- Galaxy Morphology -
There are over one hundred billion galaxies in the observable Universe — each one is unique, and their structures/morphology harbour an imprint of their own formation history. In order to unveil the true relation between the structures and evolutionary history of galaxies as well as understand the mechanisms that transfrom galaxy structures, one requires an "accurate" classification of galaxy's morphologies. This can be intepreted in two perspectives: (1) correctly classifying the morphology of a galaxy based on current visual classification schemes such as Hubble sequence (Cheng 2020a; Cheng 2021b; Cheng 2023); (2) creating a new categorisation scheme that are physically meaningful for galaxy morphology (Cheng 2021a).
My works cover both perspectives, but I am aiming to revolutionaise the convetionally visual classification scheme of galaxy morphologies by a novel objective classification scheme without human involvement using unsupervised machine learning techniques. For example, in my pioneering work (Cheng 2021a), I developed an unsupervised machine learning pipeline composed of a vector-quantised autoencoder and a hierarchical clustering algorithm to categorise monochromatic SDSS r-band galaxy images. The categorisation process is fully numerical without any human involvement, and the machine classess show distinctive physical properties including galaxy colour from each other. Leverging the fact that a machine can determine precise numerical decision boundaries on a much higher dimensional feature space than humanly possible, one can expect that the machine categorisation can lead to a more physically meaningful scheme with the capability of distinguising some visually indistinguishable morphological features.