Machine learning techniques for astrophysical modelling and photometric redshift estimation of quasars in optical sky surveys

Abstract

Machine learning techniques are utilised in several areas of astrophysical research today. This dissertation addresses the application of ML techniques to two classes of problems in astrophysics, namely, the analysis of individual astronomical phenomena over time and the automated, simultaneous analysis of thousands of objects in large optical sky surveys. Specifically investigated are (1) techniques to approximate the precise orbits of the satellites of Jupiter and Saturn given Earth-based observations as well as (2) techniques to quickly estimate the distances of quasars observed in the Sloan Digital Sky Survey. Learning methods considered include genetic algorithms, particle swarm optimisation, artificial neural networks, and radial basis function networks. The first part of this dissertation demonstrates that GAs and PSO can both be efficiently used to model functions that are highly non-linear in several dimensions. It is subsequently demonstrated in the second part that ANNs and RBFNs can be used as effective predictors of spectroscopic redshift given accurate photometry, especially in combination with other learning-based approaches described in the literature. Careful application of these and other ML techniques to problems in astronomy and astrophysics will contribute to a better understanding of stellar evolution, binary star systems, cosmology, and the large-scale structure of the universe.