Preview

PDF, English Download (11MB) | Terms of use

Abstract

Star formation is one of the most fundamental subjects in astronomy where astronomers have been seeking answers to key questions: how efficiently stars form and how newly born stars affect their surroundings. Our understanding of star formation relies mostly on the observations of star-forming regions. However, it is a non-trivial task to interpret the observations because diverse physical processes are non-linearly coupled so the observational data are highly degenerate. Additionally, the ever-expanding volume of observational data in recent days necessitates a new method that analyses large amounts of data more quickly and effectively.

In this thesis, we introduce deep learning-based tools we have developed to efficiently and effectively interpret massive data of observed star-forming regions. We adopt the conditional invertible neural network (cINN) architecture specialised in solving the inverse problem of degenerate systems. We introduce the cINNs developed for cloud-scale observations and cINNs for individual star-scale observations. Our networks are very useful tools that can consistently and quickly analyse large amounts of data. We evaluate the performance of the networks, demonstrating that our networks predict physical properties accurately and precisely.