A new machine learning technique can be used to sift through massive datasets to discern exoplanets from false positives.
In recent years, researchers have tapped artificial intelligence (AI) for a host of applications across industries from mitigating regional wildfires to identifying potential COVID-19 treatments. Now, astronomers are using machine learning algorithms to search for planets beyond our solar system formally known as exoplanets. In a recent astronomical first, a machine learning algorithm confirmed dozens of distant exoplanets drifting through the cosmos.
“Almost 30% of the known planets to date have been validated using just one method, and that’s not ideal. Developing new methods for validation is desirable for that reason alone. But machine learning also lets us do it very quickly and prioritise candidates much faster,” said David Armstrong at the University of Warwick Department of Physics.
SEE: TechRepublic Premium editorial calendar: IT policies, checklists, toolkits, and research for download (TechRepublic Premium)
The Monthly Notices of the Royal Astronomical Society recently published a University of Warwick study detailing the use of machine learning to identify potential exoplanets. Using a selection of potential exoplanets, astronomers leveraged a machine learning-based process to detect real planets and identify false positives, “calculating the probability of each candidate to be a true planet.”
During typical exoplanet surveys, massive telescope datasets are analyzed to detect an exoplanet passing between the light source (a star) and the viewing telescope. This is known as transiting. A registered decrease in light could be indicative of an exoplanet passing through the distant star’s rays. However, other factors can also cause such a reduction in light such as background object interference, telescopic camera errors, or binary star systems; setting the cosmic stage for false-positive exoplanet identification. During the validation process, researchers can screen for false-positive occurrences.
Researchers from The Alan Turing Institute and Warwick’s Departments of Physics and Computer Science created the exoplanet algorithm capable of sifting out “real” exoplanets from samples of thousands of potential exoplanets. The algorithm was trained using two large data samples from the Kepler mission. This dataset included confirmed exoplanets as well as false positives.
SEE: Cleaning up space debris around our home planet (TechRepublic)
Next, the researchers fed the algorithm a Kepler dataset of unconfirmed exoplanets. This resulted in “fifty new confirmed planets and the first to be validated by machine learning,” per the release. In the past, machine learning algorithms have been used to rank potential exoplanet candidates, however, these techniques “never determined the probability that a candidate was a true planet by themselves, a required step for planet validation.”
“We still have to spend time training the algorithm, but once that is done it becomes much easier to apply it to future candidates. You can also incorporate new discoveries to progressively improve it,” Armstrong said.