It is very exciting to hear that artificial intelligence found two new exoplanets. A machine learning method referred to as a neural network has actually found two brand new exoplanets within our own galaxy – according to software engineers from Google and NASA scientists. The amazing part of the story is how these new worlds were made know because of the awesome ability of artificial intelligence.
The Art of Finding Exoplanets
The discovery of new exoplanets—which are planets that reside outside our own solar system—is actually a very common event. The one tool that scientists have used to find them in the past has been using the Kepler Space Telescope – a tool that has previously found and confirmed around 2,525 exoplanets to date.
But the interesting wrinkle of finding these two exoplanets is because researchers were able to use an AI system to locate these two new worlds – which have been named as Kepler-80g and Kepler-90i. The 90i planet is particularly interesting to astronomers because makes the total number of planets orbiting this particular star to 8, which matches our own solar system. The temperature on 90i is believed to be quite warm – higher than 800 degrees Fahrenheit actually.
Using Neural Networks
Neural networks are those where the software actually learns from the data they accumulate. The fact is that these networks are as common as the discoveries of exoplanets. For instance, it is neural networks that are powering the language-translation function on Facebook, that power the FaceID process now available on the brand new iPhone X, and that power image recognition available for Google Photos. A prime example of how neural networks learn is by considering pictures of dogs and cats—if you present labeled pictures of dogs in a neural network, at a later date it ought to be quite able to discover new images of dogs just as it has been trained.
“Neural networks have been around for decades, but in recent years they have become tremendously successful in a wide variety of problems,” says Christopher Shallue, who is a senior software engineer at Google AI, during a recent NASA teleconference. “And now we’ve shown that neural networks can also identify planets in data collected by the Kepler Space Telescope.”
Astronomers will need great tools such as telescopes to look for exoplanets, and will also need artificial intelligence scientists to evaluate huge quantities of labeled data. In this particular instance, Shallue had trained this neural network by employing 15,000 labeled signals obtained from Kepler. These signals which are actually called light curves are measurements of just how the light of a star dips whenever an orbiting planet passes between Kepler’s eye and a star – this method is referred to as the transit method. Of these 15,000 total signals, around 3,500 of them were actually light curves from passing planets, and the remaining ones were false positives—which are light curves made by something other than an orbiting planet.
Eventually, Shallue and his associate, Andrew Vanderburg, who is a distinguished NASA Sagan postdoctoral fellow from the University of Texas at Austin, submitted the Kepler data to this neural network. It evaluated data from the systems of some 670 stars and focused on the weak signals that might potentially be coming from an undiscovered planet. And this was how they discovered these two new worlds.
“Machine learning really shines in situations where there is too much data for humans to examine for themselves,” Shallue stated.
While artificial intelligence methods and tools have been employed in the past for this type of research before, “this is the first time a neural network specifically has been used to identify a new expoplanet,” Shallue pointed out during a recent conference.