Fig. 1. U-Sleep overall architecture.

U-Sleep is a fully convolutional deep neural network. It takes as input a sequence of length L of 30-second sleep epochs and it outputs the predicted sleep stage for each epoch. We slightly modified the original figure (see Figure 2: Model architecture in¹¹) reporting the additional SaBN layers exploited in the conditional learning procedure (see subsection Conditional learning). Please refer to¹¹ for details on the U-Sleep model architecture and training parameters.