Studying AGN Host Galaxies From Hyper-Suprime-Cam Using Convolutional Neural Networks
DescriptionStudying the host galaxy properties of active galactic nuclei (AGN) is crucial in studying black-hole galaxy co-evolution. With this knowledge, we could decipher the interrelationships between characteristic properties of the galaxy, like its morphology, and properties of the AGN, like the rate of accretion onto the black hole. The AGN’s light often overwhelms the host galaxy light, making it difficult to study galactic characteristics. Point Spread Function Generative Adversarial Network (PSFGAN), a neural network designed to remove AGN point source light in large datasets of galaxies, has been trained and tested using Sloan Digital Sky Survey (SDSS) images. Providing PSFGAN with higher resolution images should improve performance to have a higher accuracy of host galaxy image reconstruction. A new dataset imaging the Stripe 82 equatorial field has been released from the Hyper Suprime-Cam (HSC) project; it has both a deeper flux limit and higher spatial resolution. We modified PSFGAN to accommodate HSC images for training and testing, producing higher accuracy models. Galaxy Morphology Network (GaMorNet) then determined the morphology of the HSC AGN hosting galaxies. Currently GaMorNet has been tested using a classification methodology, where galaxies are classified into discrete groupings (disk or bulge). We are constructing a regression style GaMorNet that will output bulge to disk ratio, which is a quantitative way to parametrize galactic morphology. First using PSFGAN to reconstruct HSC AGN host galaxy images, and then using GaMorNet for determining their accurate morphologies will be a step forward in the investigation of AGN and galaxy co-evolution.