Machine Learning for Stellar Spectra : Anomaly Detection in stellar spectra using Unsupervised Random ForestSpectral Analysis using Variational Autoencoders

Abstract: This thesis was carried out in two parts. The stellar spectral data was used from the Gaia-ESO survey. The data used was fromthe public archive as well as data received from Dr. Recio-Blanco at Observatoire Cote D'Azure. 1) I performed anomaly detection using unsupervised random forests, by applying the concept of weirdness scores to identify outliers. 2) Using spectral data along with physical parameters of objects in the galactic bulge of the Gaia-ESO survey, I built a variational autoencoder neural network to reconstruct stellar spectra and explore latent features learning physical parameters by themselves.