Accelerating SETI: Deep-Learning Search for Technosignatures of 820 Nearby Stars

A deep-learning search for the technosignatures of 820 nearby stars

The Search for Extraterrestrial Intelligence’s (SETI) goal is to determine the presence of technological life on other planets by analyzing their \”technosignatures\”. The narrowband Doppler radio signals are one of the theories for a technosignature.

The main challenge for SETI in radio is to develop a technique that can reject radio frequency interference from humans. We present here the most comprehensive deep learning based technosignature searches to date. The results returned 8 promising ETI signals that are of interest and should be re-observed as part of Breakthrough Listen.

The search includes 820 unique targets, which were observed by the Robert C. Byrd Green Bank Telescope. This totals over 480 hr on-sky data. We use a beta-Convolutional-Variational-Autoencoder, a novel algorithm that identifies technosignature candidates semi-unsupervised while maintaining a low false-positive rate. This new approach is a promising solution for accelerating SETI research and other transient astronomy in the age of data driven astronomy.


A Deep-learning Search For Technosignatures Of 820 Nearby Stars