Conditional invertible neural networks for enhanced analysis of young low-mass stars

Determining basic stellar parameters such as effective temperature and extinction is an essential step in analysing stellar spectra because the accuracy of these parameters significantly influences measurements of accretion properties and the interpretation of stellar evolution and planet formation of young stellar systems. As various physical properties are reflected on the observed spectra, it is important to measure the essential properties considering the correlations between them. In this talk, I will introduce a novel method that couples a conditional invertible neural network (cINN) architecture and Phoenix stellar atmosphere models to estimate stellar parameters (e.g., effective temperature, surface gravity, extinction, and veiling factor) of young stars from the optical stellar spectra with intermediate spectral resolution. This tool is currently designed for young low-mass stars observed with VLT/MUSE. The cINN architecture allows us to obtain a multi-dimensional posterior distribution of the parameters in a very time-efficient way without additional calculations, helping us understand the degeneracy within physical properties. I will introduce my recent two works, where I first tested the applicability of this methodology on class III template stars (Kang et al. 2023) and the application of the method on ~2000 young stars in Trumpler 14 in the Carina Nebula Complex (Kang et al.submitted).