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

This dataset contains strong gravitational lens (SGL) candidates identified in imaging data from the Kilo-Degree Survey Data Release 4 (KiDS DR4). The search is performed over the KiDS DR4 bright galaxies, comprising approximately 3.7 million galaxies, using a combination of self-supervised feature extraction, an iterative active learning framework (Astronomaly: PROTEGE), and nearest-neighbour similarity search in feature space. All candidates in this dataset have been visually inspected by three experts, and assigned a confidence grade based on their likelihood of being a SGL. Grading scheme Grade A: high-confidence strong lens candidates Grade B: likely lens candidates requiring further confirmation Grade C: low-confidence candidates showing some lens-like features The released catalogue includes the candidates found in this work, with grades reflecting the outcome of the visual inspection stage. Note: these grades reflect visual inspection by three experts and are intended to provide an internally consistent relative confidence ranking for this work, not a definitive discovery-grade classification of each system; Rojas et al. (2023) recommend a minimum of six inspectors for a classification that is stable at the individual-object level. Columns KiDS_ID: KiDS Unique object identifier RA: Right Ascension (degrees) DEC: Declination (degrees) Grade: Expert-assigned confidence grade (A, B, or C) Method: Stage(s) through which the candidate was identified (AL, NN, or AL+NN) Recovered_known: Whether the candidate is a previously known SGL recovered by this search