Posted by Robert Vamosi on December 19, 2017
Using previously studied data from NASA’s Kepler Space Telescope, a Google machine learning algorithm discovered two new planets orbiting a nearby star already known to have six planets. At a total of eight planets, that system matches our own solar system. It also begs a question: How many other planetary systems might be hiding in previously analyzed data?
“Just as we expected, there are exciting discoveries lurking in our archived Kepler data, waiting for the right tool or technology to unearth them,” said Paul Hertz, director of NASA’s Astrophysics Division in Washington, in a press release. “This finding shows that our data will be a treasure trove available to innovative researchers for years to come.”
In looking for exoplanets, planets that orbit other suns, the Kepler Space Telescope photographs the same star field every half hour for several years. For example, between 2009 and 2013, Kepler took pictures of nearly 200,000 stars every half hour, producing images of about 10 pixels per picture. That’s a lot of data to sift through.
The idea is that if a star dims and brightens, in a repeating cycle, that could mean a few things to astronomers. It could be a binary star, meaning two stars orbit a common point of gravity. Or it could be that one or more planets are orbiting it.
The star Kepler-90 was already known to have six planets. Finding that out required astronomers to comb through the Kepler data manually. It has been estimated that to rule out a single star it would take an astronomer on average of half an hour for each. There were literally thousands of stars identified by Kepler that require this extra attention.
Kepler’s four-year data set consists of more than 35,000 possible planetary signals. This could take a very long time.
So, Christopher Shallue, a Google researcher, and Andrew Vanderburg, a Sagan Fellow at the University of Texas at Austin, decided, as a test case, to take a closer look at 670 stars with previously known exoplanets. They wanted to use machine learning to both learn how to identify the characteristics of these objects and also search for weaker signals others might have missed.
The idea isn’t far fetched. In February 1930, the minor planet Pluto was discovered by Clyde Tombaugh, who returned to notebooks created by astronomers unsuccessful before him. Even so, he had to compare the new photographic plates taken on different nights of the same star field using a blink microscope to mask one side and then other so that he could spot any changes between the two. Gradually the nightly progression of one of the objects emerged. This it took hours of painstaking examination.
Something similar happened here. Astronomers combing through the raw Kepler data used a filter. Even so, there was a large set of data that required a lot of human input. It was estimated that to rule out one star would take on average of half an hour. There were literally thousands that needed this extra attention.
Exoplanets – planets outside our own solar system — can be found by looking for a change in a star’s brightness which could indicate a planet passing in front of it. The neural network developed by Google learned to identify planets from 15,000 Kepler signals that had already been known by scientists to have exoplanets. When the machine learning then looked at data from these 670 stars, it found the two planets around Kepler-90, a hotter, more massive star than our own. It is 2,500 light years away.
One of the new planets the team discovered through machine learning was Kepler-90i, which is about 30% larger than our own planet. Because all the known planets in the Kepler-90 system orbit closer to their sun than our own, Kepler-90i likely has an average temperature of about 800°F. It orbits its star every 14.4 days.
“This discovery of an eighth planet ties Kepler-90 with our own solar system for having the most known planets,” said NASA astrophysicist Paul Hertz during a press conference about the discovery.
Now, based on this limited success, the researchers hope to study the larger set of Kepler’s data, from more than 150,000 stars, to see if they can spot any other weak signals previous researchers may have missed.
Melissa Kirschner is Web Editor-in-Chief at Synopsys. She has been a writer, editor, and content strategist in high-tech for more than 15 years. Melissa is fascinated by the socio-cultural implications of the digital age. When not researching AI, autonomous driving, 5G, cryptography, and medical advancements, she enjoys reading the works of Neil Gaiman, watching dystopian dramas, and rescuing abused animals. Most of all, she likes to write things that people like to read.