Table 6.
Misclassification detection performance of the BBP, MC-dropout, ConvCNPs-SDE, and SDE-Net models on CIFAR10 with MR = [0.1, 0.3, 0.5, 0.7, 0.9, RMR] and SVHN.
| CIFAR10 with MR | Model | TNR at TPR 95% |
AUROC | Detection Accuracy |
AUPR Succ |
AUPR Err |
|---|---|---|---|---|---|---|
| MR = 0.1 | BBP | 10.97 ± 0.66 | 58.75 ± 0.62 | 56.19 ± 0.70 | 25.69 ± 0.32 | 85.16 ± 0.77 |
| MC-dropout | 10.89 ± 0.60 | 67.19 ± 0.12 | 63.17 ± 0.23 | 61.69 ± 0.13 | 69.60 ± 0.24 | |
| ConvCNPs–SDE-Net | 27.47 ± 0.80 | 83.77 ± 0.18 | 77.33 ± 0.26 | 94.88 ± 0.12 | 54.12 ± 0.64 | |
| SDE-Net | 21.85 ± 1.60 | 72.51 ± 0.29 | 66.64 ± 0.35 | 44.85 ± 0.58 | 88.73 ± 0.21 | |
| MR = 0.3 | BBP | 9.84 ± 0.34 | 56.17 ± 0.24 | 54.57 ± 0.34 | 24.77 ± 0.14 | 84.32 ± 0.67 |
| MC-dropout | 6.47 ± 0.34 | 58.22 ± 0.26 | 56.90 ± 0.46 | 32.83 ± 0.66 | 78.92 ± 0.24 | |
| ConvCNPs–SDE-Net | 26.96 ± 0.83 | 82.70 ± 0.24 | 76.04 ± 0.37 | 93.83 ± 0.17 | 55.70 ± 0.50 | |
| SDE-Net | 8.61 ± 1.24 | 65.01 ± 0.68 | 63.04 ± 0.57 | 22.85 ± 0.88 | 91.82 ± 0.34 | |
| MR = 0.5 | BBP | 8.19 ± 0.02 | 55.73 ± 0.68 | 55.17 ± 0.70 | 23.32 ± 0.43 | 83.74 ± 0.02 |
| MC-dropout | 6.11 ± 0.14 | 57.18 ± 0.23 | 56.69 ± 0.22 | 24.32 ± 0.26 | 84.06 ± 0.34 | |
| ConvCNPs–SDE-Net | 23.55 ± 0.79 | 80.31 ± 0.16 | 73.78 ± 0.33 | 91.32 ± 0.05 | 57.15 ± 0.57 | |
| SDE-Net | 5.89 ± 1.22 | 53.64 ± 0.83 | 53.40 ± 0.19 | 13.31 ± 0.32 | 90.45 ± 0.33 | |
| MR = 0.7 | BBP | 9.00 ± 0.09 | 53.62 ± 0.66 | 53.73 ± 0.61 | 20.29 ± 1.04 | 84.15 ± 0.25 |
| MC-dropout | 5.89 ± 0.78 | 56.51 ± 0.58 | 56.08 ± 0.38 | 18.90 ± 0.22 | 87.23 ± 0.38 | |
| ConvCNPs–SDE-Net | 18.24 ± 0.69 | 75.85 ± 0.39 | 69.81 ± 0.40 | 84.37 ± 0.38 | 61.18 ± 1.22 | |
| SDE-Net | 5.05 ± 0.98 | 50.28 ± 0.51 | 51.17 ± 0.40 | 10.67 ± 0.44 | 89.87 ± 0.28 | |
| MR = 0.9 | BBP | 5.06 ± 0.13 | 53.45 ± 2.20 | 53.78 ± 1.96 | 16.89 ± 1.98 | 86.25 ± 0.14 |
| MC-dropout | 5.73 ± 0.78 | 55.93 ± 0.35 | 55.23 ± 0.28 | 14.60 ± 0.75 | 89.49 ± 0.23 | |
| ConvCNPs–SDE-Net | 11.53 ± 0.48 | 66.46 ± 0.45 | 62.43 ± 0.48 | 56.21 ± 0.53 | 74.77 ± 0.72 | |
| SDE-Net | 4.81 ± 0.88 | 50.93 ± 0.40 | 51.68 ± 0.36 | 11.26 ± 0.68 | 89.86 ± 0.25 | |
| MR = RMR | BBP | 8.12 ± 0.17 | 55.30 ± 1.18 | 53.92 ± 0.85 | 19.66 ± 0.15 | 86.11 ± 0.74 |
| MC-dropout | 6.66 ± 0.58 | 56.51 ± 0.88 | 55.59 ± 0.83 | 19.04 ± 0.55 | 87.32 ± 0.25 | |
| ConvCNPs–SDE-Net | 16.33 ± 1.57 | 75.42 ± 1.58 | 69.77 ± 1.39 | 80.73 ± 3.76 | 66.10 ± 2.00 | |
| SDE-Net | 5.34 ± 0.77 | 51.45 ± 0.94 | 51.74 ± 0.89 | 11.69 ± 0.56 | 90.21 ± 0.24 |