Speaker
Description
Unsupervised Machine Learning can create classifications by learning from data features and has the capacity to process the vast number of galaxies observed by contemporary large sky telescopes. We designed an Unsupervised ML technique that performs a dimensionality reduction with UMAP and then applies clustering methods such as GMM and HDBSCAN to the resulting 2D embedding. It was implemented on KiDS + VIKING galaxies with GAMA spectroscopy, selecting various mass limited samples and using k-corrected fluxes estimated with TOPz. We present a proof of concept: we test how to mitigate the UMAP stochasticity, examine the cluster's populations within narrow redshift bins, and study how well the resulting ML-based clusters conserve properties when compared with other classifications. We highlight the potential of the technique, its strengths, and some aspects for refinement prior to broader application.