3D Galaxy Shape Prediction with Artificial Neural Networks  [slides]

Abstract

It has been nearly a century since the great debate was put to rest by Hubble, who definitively proved that many nebulae were truly "island universes" akin to our own Milky Way. Since then, millions of galaxies have been discovered, yet one relatively simple question remains open: what are the true three dimensional shapes of galaxies? The intangible nature of observational astronomy means that any one galaxy provides a single, two dimensional perspective, and it is this aspect of our field that makes this question so difficult to answer. In this talk, I will give an overview of the history of galaxy 3D shape research from statistical photometric studies up to more recent advances incorporating galaxy kinematics with integral field spectroscopy. I will then present critical tests we have performed of these recent methods that incorporate mock IFS from cosmological hydrodynamics simulations (from which the true 3D shapes are known). These tests have revealed striking biases in the recovered shapes of galaxies that have spurred us in a new direction: machine learning and neural networks. I will finish with our preliminary results in galaxy shape prediction with these modern techniques, which may provide an important step forward in this fundamental area of research.