AI to find other planets

Space telescopes travel through space in search of new habitable worlds. (NASA)

Discovery of a new planet, now possible machine learning

In recent years, several space missions have been launched to look for habitable worlds in our galaxy.

In 2013, the Kepler space telescope detected one star and seven planets orbiting it: the Kepler-90 system. Four years later, it has been discovered that this system does not have seven planets but eight, just like our very own Solar System.

The planets of the Kepler-90 system are arranged in a similar way to those of our solar system: the ones closest to the sun are smaller than the ones farther away. (NASA)

Probes and telescopes that travel through space send tons of data back to Earth that must be analyzed … not an easy task! The discovery of the eighth planet, Kepler-90i, has been possible thanks to ‘machine learning’, a new AI technology developed by Google.

Teaching computers

Thanks to space missions, thousands of new planets have been discovered, which could easily become our new planet Earth. However, these planets are so far away that they can’t be seen. How do we actually know that they are there?

Space telescopes such as Kepler detect variations in the light emitted by stars. When a star shines less, this could be because a planet is orbiting and passing by (known as concealment).

But work does not end there yet. The collected data is sent to Earth, where scientists must analyze and compare it. There is so much information that computers are needed to process it, but a computer does not have the human ability to analyze and deduce… or does it?

“Machine learning” is a kind of AI: through software programming, computers are taught to learn and think for themselves.

In order to discover new planets, software engineers Christopher Shallue and Andrew Vanderburg have ‘trained’ a computer to learn how to interpret light variations and conclude whether they are actually planets.

In their blog, both scientists explain how they used machine learning to analyze Kepler mission data. Over four years, the telescope has observed 200,000 stars, capturing an image every 30 minutes and creating 14,000 million data points.

This is a huge amount of information, even if it is analyzed by the most powerful computers. Analyzing this information would be hard and would take a long time, which is why machine learning is more effective.

Translated by Chaplin’s Languages | Find out more in Junior Report


Please enter your comment!
Please enter your name here