TABLE 1.

Distributions of the training, validation, and test sets in fine-tuning l₁, l₂, and l₃. Before fine-tuning, the model was first pre-trained on the source data. While train/validation/test set distributions were 80%/10%/10% in pre-training and training-from-scratch, the train data in fine-tuning systematically lowered. Set sizes of fine-tuning l₃ have been used in analyses where the training strategy was compared to pre-training and training-from-scratch. Distributions in absolute subject numbers with percentages in parentheses.

Name	Train/validation/test set distributions in OSA and insomnia datasets (N = 60)	Train/validation/test set distribution for RBD datasets (N = 22)
Fine-tuning l1	6/6/6 (10%/10%/10%)	6/2/2 (30%/10%/10%)
Fine-tuning l2	12/6/6 (20%/10%/10%)	12/2/2 (55%/10%/10%)
Fine-tuning l3	24/6/6 (40%/10%/10%)	18/2/2 (80%/10%/10%)