Disentangling stellar atmospheric parameters in astronomical spectra using Generative Adversarial Neural Networks

We present an encoder-decoder architecture with adversarial training, a modification of thetraditional autoencoder, whose aim is to unravel in the latent space the desired parameters ofthe rest of the information contained in astronomical spectra. Concretely, we have developedan algorithm for the chemical disentanglement in APOGEE and Gaia/RVS spectroscopicsurveys. In this poster, we show the results obtained for deriving stellar parameters Teff,log, [M/H] and [α/Fe] using Gaia/RVS spectra.