In an effort to aid in the ongoing quest for extraterrestrial intelligence, scientists have created a machine-learning technique that they believe may help filter out interference and more effectively detect odd radio signals from space.
Since many years ago, radio telescopes have been utilized by SETI (Search for extraterrestrial intelligence) programs to find clear artificial signals originating from the firmament. However, interference from human technology makes this search difficult because it may result in false positive identifications that take a long time to remove from huge data sets.
The University of Toronto’s Peter Ma, a third-year physics and mathematics student, led research that utilized 115 million data snippets collected from observations made of 820 stars. The team’s deep learning models detected about 3 million signals of interest using the machine learning framework TensorFlow and the Python module Keras. The group was reduced to 20,515 relevant signals, which the authors said is more than 100 times less than earlier analysis of the same dataset.
According to an article published in Nature Astronomy, they then discovered eight previously unidentified signals of interest, despite the fact that subsequent observations have failed to successfully re-identify these targets.
To speed up SETI and other data-driven surveys, the authors advise using their method on other large datasets.
By examining the ‘technosignatures’ produced by their technological advancements, SETI seeks to provide an answer to this question by hunting for signs of intelligent life elsewhere in the galaxy. Due to the simplicity of radio signal propagation into interstellar space and the relatively efficient design of powerful radio transmitters and receivers, the majority of technosignature searches to date have been carried out at radiofrequencies, according to the scientists.
They claimed that “both scientists and the general public are acutely interested in the identification of an unambiguous technosignature” because it would prove the existence of extraterrestrial intelligence (ETI).
Other uses of machine learning in SETI, according to the authors, include a general signal classifier for observations made with the Allen Telescope Array and the Five-hundred-meter Aperture Spherical Radio Telescope, convolutional neural network-based algorithms for identifying radio frequency interference, and algorithms for anomaly detection.
One of the most well-known initiatives in the subject was SETI@home, which for more than 20 years transmitted radio telescope readings to volunteers’ home computers to search for possible indications of extraterrestrial life, but stopped sending data in 2020.
The Berkeley SETI Research Center, which oversees a number of related projects, has been in charge of the project since 1999. It has consumed around 1.5 million days of computer time. It failed to pinpoint intelligent extraterrestrial life, but it did show that volunteer computing initiatives might use Internet-connected computers as a useful analysis tool, outperforming the biggest supercomputers.
