In the light of more and more new instrumentation to get a deeper insight into the universe, tons of data are collected. While traditional machine-learning methods have been used in processing stellar spectral data, such large new datasets are better handled with Deep Learning (DL) techniques. In this work, we present a Deep Convolutional Neural Network (CNN) approach to derive fundamental stellar parameters (effective temperature, surface gravity, metallicity and rotational velocity) from high-resolution high signal-to-noise ratio spectra. We construct an individual CNN architecture for each of the four parameters and train them on synthetic PHOENIX-ACES spectra. After that, we apply the trained networks to the observed spectra of 50 M dwarfs observed with CARMENES. The CARMENES spectrograph, installed on the 3.5 m telescope at the Calar Alto Observatory (Spain) has two channels, covering the visible (0.52 to 0.96 µm, R = 96,400) and near-infrared (0.96 to 1.71 µm, R = 80,600) spectral ranges. We compare our results to literature values, and demonstrate that our method can be used for stellar parameter determination without the need of having a huge sample of stellar spectra with known parameters, because our networks can be trained on synthetic spectra. Introducing Deep Transfer Learning (DTL) in our approach allows us to transfer external knowledge about the stellar parameters (e.g., from interferometry) to our training set and therefore improve our results compared to literature.
{"references":["Abadi et al. (2015), https://github.com/tensorflow/tensorflow","Abraham et al. (2018), 10.1093/mnras/sty627","Anthony and Bartlett (2009), ISBN9780521118620","Antoniadis-Karnavas et al. (2020), 10.1051/0004-6361/201937194","Birky et al. (2020), 10.3847/1538-4357/ab7004","Boyajian et al. (2008), 10.1086/589554","Boyajian et al. (2012), 10.1088/0004-637x/757/2/112","Bressan et al. (2012), 10.1111/j.1365-2966.2012.21948.x","Chen et al. (2014), 10.1093/mnras/stu1605","Chen et al. (2015), 10.1093/mnras/stv1281","Chollet (2015), \\url{https://github.com/keras-team/keras}","Czesla et al. (2019), ascl:1906.010","Dieleman et al. (2015), 10.1093/mnras/stv632","Dittmann et al. (2016), 10.3847/0004-637x/818/2/153","Fabbro et al. (2018), 10.1093/mnras/stx3298","Figueira et al. (2016), 10.1051/0004-6361/201526900","Freedman and Diaconis (1981), 10.1007/BF01025868","Gaidos and Mann (2014), 10.1088/0004-637X/791/1/54","Gong and Ordieres-Meré (2016), 10.1016/j.envsoft.2016.06.020","Hauschildt (1992), 10.1016/0022-4073(92)90105-D","Hauschildt (1993), 10.1016/0022-4073(93)90080-2","Hauschildt and Baron (1999), arXiv:astro-ph/9808182","Hon et al. (2017), 10.1093/mnras/stx1174","Houdebine et al. (2019), 10.3847/1538-3881/ab23fe","Husser et al. (2013), 10.1051/0004-6361/201219058","LeCun et al. (2015), 10.1038/nature14539","Mahabal et al. (2017), arXiv:1709.06257","Maldonado et al. (2015), 10.1051/0004-6361/201525797","Mann et al. (2015), 10.1088/0004-637X/804/1/64","McInnes et al. (2018), 10.21105/joss.00861","Montes et al. (2018), 10.1093/mnras/sty1295","Nagel et al. (2021), submitted to A&A","Ness et al. (2015), 10.1088/0004-637X/808/1/16","Parzen (1962), 10.1214/aoms/1177704472","Passegger et al. (2018), 10.1051/0004-6361/201732312","Passegger et al. (2019), 10.1051/0004-6361/201935679","Passegger et al. (2020), 10.1051/0004-6361/202038787","Quirrenbach et al. (2018), 10.1117/12.2313689","Reiners et al. (2018), 10.1051/0004-6361/201732054","Rojas-Ayala et al. (2012), 10.1088/0004-637X/748/2/93","Rosenblatt (1956), 10.1214/aoms/1177728190","Sarro et al. (2018), 10.1093/mnras/sty165","Schmidhuber (2015), 10.1016/j.neunet.2014.09.003","Scott (1979), 10.1093/biomet/66.3.605","Tang et al. (2014), 10.1093/mnras/stu2029","Terrien et al. (2015), 10.1088/2041-8205/802/1/l10","Vilalta (2018), 10.1088/1742-6596/1085/5/052014","vonBraun et al. (2014), 10.1093/mnras/stt2360","Whitten et al. (2019), 10.1051/0004-6361/201833368","Wu et al. (2018), 10.1093/mnras/sty2646","Zechmeister et al. (2018), 10.1051/0004-6361/201731483","Zhang et al. (1998), 10.1016/S0169-2070(97)00044-7","Zheng et al. (2018), 10.3390/s18072146"]}