Recently, a neural network contributed to the discovery of the first star with eight known planets, like our own solar system. NASA teamed up with Google to troll through data collected by the Kepler Space Telescope since 2009. Google engineers created a neural network designed to analyze the historic data, looking for planetary transits.
Researchers used previous data showing confirmed exoplanet signals to train the system. As a planet crosses between its star and the sensor, the sensor records a dip in brightness. Most of the previous discoveries were done manually or with the help of algorithms. Armed with this information, the neural network looked for patterns in these datasets. Researchers gave the network 670 weaker signals from Kepler and found two likely candidates, Kepler-90i and Kepler-80g. Kepler-90i was found in the Kepler-90 system, making it the eighth known exoplanet in that system.
While Kepler-90 may have the same number of planets, there are many differences. The planet furthest from the star orbits at a similar distance as Earth does to the Sun. All the other planets orbit much closer and are relatively small.
There is a wealth of information kept in archives from past missions, more than can ever be deciphered manually. Machine learning opens new doors and windows into previously collected data, seeing patterns that were undetectable to the human eye. The researchers carrying out the study plan on expanding their study to include the full dataset of 150,000 stars.
In other transit news, Thomas Jacobs, a citizen scientist, was the first to discover a comet orbiting a distant star. Instead of just seeing the dip in brightness a comet tail was detectable, blocking slightly more light than usual. In this case, the human eye was more capable of detecting the faint differences than an algorithm