fig3

Figure 3. (A) Training and test set MSE of ANN models under FPS in interpretable space (FPS-A, blue), regression space (FPS-B, cyan), casually selected space (FPS-C, red), and RS (gray) across different training sizes; (B) Distribution of ΔMSE for different sampling methods across various training sizes, with FPS-A and FPS-B exhibiting negative ΔMSE between training sizes of 0.2 and 0.5; (C) Loss curves for training and test sets under different sampling methods at a training size of 0.3. Loss curves for FPS-A and FPS-B remain stable with increasing training epochs, indicating lower variability, with A and B showing better performance than C. MSE: Mean squared error; ANN: artificial neural network; FPS: farthest point sampling; RS: random sampling.