Fig. 8From: Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantificationPerformance dependence on the number of reference compounds for the MLP-SNN models. For an easy comparison, the global minimum of each curve was shifted to 0 by 0.72 at n = 10 for the lipophilicity, 1.50 at n = 6 for the freesolv, and 0.79 at n = 7 for the ESOL datasetBack to article page