Fig. 1From: Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantificationIllustration of the three model designs. a The traditional model aims to predict the true value based on a single input. b In the delta (Δ) model the difference vector between two samples is used as input and the model predicts their value difference. c The Siamese model makes use of the two networks of sharing weights, and the difference between the hidden states is fed through an additional network that predicts the value difference of the two samplesBack to article page