Eight Technosignature Candidates Uncovered by a Deep-Learning Pipeline in Breakthrough Listen's 2021 SETI Search

To the point

Peter Xiangyuan Ma and colleagues used a semi-supervised deep‑learning pipeline that blends an autoencoder with a random forest to reanalyze about 150 terabytes of data from 820 nearby stars over two months using 12 GPUs, uncovering eight potential signs of extraterrestrial technology that classic methods missed, with help from Cherry Ng, Leandro Rizk and UC Berkeley’s SETI group and support from the Laidlaw Foundation and Breakthrough Listen, and they aim to scale the search to a million stars with MeerKAT and beyond.