AstroAI Workshop 2026
Aleksandr Belotserkovtsev
Super-Resolving Euclid to Find Strong Gravitational Lenses
Presenter: Aleksandr Belotserkovtsev (Harvard University)
Title: Super-Resolving Euclid to Find Strong Gravitational Lenses
Date/Time: Monday, June 15, 4:00 PM - 5:30 PM
Abstract: Strong gravitational lensing yields systems that are valuable because they allow for studies of distant magnified galaxies that would otherwise be too faint to see, the mass distribution of the foreground galaxy or cluster, dark matter, and even the geometry of the Universe. In particular, lenses with small Einstein radii (the angular distance between the lens and the source) are especially valuable, since they constrain the dark matter halo profile. However, when this radius is too small, the lens cannot be visually separated from the source in low spatial resolution, wide-area surveys, which causes underrepresentation of lenses with small Einstein radii in studies. Recent advances in super-resolution techniques offer a way to recover this missing population by improving the identification of strong gravitational lenses in Euclid surveys, the first wide-area space-based optical survey. We build on POLISH (Connor et al. 2022) and POLISH++ (Wu et al. 2026) architectures, originally developed for radio astronomy: they are deep convolutional neural networks with a global residual connection. We adapt POLISH to increase the resolution of Euclid images to the level of higher-quality Hubble images, training on synthetic Euclid-like images built from simulated galaxies, with Hubble imaging as the high-resolution reference. Our preliminary results show that, applied to real Euclid data, the model both super-resolves and denoises, revealing structure that is blurred or buried in the originals — a promising first step toward recovering the missing small-radius lenses.