Type: Poster
Session: Virtual (Rubin Research Bytes)
Author: Priyanka Gawade

This work will also be presented as a flash talk in the Machine Learning and Artificial Intelligence session.

Abstract: Gravitational lensing is an important tool to probe the matter distribution in the Universe. Analysing strongly lensed systems, like a galaxy lensed by another foreground galaxy or a cluster, where we can get multiple distorted images of the background galaxy, provides us crucial information about the distribution of the total matter in the foreground lens. Conventional lens modelling techniques used to analyse the strongly lensed systems are in general time and resource consuming, require sophisticated lensing codes and expertise in the physics involved. To inspect the tens of thousands of lenses expected from upcoming advanced imaging surveys like Rubin LSST, we need fast and automated techniques. In our work, we construct a simple convolutional neural network to rapidly predict the lens mass model parameters of lenses from the ground-based imaging surveys and demonstrate this on existing data from HSC. We focus on the lens mass model parameters such as Einstein radius, axis ratio and position angle of the major axis of the mass distribution. The network is first trained on simulated data where we have added lensed features on top of real data. The optimised network is then use to predict the lens parameters of galaxy-scale lenses from HSC. We find good agreement with the Einstein radii of these lenses, and a poorer performance on the axis ratio and position angles of the mass distribution, when compared to conventional modelling methods. We present comparison between our predictions of 10 high quality HSC lenses with conventional modelling methods from the literature.