Google Uses Machine Learning to Find New Exoplanets

Astronomers have been spotting exoplanets for decades, but the overwhelming majority of newly discovered planets outside our solar system have come courtesy of the Kepler mission. This space observatory has found well over a thousand exoplanets, but Kepler accumulated a ton of data that’s difficult to analyze. Google has stepped in to lend astronomers a hand with machine learning technology, and this partnership is already bearing fruit. Google’s TensorFlow platform has identified two previously missed planets orbiting faraway stars.

In the first phase of Kepler’s mission, the spacecraft watched large swaths of the sky, taking images every 30 minutes. Kepler watches for dips in brightness, which indicate a planet passed in front of a distant star. This so-called “transit” method of exoplanet detection has proven very effective, but Kepler produces enormous volumes of data. Over its first four year mission, Kepler produced about 14 billion data points, so you need some sort of software analysis to cull the list of potentially important signals to the point humans can have a look. Google’s software can do that more efficiently with the power of machine learning.


Machine learning is all about teaching a computer how to recognize patterns, and it requires large amounts of data for training the system. That makes exoplanet hunting the perfect application in many ways. In this case, Google AI researcher Chris Shallue worked with astrophysicist Andrew Vanderburg from UT Austin to train the system with more than 15,000 labeled Kepler signals. The TensorFlow model learned what planets and not-planets looked like in the Kepler data, eventually reaching a 96 percent accuracy rate when shown new data.

In the first application of the new planet-hunting AI, Shallue and Vanderburg unleashed it on the 670 stars that were known to have exoplanets. It found two new planets, one orbiting Kepler 80 and another around Kepler 90. The one orbiting Kepler 90 (known as Kepler 90i) is particularly interesting.


The discovery of Kepler 90i boosts the number of known planets around this star some 2,545 light years away to eight, just like our own solar system. It’s the only other eight-planet system known to exist. Kepler 90i is 30 percent larger than Earth and orbits its host star in just 14 Earth days. The surface temperature is somewhere in the neighborhood of 800 degrees Fahrenheit.

Google plans to feed more data into the model to see if more planets are hiding in there. After all, it’s only scanned 6,470 out of 200,000.