Laidlaw Proposal

Abstract: The goal for The Search for Extra-terrestrial Intelligence (SETI) is to find evidence of technological signals beyond Earth. Radiofrequency SETI searches are often conducted in environments characterized by the high volume of interference and a vast quantity of unlabeled data. The main problem in Radio SETI is developing a generalizable technique in rejecting human radio frequency interference (RFI) and help narrow the searches for technosignatures. In this research project, we present a β−Convolutional Variational Autoencoder with an embedded discriminator combined with spectral clustering to help classify between RFI and SETI candidates in a semi-unsupervised fashion. We develop and evaluate the performance of this algorithm on a test bench of synthetic SETI events created with the data collected from the Breakthrough Listen project at the GreenBank telescope. Finally, this approach is being executed on the real GBT L-band dataset of over 1327 observational targets