Table 3 EA: UQ evaluation metrics for the evidential uncertainties of an evidential NN model trained on the vertical IP dataset. LS flex: UQ evaluation metrics for the LS flex uncertainties of one of the five feed forward NN models trained with different random splits of the vertical IP training data. Random seed 42 is used in both cases
UQ | \(R^2\) | a | b (eV) | \(\rho _{rank}\) | \(\rho _{rank}^{sim}\) | \(A_{mis}\) | NLL | NLL\(^{sim}\) |
|---|---|---|---|---|---|---|---|---|
EV | 0.83 | 0.27 | 0.35 | 0.41 | 0.56(0.05) | 0.07 | 0.90 | 0.89(0.04) |
LS flex | 0.74 | 0.85 | 0.06 | 0.22 | 0.27(0.06) | 0.02 | 0.87 | 0.90(0.04) |