An Ai Algorithm-based Method Developed by NYUAD Researchers Offers a New Approach for Detecting Moons Outside Our Solar System

Novel computing system avoids challenges such as stellar variability (variations in the brightness of stars) and photon noise (conflicting signals) experienced by other techniques

Abu Dhabi, UAE – August 5, 2020:  Over the past two decades, more than 3,500 exoplanets (planets that orbit stars outside our solar system) have been discovered, but none of the moons that orbit them. A new AI algorithm-based method created by researchers at the NYU Abu Dhabi Center for Space Science promises to avoid some of the challenges of finding moons that live outside the solar system, also known as exomoons, offering new hope for locating these extrasolar satellites.

Most known exoplanets were located using the transit method, which utilizes space telescopes to observe a star over a period of time and generate a photometric light curve. If an exoplanet is present, it would be detected as it orbited around the star and blocked light, altering the light curve.  Detecting exomoons using this method has proven to be a challenge as stellar variability (variations in the brightness of stars) and photon noise (conflicting signals) create false positive identifications.

A team of researchers led by Rasha Alshehhi, a postdoctoral associate at the NYU Abu Dhabi Center for Space Science, has designed a new AI algorithm-based method to more accurately detect exomoons. Using numerical experiments, this method outperforms classical methods used in exoplanet science to identify moon-like signals and avoids false positives caused by stellar variability and instrumental noise.

In the paper titled Detection of exomoons in simulated light curves with a regularized convolutional neural network, published in the journal Astronomy & Astrophysics, Alshehhi and her colleagues proposed a 1D convolutional neural network (CNN) to accurately identify possible exomoon transits around exoplanets. CNNs are the latest machine learning methods which do not require the preprocessing stage (i.e. feature engineering). In the study, the CNNs were used to specifically classify photometric transit light curves that would display a difference if an exomoon was present. As this is an AI algorithm-based method, the researchers were able to remove unwanted variations in data (stellar variability and instrumental noise) that other methods have not.

Several methods have utilized the transit system to attempt to discover exomoons by monitoring light curves for any changes caused by their presence, but the CNN is faster and more effective at predicting a moon-like signal.

“This new method, which uses the latest machine learning technique, is a more promising approach for analyzing real transit light curves in the future to discover exomoons,” said Alshehhi. “Our hope is that it will open up new opportunities to finally identify and study exomoons, which have been a source of wonder – and frustration – for some time.”

Caption: Sketch of orbital configuration for the numerical simulations of transit light curves. The planet and the moon orbit their common center of mass C.