Table 6 Performance metrics for UQ models for uniform elongation and total elongation, given as average value ± standard deviation (in subscript).
From: Empirical validation of size effects in sub-sized tensile specimens for nuclear structural materials
Model | Uniform elongation | Total elongation | ||||
|---|---|---|---|---|---|---|
R2 ↑ | RMSE ↓ | Coverage (%) ↑ | R2 ↑ | RMSE ↓ | Coverage (%) ↑ | |
Quantile regression | 0.872± 0.05 | 7.01± 2.3 | 70.6± 22.9 | 0.837± 0.13 | 8.1± 3.2 | 84.2± 9.8 |
Natural grad boosting | 0.961± 0.02 | 3.89± 1.4 | 94.1± 5.3 | 0.839± 0.11 | 8.2± 2.8 | 90.5± 11.7 |
Gaussian process regression | 0.909± 0.05 | 5.8± 1.6 | 91.8± 8.8 | 0.819± 0.11 | 8.8± 2.4 | 89.5± 6.9 |
Deep ensemble | 0.866± 0.11 | 6.59± 2.8 | 33.5± 12.8 | 0.835± 0.09 | 8.5± 2.0 | 46.8± 4.2 |
MC dropout | 0.882± 0.05 | 6.71± 1.8 | 42.7± 17.8 | 0.786± 0.10 | 9.8± 2.1 | 45.8± 9.9 |
Bayesian NNs—VarInf | 0.885± 0.05 | 4.89± 1.3 | 88.3± 5.3 | 0.820± 0.10 | 8.7± 2.5 | 76.8± 9.9 |
Bayesian NNs—MCMC | 0.928± 0.05 | 5.08± 2.0 | 92.9± 11.4 | 0.884± 0.07 | 7.1± 1.8 | 95.3± 4.2 |
