Have you always wanted to explore space and find strange new alien worlds? Are you too lazy to leave your comfortable chair and the warm, reassuring glow of your computer screen? Google has some good news for you armchair star-ship captains out there. The machine-learning code responsible for the discovery of two exoplanets back in December has been released to the public, so you can now join the ongoing search for exoplanets and help uncover the strange secrets of our universe.
In a blog post published Thursday, March 8, senior Google software engineer Chris Shallue detailed the machine-learning code and how it can be used to help search for alien planets. To detect planets outside our solar system using tools like the Kepler space telescope, astronomers look at the light and other cosmic radiation that hits the telescope’s photometer. When there’s a conspicuous dip in an otherwise stable amount of light being detected by the telescope, there’s a chance that a planet, star, or something else may be responsible for blocking out some the light. There’s a chance, too, that it might just be instrumental noise. Once an anomaly in the signal is noticed, an algorithm makes a calculation as to the probability of an exoplanet’s existence. It’s not confirmed, however, until an astronomer manually looks through the data and can make an informed decision about what is causing the anomaly.
Because of the immense amount of data being analyzed, astronomers had to develop a way to avoid being overwhelmed by false positives caused by instrumental noise. A signal-to-noise cutoff ratio is applied to the data and any signals below the cutoff point are deemed too likely to be noise to warrant further review. While necessary, such a practice means there may be a number of actual exoplanets who’s signal was below the cutoff ratio, most likely smaller Earth-sized planets, the planets most likely to harbor alien life. That’s where Google’s code comes in.
Google’s code looks through the rejected Kepler signals and, like the terrifying piece of artificial intelligence it is, learns more ways to separate signal noise and actual anomalies. To test this, Google put its code to work on the data from 670 stars observed by the Kepler telescope and rejected by the automated analysis. From that formerly rejected data, Google identified two more exoplanets, Kepler-90 i and Kepler-80 g.
There is much more rejected Kepler data to sort through and Google has released its code to the public in hopes that open-sourcing it can help improve it. There is another space telescope being launched next month, too: the Transiting Exoplanet Survey Satellite (TESS) will launch on April 16, 2018 for a two year mission to observe potential exoplanets. This mission also uses the transiting method of detection, and Google’s code will likely be instrumental in distinguishing noise from potential alien worlds.
You can find the download for the code, and instructions on how to use it, on GitHub. If the thought of a machine-learning code on your computer makes you uneasy, you can also help sort data with your eyes, the old fashioned way.