| Abstract [eng] |
The new generation of large-scale spectroscopic surveys aims to observe millions of stars across the whole Milky Way galaxy. To analyze this enormous number of stellar spectra, new techniques are being developed. This work will show convolutional neural networks (CNNs) can parameterize thousands of stellar spectra within a few minutes. The precision and accuracy of the CNN results are good, and they can be used to investigate key properties of the Milky Way galaxy. This thesis lays out the details of the CNN architecture, its training and testing phase, and the estimation of the internal uncertainties. Spectra from the Gaia-ESO survey are then used to train a CNN to predict the stellar parameters effective temperature and surface gravity, and the atmospheric abundances of lithium, magnesium, aluminum, and iron. The results are used to investigate the properties of several globular clusters, as well as the thin and thick disks of the Milky Way galaxy. By re-analyzing the spectra from the Gaia-ESO survey, 31 previously unidentified lithium-rich giants have been found. |
