Fig. 5: Comparison between TL and DL for different training set sizes. | Nature Communications