The artificial intelligence helps to break through the audit Find new fast radio bursts

Green Bank Telescope (Credit: AUI / GBO)

Breakthrough listen – the astronomical program looking for signs of a smart life in the universe – applied machine learning techniques to detect 72 new fast radio bursts (FRB) emanating from the "repeater" FRB 121102 .

Fast radio bursts, or FRBs, are light pulses of radio emission of a few milliseconds, which are thought to come from distant galaxies. Most FRBs were observed during a single explosion. On the other hand, FRB 121102 is the only one to date to emit repeated bursts, of which 21 detected during Breakthrough listen observations made in 2017 with the Green Bank Telescope (GBT) in West Virginia¹.

The source and mechanism of the FRB are still mysterious. Previous studies have shown that the bursts of 121102 emanated from a galaxy 3 billion light-years away from Earth, but the nature of the object that emits them is still unknown. Theories range from highly magnetized neutron stars, dynamited by gas flows near a supermassive black hole, to suggestions that bursting properties match the signatures of technologies developed by an advanced civilization.

"Not all of the discoveries come from new observations," said Pete Worden, executive director of revolutionary initiatives, including: Listening"In this case, it was an intelligent and original reflection applied to an existing dataset. He advanced our knowledge of one of the most enticing mysteries of astronomy. "

In search of a deeper understanding of this intriguing object, the Listening Scientific team at the University of California, Berkeley SETI Research Center² observed FRB 121102 for five hours on August 26, 2017, using Breakthrough Listen digital instrumentation at GBT. Combining through 400 TB of data, they reported (in an article led by Vishal Gajjar, a postdoctoral researcher at Berkeley SETI, recently accepted for publication in the Astrophysical Journal).³) a total of 21 bursts. All were seen in one hour, suggesting that the source alternates between periods of rest and frenetic activity.

Today, Gerry Zhang, a UC Berkeley Ph.D. student, and his collaborators have developed a powerful new machine-learning algorithm and re-analyzed the 2017 GBT dataset, uncovering a further 72 bursts. detected at the origin. Zhang's team used some of the techniques used by Internet technology companies to optimize search results and rank images. They formed an algorithm known as a convolutional neural network to recognize the bursts found with the classical search method used by Gajjar and his collaborators, and then broke away from the 400 TB dataset to find the gusts missed by Gajjar and his collaborators. classic approach.

The results helped to establish new constraints on the frequency of FRB pulses 121102, suggesting that the pulses are not received with a regular pattern (at least if the period of this pattern is greater than about 10 milliseconds). Just as pulse patterns of pulsars have helped astronomers limit computer models to the extreme physical conditions of these objects, the new measurements of the FRBs will help determine what are the powers of these enigmatic sources.

"This work is only the beginning of using these powerful methods to detect radio transients," said Gerry Zhang. "We hope our success could inspire further serious efforts in the application of machine learning to radio astronomy."

"Gerry's work is exciting not only because it helps us to better understand the dynamic behavior of FRBs," said Berkeley SETI Research Center Director and Senior Research Scientist Andrew Siemion, "but also to the promise of using the machine. learn to detect missed signals by conventional algorithms. "

That the FRBs themselves turn out to be signatures of extraterrestrial technologies, Breakthrough Listen helps to push the limits of a new and growing field of our understanding of the universe around us.

The new findings are described in an article (Zhang et al., 2018) accepted for publication in the Astrophysical Journal. A pre-print of the paper, an animation of detected bursts, the data and the code used in the analysis, as well as other details on the observations are available on seti.berkeley.edu/frb-machine.