Euclid Quick Data Release (Q1) From simulations to sky: Advancing machine

The Euclid mission will provide a revolutionary sample of over a hundred thousand galaxy-scale strong gravitational lenses. Identifying thesescientifically valuable objects in Euclid’s vast data set is impossible without the help of machine learning. However, the number of known stronglenses is typically too small to train robust supervised models using real data alone, and consequently machine learning approaches rely heavilyon simulated lenses for training. A well-known challenge is that machine learning models trained on one data domain often underperform whenapplied to a different domain. In the context of lens finding, this means that strong performance on simulated lenses does not necessarily translateinto equally good performance on real observations. With the discovery of around 500 strong lenses in Q1, we can now directly measure thisperformance gap and explore how incorporating real lenses improves results. We find that a network trained only on simulations recovers up to92% of simulated lenses with 100% purity but only achieves 50% completeness with 24% purity on real Q1 data. By augmenting training datawith real Q1 lenses and non-lenses, completeness improves by 25–30% in terms of the expected yield of discoverable lenses in Euclid ’s DataRelease 1 and the full Euclid Wide Survey. Roughly 20% of this improvement comes from the inclusion of real lenses in the training data, while5–10% comes from the added non-lenses from Q1. We show that the most effective lens-finding strategy for real-world performance combines thediversity of simulations with the fidelity of real lenses. This hybrid approach, enabled by the initial Q1 discoveries, establishes a clear methodologyfor maximising lens discoveries in future data releases from Euclid, and will also be applicable to other surveys such as LSST.Key words. Gravitational lensing: strong – Methods: data analysis – Surveys