The Gaia-ESO Survey: Chemical evolution of Mg and Al in the Milky Way with Machine-Learning

Abstract

We aim to prepare the machine-learning ground for the next generation of spectroscopic surveys, such as 4MOST and WEAVE. Our goal is to show that convolutional neural networks can predict accurate stellar labels from relevant spectral features in a physically meaningful way. We built a neural network and trained it on GIRAFFE spectra with associated stellar labels from the sixth internal Gaia-ESO data release. Our neural network predicts the atmospheric parameters Teff and log(g) as well as the chemical abundances [Mg/Fe], [Al/Fe], and [Fe/H] for 30115 stellar spectra. The scatter of predictions from eight slightly different network models shows a high internal precision of the network results: 24 K for Teff, 0.03 for log(g), 0.02 dex for [Mg/Fe], 0.03 dex for [Al/Fe], and 0.02 dex for [Fe/H]. The network gradients reveal that the network is inferring the labels in a physically meaningful way from spectral features. Validation with benchmark stars and several scientific applications confirm that our network predictions are accurate for individual stars and recover the properties of different stellar populations in the Milky Way galaxy. Such a study provides very good insights into the application of machine-learning for the spectral analysis of large-scale spectroscopic surveys, such as WEAVE and 4MIDABLE-LR and -HR (4MOST Milky Way disk and bulge low- and high-resolution). The community will have to put a substantial effort into building proactive training sets for machine-learning methods to minimize the possible systematics.