AI4Research Seminar: Bayesian Machine Learning for Astronomical Object Classification

In anticipation of the forthcoming 4MOST telescope, we are developing a software pipeline for the automated classification of high-resolution spectra of both stars and galaxies. In order to quantify the prediction confidence we are aiming to design and train Bayesian Neural Networks (supervised), which work with parameter distributions instead of their point estimates. Simultaneously, we are exploring Variational Autoencoders (unsupervised) for effective anomaly scoring. This project is part of my ongoing PhD and it is work in progress.