AI Now Advancing the Discovery of Alien Intelligence
Led by an Undergraduate Student from the University of Toronto, ML is now being used to discover artifacts likely originating from intelligent sources
Created on February 10|Last edited on February 10
Comment
An undergraduate student at the University of Toronto, Peter Ma, has improved our ability to discover signs of alien life.
While just in highschool, Ma began an effort that ultimately catalyzed the development of a specialized ML technology for aiding the search for extraterrestrial intelligence (SETI).
Ma partnered with university of Toronto and UC Berkely researchers to build a system capable of deciphering between signals of radio interference and signals that could indicate evidence of alien intelligence, which are called technosignatures. These technosignatures tend to be narrowband signals with a nonzero drift rate, which are unlikely to come from a non-intelligent source. The work was recently published in Nature Astronomy, and achieve state of the art performance.
The authors introduce new a model architecture which involves an encoder that utilizes convolution kernels to preprocess input signals, and then finally feeds the encoded signal into a random forest classifier.
As there is not any verified ground truth data that could be used for training, the authors rely on synthetic data to train their classifier. The objective of the network is to ultimately predict three different classes, which include false data with no ETI signal, true data with ETI signals, and true data with ETI signals and radio interference.
As the evaluation for a task like locating extraterrestrial intelligence is not exactly standardized such as object detection on the Imagenet dataset, a key metric for performance is the models ability to discover signals that have yet to be discovered. The model was able to find 8 new signals indicating high probability of extraterrestrial significance, and also returns fewer false positives compared to previous methods.
Future Application
Although current telescope and rocket technologies may be limiting the application of a system that locates extraterrestrial intelligence, the task serves as a unique problem for researchers to develop novel methods that could be applied for other tasks. Specifically, the methods used to preprocess the signals have high overlap with many audio processing techniques, which have immediate application in areas like audio synthesis. AI is continuing to show nearly unlimited application, and has shown that not even the sky is the limit.
The Paper
Original Article
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.