Mining for Strong Gravitational Lenses with Self-supervised Learning

Abstract We employ self-supervised representation learning to distill information from 76 million galaxy images the Dark Energy Spectroscopic Instrument Legacy Imaging Surveys’ Data Release 9. Targeting identification of new strong gravitational lens candidates, we first create a rapid similarity search tool discover lenses given only single labeled example. then show how training simple linear classifier on representations, requiring few minutes CPU, can automatically classify with great efficiency. present 1192 candidates that identified through brief visual campaign and release an interactive web-based top network predictions facilitate crowd-sourcing discovery additional other rare objects: github.com/georgestein/ssl-legacysurvey .