The promise of self-supervised and active learning for Strong Lens discovery: Astronomaly applied to KiDS

Abstract

Strong gravitational lenses (SGLs) are rare systems whose discovery currently relies primarily on supervised machine learning methods trained on large simulated datasets. We present the first application of Astronomaly:PROTEGE to SGL discovery, demonstrating that a human-in-the-loop active learning framework can efficiently identify lenses in large imaging surveys without the need for simulated training data. We consider a sample of 3.7 million bright galaxies from the Kilo-Degree Survey (KiDS)…