- June 28, 2021, 2:00 pm US/Central
Speaker: Rebecca Nevin, Harvard
Abstract: The difficulty of accurately identifying galaxy mergers, let alone classifying galaxy mergers of different mass ratios and stages, limits our understanding of the contribution of mergers to galaxy evolution. In recent years, machine learning has emerged as an exciting new avenue to increase the accuracy, completeness, and precision of merger identification. In this seminar, I will present two approaches I have taken to this challenge.
First, I’ll discuss my approach to identifying mergers using linear discriminant analysis applied to Sloan Digital Sky Survey (SDSS) imaging and to SDSS-IV’s MaNGA integral field spectroscopy (IFS) kinematic maps. I will describe how I create two complementary classification schemes from mock images and kinematic maps from N-body/hydrodynamics simulations of merging galaxies. I will discuss the strengths and limitations of this classification technique and my progress in applying the classification to the 1.3 million observed galaxies in the SDSS photometric survey.
Second, I will talk about a contrasting approach that uses the strength of deep convolutional neural networks (CNNs) to classify different types of mergers in mock JWST NIRCam imaging created from the Illustris TNG50 cosmological simulation. This project uses a novel combination of interpretability techniques coupled with CNNs to understand how the classification of galaxy mergers changes with galaxy properties such as redshift.
For more information, please contact Yu-Dai Tsai at ytsaiATfnal.gov.