In this talk, I will introduce a novel machine learning framework for estimating the Bayesian posteriors of morphological parameters for arbitrarily large numbers of galaxies. The Galaxy Morphology Posterior Estimation Network (GaMPEN) estimates values and uncertainties for a galaxy's bulge-to-total light ratio, effective radius, and flux. GaMPEN also contains a Spatial Transformer Network (STN) that automatically crops input galaxy frames to an optimal size before determining their morphology. The STN will be crucial in applying GaMPEN to new survey data with no radius measurements. GaMPEN is the first machine learning framework for determining posterior distributions of morphological parameters and is one of the first applications of an STN to astronomy.
Using GaMPEN, we have determined the full Bayesian posteriors for the morphological parameters of ~ 8 Million galaxies in the Hyper Suprime-Cam (HSC) Wide Survey with z < 0.75 and m < 23. Using a novel technique of first training on simulations and then transfer-learning on real data, we have been able to train GaMPEN with < 1% of our dataset. By analyzing a sub-sample using light-profile fitting, we have shown that the posterior distributions predicted by GaMPEN are accurate and well-calibrated for all three parameters with < 5% deviation. This is one of the largest morphological catalogs of galaxy parameters currently available and is allowing us to use morphology to probe galaxy evolution for lower mass galaxies with extremely high statistical significance.
Join Zoom Meeting https://carnegiescience.zoom.us/j/91681612163