In a new paper published in the journal Nature Astronomy, astronomers with Breakthrough Listen Initiative — the largest ever scientific research program aimed at finding evidence of alien civilizations — present a new machine learning-based method that they apply to more than 480 hours of data from the Robert C. Byrd Green Bank Telescope, observing 820 nearby stars. The method analyzed 115 million snippets of data, from which it identified around 3 million signals of interest. The authors then inspected the 20,515 signals and they identified 8 previously undetected signals of interest, although follow-up observations of these targets have not re-detected them.
“The key issue with any techno-signature search is looking through this huge haystack of signals to find the needle that might be a transmission from an alien world,” said Dr. Steve Croft, an astrophysicist at the University of California, Berkeley and a member of the Breakthrough Listen team. “The vast majority of the signals detected by our telescopes originate from our own technology — GPS satellites, mobile phones, and the like. Our algorithm gives us a more effective way to filter the haystack and find signals that have the characteristics we expect from techno-signatures.”
Classical techno-signature algorithms compare scans where the telescope is pointed at a target point on the sky with scans where the telescope moves to a nearby position, in order to identify signals that may be coming from only that specific point.
These techniques are highly effective. For example, they can successfully identify the Voyager 1 space probe, at a distance of 20 billion km, in observations with the Green Bank Telescope. But all of these algorithms struggle in crowded regions of the radio spectrum, where the challenge is akin to listening for a whisper in a crowded room.
The process developed by the team inserts simulated signals into real data, and trains an artificial intelligence algorithm known as an auto-encoder to learn their fundamental properties. The output from this process is fed into a second algorithm known as a random forest classifier, which learns to distinguish the candidate signals from the noisy background. “In 2021, our classical algorithms uncovered a signal of interest, denoted BLC1, in data from the Parkes telescope,” said Breakthrough Listen’s principal investigator Dr. Andrew Siemion, an astronomer at the University of California, Berkeley.
Read the full article at: www.sci.news
Abstract We introduce MusicLM, a model generating high-fidelity music from text descriptions such as “a calming violin melody backed by a distorted guitar riff”. MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music…