Exploring the Morphologies of High Redshift Galaxies with Machine Learning

TL;DR

This paper investigates the use of unsupervised machine learning to classify high redshift galaxy morphologies, addressing observational biases and linking clusters to physical properties.

Contribution

It introduces an improved unsupervised feature extraction method that accounts for observational biases, enhancing galaxy classification at high redshift.

Findings

Reduced number of clusters after bias correction

Clusters correlate with physical galaxy properties

Method improves classification accuracy

Abstract

The morphology of a galaxy has been shown to encode the evolutionary history and correlates strongly with physical properties such as stellar mass, star formation rates and past merger events. While the majority of galaxies in the local universe can be classified on the Hubble sequence, little is known about the different types of galaxies we observe at high redshift. The irregular morphology of these galaxies makes visual classifications difficult, and with the future of astronomy consisting of many "Big Data" surveys we need an efficient, and unbiased classification system in place. In this work we explore the use of unsupervised machine learning techniques to preform feature extraction from galaxy images to separate high redshift galaxies into different morphological types based on the machine learning clusters. We expand on previous work by addressing observation biases such as the orientation, apparent size of the galaxies and noise before extracting features, thus reducing the number of clusters and forcing the network to learn meaningful features. We then compare the extracted clusters' physical properties, to investigate the separation between the groups.