Artificial intelligence helps SETI sleuths find more radio bursts from mystery source
Researchers used artificial intelligence to search through data from a radio source, capturing many more fast radio bursts than humans could. (Breakthrough Listen Illustration / Danielle Futselaar)
Researchers at Breakthrough Listen, a multimillion-dollar campaign to seek out signals from alien civilizations, still don’t know exactly what’s causing repeated bursts of radio waves from an distant galaxy — but thanks to artificial intelligence, they’re keeping closer tabs on the source, whatever it turns out to be.
A team led by Gerry Zhang, a graduate student at the University of California at Berkeley, developed a new type of machine-learning algorithm to comb through data collected a year ago during an observing campaign that used the Green Bank Telescope in West Virginia.
The campaign focused on a radio source known as FRB 121102, located in a dwarf galaxy sitting 3 billion light-years away in the constellation Auriga. Astronomers have observed plenty of fast radio bursts over the past decade, each lasting only a few milliseconds. Only FRB 121102 has been found to send out repeated bursts, however.
A number of theories have been proposed to explain the bursts, ranging from interactions involving magnetized neutron stars and black holes to deliberate signaling by advanced civilizations.
The researchers from Breakthrough Listen, one of several space projects backed by Russian-born billionaire Yuri Milner, added to the mystery last year when they conducted a six-hour listening session on Aug. 26, 2017.
Their initial analysis, using standard search algorithms, found that 21 bursts came during the first hour of monitoring. Then the radio source seemed to go silent.
But did it?
To double-check the data, Zhang and his team used machine-learning techniques originally developed for optimizing search results and classifying images. They trained a different kind of algorithm known as a convolutional neural network on the burst examples that were found using more traditional methods, and then set the AI algorithm loose on the complete data set to look for other bursts that might have been missed.
The AI algorithm found 72 more bursts, bringing the total number of fast radio bursts traced to FRB 121102 to about 300 since its discovery in 2012. A paper reporting the latest results has been accepted for publication in The Astrophysical Journal.
The updated analysis indicates that there’s no predictable pattern to the recurrence of the bursts, at least on timescales of more than 10 milliseconds. The new findings are likely to put new constraints on the various hypotheses that are being considered as the cause of the bursts — and thus contribute to solving the mystery.
“This work is only the beginning of using these powerful methods to find radio transients,” said Zhang. “We hope our success may inspire other serious endeavors in applying machine learning to radio astronomy.”
Andrew Siemion, director of the Berkeley SETI Research Center and principal investigator for Breakthrough Listen, said the methods could also address other challenges in the search for alien signals.
“Whether or not FRBs themselves eventually turn out to be signatures of extraterrestrial technology, Breakthrough Listen is helping to push the frontiers of a new and rapidly growing area of our understanding of the universe around us,” he said.
Could artificial intelligence find extraterrestrial intelligence? Stay tuned…
In addition to Zhang and Siemion, the authors of “Fast Radio Burst 121102 Pulse Detection and Periodicity: A Machine Learning Approach” include Vishal Gajjar, Griffin Foster, James Cordes, Casey Law and Yu Wang. Check out the Berkeley SETI Research Center’s website for more information, including the full pre-print manuscript.