The CASSL pipeline analyzes strong lensing in JWST data. First, the code forecasts galaxy-scale strong lenses for JWST, including Einstein radii down to θ_E=0.02″ systems (generalizable to other telescopes and selection criteria). The code also simulates images of galaxy-scale strong lenses for JWST in the range 0.02″<θ_E<1.5″. These simulations use the VELA hydrodynamical simulations as very realistic lensed source galaxies. CASSL employs empirically motivated parameter distributions to generate realistic datasets of simulated strong lensing images.

The RUN Pipeline trains, evaluates, and tests two convolutional neural network models used to detect strong lensing in images. The code first classifies strong lenses from non-lenses with a ResNet model, and further detects the locations of small Einstein radius systems in cutout images down to θ_E∼0.03″ with a U-Net model. After both models are trained, the software analyzes the performance of the pipeline on validation and testing data.