Oral Presentation

Unravelling the merger histories of galaxies with deep learning

Author(s): Connor Bottrell

Presenter: Connor Bottrell (Kavli IPMU)

Mergers between galaxies can be drivers of morphological transformation and various physical phenomena in galaxies, including star-formation, black-hole accretion, and chemical redistribution. These effects are seen clearly among galaxies that are currently interacting (pairs) -- which can be selected with high purity spectroscopically with correctable completeness. Galaxies in the merger remnant phase (post-mergers) exhibit some of the strongest changes, but are more elusive because identification must rely on the remnant properties alone. I will present results from my recent paper combining images and stellar kinematics to identify merger remnants using deep learning (arXiv:2201.03579). I find that kinematics are not the smoking-gun for improving remnant classification purity and that high posterior purity remains a significant challenge for remnant identification using imaging, kinematics, or both. However, an alternative approach which treats all galaxies as merger remnants and reframes the deep learning task as regression rather than classification yields exciting results and promising applications.