Architecture of our CNN. The input layer reads in the flux information of the stellar spectra. It is followed by three pairs of convolution and max-pooling layers. The filter outputs from the third convolution and max-pooling pair are then flattened to serve as inputs for the dense layers. Three dense layers (with a dropout layer after each) interpret the spectral features, found by the convolution layers, into output labels. The outputs from a last dense layer are the values of our five stellar labels (atmospheric parameters and elemental abundances).