Table 4.
Misclassification detection performance of the BBP, MC-dropout, ConvCNPs–SDE, and SDE-Net models on MNIST with MR = [0.1, 0.3, 0.5, 0.7, 0.9, RMR] and SVHN.
| MNIST with MR | Model | TNR at TPR 95% |
AUROC | Detection Accuracy |
AUPR Succ |
AUPR Err |
|---|---|---|---|---|---|---|
| MR = 0.1 | BBP | 40.78 ± 2.34 | 89.55 ± 0.92 | 82.52 ± 1.25 | 98.79 ± 0.22 | 44.14 ± 3.22 |
| MC-dropout | 86.52 ± 1.22 | 95.59 ± 0.88 | 91.76 ± 0.68 | 99.93 ± 0.01 | 36.68 ± 1.88 | |
| ConvCNPs–SDE-Net | 92.44 ± 1.06 | 98.15 ± 0.12 | 95.01 ± 0.50 | 99.99 ± 0.00 | 31.49 ± 5.90 | |
| SDE-Net | 84.43 ± 3.22 | 96.75 ± 0.60 | 92.59 ± 0.76 | 99.96 ± 0.01 | 31.32 ± 2.89 | |
| MR = 0.3 | BBP | 24.90 ± 2.26 | 82.15 ± 2.82 | 75.45 ± 1.32 | 94.61 ± 2.34 | 54.26 ± 2.04 |
| MC-dropout | 60.92 ± 1.24 | 91.73 ± 0.41 | 54.25 ± 1.51 | 99.21 ± 0.01 | 46.33 ± 1.31 | |
| ConvCNPs–SDE-Net | 85.80 ± 3.63 | 97.22 ± 0.42 | 93.00 ± 1.02 | 99.97 ± 0.01 | 35.81 ± 4.71 | |
| SDE-Net | 54.25 ± 1.51 | 91.11 ± 0.46 | 84.01 ± 0.93 | 99.44 ± 0.04 | 40.00 ± 1.75 | |
| MR = 0.5 | BBP | 12.17 ± 2.16 | 72.48 ± 1.26 | 68.08 ± 1.52 | 78.42 ± 4.25 | 68.32 ± 2.55 |
| MC-dropout | 27.28 ± 0.33 | 82.15 ± 0.33 | 76.10 ± 0.23 | 92.85 ± 0.13 | 56.85 ± 0.43 | |
| ConvCNPs–SDE-Net | 72.53 ± 3.26 | 95.07 ± 0.51 | 89.36 ± 0.39 | 99.86 ± 0.01 | 36.57 ± 4.60 | |
| SDE-Net | 33.07 ± 0.53 | 83.79 ± 0.35 | 76.12 ± 0.24 | 95.33 ± 0.14 | 56.45 ± 0.83 | |
| MR = 0.7 | BBP | 10.17 ± 1.27 | 65.10 ± 1.21 | 65.09 ± 0.82 | 51.87 ± 0.57 | 82.44 ± 1.22 |
| MC-dropout | 15.67 ± 0.87 | 73.82 ± 0.35 | 68.55 ± 0.32 | 72.23 ± 0.37 | 72.72 ± 0.33 | |
| ConvCNPs–SDE-Net | 43.58 ± 1.67 | 89.10 ± 0.16 | 82.10 ± 0.41 | 98.44 ± 0.07 | 48.40 ± 1.01 | |
| SDE-Net | 28.88 ± 0.97 | 76.12 ± 0.50 | 68.84 ± 0.47 | 75.88 ± 0.34 | 75.61 ± 0.76 | |
| MR = 0.9 | BBP | 9.21 ± 1.22 | 69.39 ± 1.45 | 64.41 ± 0.62 | 26.28 ± 1.46 | 94.27 ± 0.32 |
| MC-dropout | 7.45 ± 0.32 | 59.26 ± 0.43 | 57.29 ± 0.12 | 25.06 ± 0.22 | 86.16 ± 0.22 | |
| ConvCNPs–SDE-Net | 15.73 ± 0.67 | 70.02 ± 0.70 | 64.73 ± 0.81 | 60.84 ± 1.19 | 77.94 ± 0.59 | |
| SDE-Net | 7.14 ± 0.42 | 60.91 ± 0.41 | 59.25 ± 0.30 | 22.04 ± 0.79 | 88.79 ± 0.25 | |
| MR = RMR | BBP | 10.85 ± 2.43 | 70.43 ± 5.52 | 67.83 ± 3.42 | 68.10 ± 6.22 | 80.86 ± 3.23 |
| MC-dropout | 34.81 ± 13.32 | 78.00 ± 5.40 | 72.60 ± 4.40 | 77.47 ± 9.45 | 76.60 ± 4.21 | |
| ConvCNPs–SDE-Net | 36.95 ± 11.08 | 88.19 ± 2.74 | 81.51 ± 2.59 | 94.38 ± 3.33 | 71.57 ± 6.18 | |
| SDE-Net | 35.11 ± 15.40 | 83.44 ± 6.02 | 76.25 ± 4.80 | 80.15 ± 15.44 | 83.92 ± 3.18 |