Table 5.
Classification and OOD detection results of the BBP, MC-dropout, ConvCNPs–SDE, and SDE-Net models on ID dataset CIFAR10 with MR = [0.1, 0.3, 0.5, 0.7, 0.9, RMR] and OOD dataset SVHN.
| CIFAR10 with MR | Model | Classification Accuracy |
TNR at TPR 95% |
AUROC | Detection Accuracy |
AUPR In |
AUPR Out |
|---|---|---|---|---|---|---|---|
| MR = 0.1 | BBP | 19.83 ± 0.45 | 64.31 ± 3.35 | 92.57 ± 2.47 | 86.49 ± 2.45 | 84.29 ± 1.55 | 96.46 ± 1.42 |
| MC-dropout | 43.01 ± 0.35 | 3.67 ± 0.13 | 50.90 ± 0.15 | 52.50 ± 0.10 | 31.15 ± 0.16 | 71.50 ± 0.05 | |
| ConvCNPs–SDE-Net | 78.66 ± 0.20 | 4.46 ± 0.31 | 59.97 ± 0.16 | 59.33 ± 0.11 | 42.23 ± 0.15 | 75.54 ± 0.07 | |
| SDE-Net | 23.23 ± 0.25 | 1.57 ± 0.10 | 37.79 ± 0.13 | 50.36 ± 0.06 | 24.04 ± 0.14 | 63.41 ± 0.06 | |
| MR = 0.3 | BBP | 19.64 ± 0.15 | 59.53 ± 3.43 | 92.62 ± 1.89 | 86.50 ± 1.55 | 86.45 ± 2.45 | 96.36 ± 1.45 |
| MC-dropout | 24.91 ± 0.25 | 1.87 ± 0.15 | 40.81 ± 0.35 | 50.01 ± 0.26 | 23.62 ± 0.12 | 65.83 ± 0.25 | |
| ConvCNPs–SDE-Net | 76.46 ± 0.20 | 4.25 ± 0.37 | 58.31 ± 0.30 | 57.89 ± 0.20 | 40.55 ± 0.18 | 74.71 ± 0.22 | |
| SDE-Net | 13.11 ± 2.89 | 4.61 ± 0.25 | 52.28 ± 0.34 | 52.55 ± 0.15 | 31.46 ± 0.12 | 72.74 ± 0.30 | |
| MR = 0.5 | BBP | 19.44 ± 0.25 | 46.25 ± 4.26 | 87.53 ± 3.72 | 81.05 ± 3.72 | 77.70 ± 3.38 | 93.95 ± 1.68 |
| MC-dropout | 18.49 ± 0.19 | 2.22 ± 0.15 | 40.79 ± 0.19 | 50.00 ± 0.15 | 23.42 ± 0.17 | 66.10 ± 0.16 | |
| ConvCNPs–SDE-Net | 72.21 ± 0.26 | 3.57 ± 0.19 | 55.08 ± 0.28 | 55.35 ± 0.13 | 37.36 ± 0.15 | 72.98 ± 0.22 | |
| SDE-Net | 10.33 ± 0.06 | 4.97 ± 0.08 | 50.74 ± 0.26 | 50.95 ± 0.17 | 29.18 ± 0.21 | 72.33 ± 0.13 | |
| MR = 0.7 | BBP | 18.39 ± 0.52 | 21.77 ± 4.09 | 77.28 ± 3.11 | 70.66 ± 2.47 | 67.15 ± 2.63 | 87.66 ± 2.29 |
| MC-dropout | 14.61 ± 0.18 | 2.76 ± 0.28 | 42.66 ± 0.22 | 50.00 ± 0.13 | 24.26 ± 0.23 | 67.36 ± 0.14 | |
| ConvCNPs–SDE-Net | 62.47 ± 0.49 | 2.50 ± 0.06 | 48.70 ± 0.28 | 52.08 ± 0.05 | 31.64 ± 0.14 | 69.60 ± 0.13 | |
| SDE-Net | 10.10 ± 0.08 | 4.91 ± 0.28 | 49.86 ± 0.50 | 50.27 ± 0.19 | 27.85 ± 0.31 | 72.05 ± 0.34 | |
| MR = 0.9 | BBP | 14.60 ± 0.67 | 1.10 ± 0.40 | 55.31 ± 2.84 | 65.55 ± 1.23 | 54.23 ± 2.43 | 68.97 ± 0.93 |
| MC-dropout | 12.01 ± 0.16 | 2.81 ± 0.11 | 43.53 ± 0.06 | 50.00 ± 0.07 | 24.45 ± 0.04 | 67.91 ± 0.16 | |
| ConvCNPs–SDE-Net | 36.48 ± 0.43 | 1.29 ± 0.01 | 37.86 ± 0.39 | 50.49 ± 0.05 | 23.48 ± 0.30 | 64.32 ± 0.18 | |
| SDE-Net | 10.18 ± 0.06 | 4.91 ± 0.18 | 49.40 ± 0.06 | 50.19 ± 0.12 | 27.02 ± 0.05 | 71.92 ± 0.11 | |
| MR = RMR | BBP | 16.37 ± 0.38 | 15.04 ± 5.27 | 70.19 ± 1.22 | 65.72 ± 0.64 | 59.93 ± 0.21 | 83.36 ± 1.87 |
| MC-dropout | 14.78 ± 0.08 | 2.31 ± 0.25 | 41.87 ± 0.35 | 50.00 ± 0.15 | 23.72 ± 0.23 | 66.80 ± 0.33 | |
| ConvCNPs–SDE-Net | 56.04 ± 4.08 | 1.98 ± 0.24 | 46.79 ± 2.04 | 51.80 ± 0.48 | 30.41 ± 1.90 | 68.45 ± 0.98 | |
| SDE-Net | 10.18 ± 0.05 | 4.82 ± 0.26 | 50.01 ± 0.36 | 50.42 ± 0.22 | 28.00 ± 0.28 | 72.03 ± 0.27 |