Title:Neural network prediction of model parameters for strong lensing samples from Hyper Suprime-Cam Survey

Abstract:Galaxies that cause the strong gravitational lensing of background galaxies provide us crucial information about the distribution of matter around them. Traditional modelling methods that analyse such strong lenses are both time and resource consuming, require sophisticated lensing codes and modelling expertise. To study the large lens population expected from imaging surveys such as LSST, we need fast and automated analysis methods. In this work, we build and train a simple convolutional neural network with an aim to rapidly predict model parameters of gravitational lenses. We focus on the most important lens mass model parameters, namely, the Einstein radius, the axis ratio and the position angle of the major axis of the mass distribution. The network is trained on a variety of simulated data with an increasing degree of realism and shows satisfactory performance on simulated test data. The trained network is then applied to the real sample of galaxy-scale candidate lenses from the Subaru HSC, a precursor survey to LSST. Unlike the simulated lenses, we do not have the ground truth for the real lenses. Therefore, we have compared our predictions with those from YattaLens, a lens modelling pipeline. Additionally, we also compare the parameter predictions for 10 HSC lenses that were also studied by other conventional modelling methods. These comparisons show a fair quantitative agreement on the Einstein radius, although the axis ratio and the position angle from the network as well as the individual modelling methods, seem to have systematic uncertainties beyond the quoted errors.