engineer from Google’s Naughty and astronomer from NASA Andrew Vanderburgh (Andrew Vanderburg) began to teach the computer to search the planet according to “Kepler”. As the press release NASA, they began with the fact that the car showed 15 thousand records that had been previously tested by other methods. Having that artificial intelligence has correctly answered the question whether there is a planet or not, 96% of the cases, the researchers began to search for yet undiscovered worlds.

Neural networks have been proposed observations 670 stars have planets. Scientists logic was simple: these bodies will have “family”, then it may be a previously unknown “households”.

And opened a planet Kepler-90i, which is only a third more Land. She was eighth (in order of opening) in a sun-like star Kepler-90 remote from our sun at 2545 light years.

Unfortunately, this world is difficult to name suitable for life. One revolution around its sun the planet makes for 14, 4 earth days, and the average surface temperature should reach 400 degrees Celsius. Other well-known worlds of this system is also very close stars, and only the outermost known its planets, Kepler-90h, lies in the habitable zone.

Soon, the program identified another planet. The world Kepler-80h has five famous “brothers”, and a size similar to Earth.

While catch e-astronomer is confined. But the first step is the hardest. The authors plan to work with their offspring any data collected “Kepler” for four years, and that the results of observation of more than 150 thousand stars.