Finding quadruply imaged quasars with machine learning – I. Methods

Strongly lensed quadruply imaged quasars (quads) are extraordinary objects. They very rare in the sky -- only a few tens known to date and yet they provide unique information about wide range of topics, including expansion history composition Universe, distribution stars dark matter galaxies, host galaxies quasars, stellar initial mass function. Finding them astronomical images is classic "needle haystack" problem, as outnumbered by other (contaminant) sources many orders magnitude. To solve this we develop state-of-the-art deep learning methods train on realistic simulated quads based real taken from Dark Energy Survey, with source deflector models, chromatic effects microlensing. The performance best mixture objects excellent, yielding area under receiver operating curve 0.86 0.89. Recall close 100% down total magnitude i~21 indicating high completeness, while precision declines 85% 70% i~17-21. extremely fast: training 2 million samples takes 20 hours GPU machine, 10^8 multi-band cutouts can be evaluated per GPU-hour. speed method pave way apply it large sources, bypassing need for photometric pre-selection that likely major cause incompleteness current quads.