When Spectral Modeling Meets Convolutional Networks: A Method for Discovering Reionization-era Lensed Quasars in Multi-band Imaging Data by Irham Taufik Andika et al. on Tuesday 29 NovemberOver the last two decades, around three hundred quasars have been discoveredat $z\gtrsim6$, yet only one was identified as being strong-gravitationallylensed. We explore a new approach, enlarging the permitted spectral parameterspace while introducing a new spatial geometry veto criterion, implemented viaimage-based deep learning. We made the first application of this approach in asystematic search for reionization-era lensed quasars, using data from the DarkEnergy Survey, the Visible and Infrared Survey Telescope for AstronomyHemisphere Survey, and the Wide-field Infrared Survey Explorer. Our searchmethod consists of two main parts: (i) pre-selection of the candidates based ontheir spectral energy distributions (SEDs) using catalog-level photometry and(ii) relative probabilities calculation of being a lens or some contaminantutilizing a convolutional neural network (CNN) classification. The trainingdatasets are constructed by painting deflected point-source lights over actualgalaxy images to generate realistic galaxy-quasar lens models, optimized tofind systems with small image separations, i.e., Einstein radii of$\theta_\mathrm{E} \leq 1$ arcsec. Visual inspection is then performed forsources with CNN scores of $P_\mathrm{lens} > 0.1$, which led us to obtain 36newly-selected lens candidates, waiting for spectroscopic confirmation. Thesefindings show that automated SED modeling and deep learning pipelines,supported by modest human input, are a promising route for detecting stronglenses from large catalogs that can overcome the veto limitations of primarilydropout-based SED selection approaches.arXiv: http://arxiv.org/abs/http://arxiv.org/abs/2211.14543v1