“Not all discoveries come from new observations,” says Pete Worden, in a comment referring to original thinking as applied to an existing dataset. Worden is executive director of the Breakthrough Initiatives program, which includes Breakthrough Listen, an ambitious attempt to use SETI techniques to search for signs of technological activity in the universe. Note that last word: The targets Breakthrough Listen examines do extend to about one million stars in the stellar neighborhood, but they also go well outside the Milky Way, with 100 galaxies being studied in a range of radio and optical bands.

A major and sometimes neglected aspect of SETI as it is reported in the media is the fact that such careful observation can turn up highly useful astronomical information unrelated to any extraterrestrial technologies. Worden’s comment underlines the fact that we are generating vast data archives as our multiplying space- and ground-based instruments continue to scan the heavens at various wavelengths. It is not inconceivable that the signature of a distant civilization or a novel astrophysical process is buried deep within data that is years or decades old.

The case in point this morning comes through Breakthrough Listen’s observations of the Fast Radio Burst (FRB) 121102. FRBs seize the attention because they are a recent and puzzling detection. The first, the so-called Lorimer Burst (FRB 010724) was found no more than eleven years ago. FRBs are also highly unusual, appearing as radio pulses of extremely short duration, usually milliseconds. They are believed to originate in distant galaxies and one of them, FRB 121102, stands out because unlike all others thus far detected, it sends out repeat bursts.

Image: The FRB in question. This is a Gemini composite image of the field around FRB 121102 (indicated). The dwarf host galaxy was imaged, and spectroscopy performed, using the Gemini Multi-Object Spectrograph (GMOS) on the Gemini North telescope on Maunakea in Hawai’i. Data were obtained on October 24-25 and November 2, 2016, before the subsequent Breakthrough Listen observations at Green Bank. Credit: Gemini Observatory/AURA/NSF/NRC.

FRB 121102 is prolific in repeat bursts, as it turns out, and this is where Worden’s words resonate. Earlier studies have shown that the source is located in a galaxy some 3 billion light years away. Breakthrough Listen’s observing team at the University of California, Berkeley SETI Research Center went to work on this ‘repeater’ on August 26, 2017, finding a total of 21 bursts in data acquired at the Green Bank Telescope in West Virginia. All of these turned up within a single hour, which implies periods of intense activity. For more, see New Activity of Repeating FRB 121102.

Working with the same dataset, UC Berkeley graduate student Gerry Zhang and collaborators have now introduced a new machine learning algorithm into the mix, subjecting the same dataset (about 400 TB) to techniques not dissimilar to those used by Internet technology companies. Zhang and team’s ‘convolutional neural network’ was trained to recognize FRBs in the same way that Vishal Gajjar applied when he and his own colleagues analyzed the data in 2017. The algorithm was then applied to the raw data to see if anything new could be acquired.

From the paper:

…we present a re-analysis of the C-band observation by Breakthrough Listen on August 26, 2017 with convolutional neural networks. Recent rapid development of deep learning, and in particular, convolutional neural networks (CNN; Krizhevsky et al. 2012; Simonyan & Zisserman 2014; He et al. 2015; Szegedy et al. 2014) has enabled revolutionary improvements to signal classification, pattern recognition in all fields of data science such as, but not limited to computer image processing, medicine, and autonomous driving. In this work, we present the first successful application of deep learning to direct detection of fast radio transient signals in raw spectrogram data.

The result: 72 new detections of pulses from FRB 121102. That takes the cumulative total from this dataset to 93 pulses in five hours of observation, including 45 pulses within the first 30 minutes. As to periodicity, the work shows that the pulses are not received in a regular pattern. Using the same methods, Zhang and team go on to explore trends in pulse fluence, pulse detection rate, and pulse frequency structure. The breadth of the analysis is made possible by the sheer number of pulses, the highest detected from a single observation, and the use of Zhang and team’s neural network technology, which can surely be adapted to other datasets.

“Gerry’s work is exciting not just because it helps us understand the dynamic behavior of FRBs in more detail,” says Berkeley SETI Research Center Director and Breakthrough Listen Principal Investigator Dr. Andrew Siemion, “but also because of the promise it shows for using machine learning to detect signals missed by classical algorithms.”

So can we be sure we are dealing with a natural astrophysical phenomenon, or is there any possibility of a technology behind this repeating FRB? The data here do not give us the answer, but it’s intriguing to speculate about galactic SETI in light of our recent discussion of the work of Yuki Nishino and Naoki Seto (Kyoto University), who have suggested that natural phenomena such as the merger of a binary neutron star system like GW170817 could be used as a marker to flag a message from a civilization attempting to announce its presence to the cosmos.

For that matter, are all FRBs the same? We’ve only found the one repeater thus far, but 10⁴ are now thought to occur in our sky on a daily basis. Avi Loeb and Minasvi Lingam (Harvard University) have worked out an FRB rate for each galaxy within 100 Gpc³ of about 10^-5 per day, if indeed we are dealing with natural sources like gamma ray bursts or neutron star mergers. And as they’ve noted in a recent paper, one of these going off in our own galaxy, as would be expected about every 300 years, would resolve the question of FRB origins, a doubtless spectacular event if occurring nearby.

The Gajjar et al. paper reporting on the 2017 repeat detections is “Highest-frequency detection of FRB 121102 at 4-8 GHz using the Breakthrough Listen Digital Backend at the Green Bank Telescope,” accepted at The Astrophysical Journal (preprint). The Zhang et al. paper covering the subsequent new analysis is “Fast Radio Burst 121102 Pulse Detection and Periodicity: A Machine Learning Approach,” accepted for publication in The Astrophysical Journal, with further details available here. Avi Loeb and Minasvi Lingam’s paper on technological possibilities in FRBs is “Fast Radio Bursts from Extragalactic Light Sails,” The Astrophysical Journal Letters Vol. 837, No. 2 (March 8, 2017). Abstract / preprint.