Neural Network Verification on MNIST 1000 images
79.3Verified Accuracyβ-CROWN FSB
Evaluation results
| Method | Links | ||
|---|---|---|---|
| β-CROWN FSBModel=ConvBig2021.03 | 79.3 | 3.1 | |
| PRIMAModel=ConvBig2021.03 | 77.5 | 11 | |
| β-CROWN FSBModel=MLP 5 × 2002021.03 | 77.4 | 86 | |
| OptC2VModel=ConvBig2021.03 | 77.1 | 102 | |
| kPolyModel=ConvBig2021.03 | 73.6 | 40 | |
| β-CROWN FSBModel=MLP 8 × 2002021.03 | 73.5 | 95 | |
| β-CROWN FSBModel=ConvSmall2021.03 | 72.7 | 7 | |
| CROWN/DeepPolyModel=ConvBig2021.03 | 71.1 | 21 | |
| β-CROWN FSBModel=MLP 5 × 100‡2021.03 | 69.9 | 102 | |
| PRIMAModel=MLP 5 × 2002021.03 | 69 | 224 | |
| PRIMAModel=MLP 8 × 2002021.03 | 62.4 | 395 | |
| β-CROWN FSBModel=MLP 8 × 1002021.03 | 62 | 103 | |
| OptC2VModel=MLP 5 × 2002021.03 | 60.1 | 403 | |
| PRIMAModel=ConvSmall2021.03 | 59.8 | 42 | |
| kPolyModel=MLP 5 × 2002021.03 | 57.4 | 187 | |
| OptC2VModel=MLP 8 × 2002021.03 | 52.8 | 3,451 | |
| PRIMAModel=MLP 5 × 100‡2021.03 | 51 | 159 | |
| kPolyModel=MLP 8 × 2002021.03 | 50.6 | 464 | |
| kPolyModel=MLP 5 × 100‡2021.03 | 44.1 | 307 | |
| OptC2VModel=ConvSmall2021.03 | 43.6 | 55 | |
| OptC2VModel=MLP 5 × 100‡2021.03 | 42.9 | 137 | |
| PRIMAModel=MLP 8 × 1002021.03 | 42.8 | 301 | |
| OptC2VModel=MLP 8 × 1002021.03 | 38.4 | 759 | |
| kPolyModel=MLP 8 × 1002021.03 | 36.9 | 171 | |
| kPolyModel=ConvSmall2021.03 | 34.7 | 477 | |
| CROWN/DeepPolyModel=MLP 5 × 2002021.03 | 29.2 | 2.4 | |
| CROWN/DeepPolyModel=MLP 8 × 2002021.03 | 25.9 | 5.6 | |
| CROWN/DeepPolyModel=MLP 8 × 1002021.03 | 18.2 | 1.4 | |
| CROWN/DeepPolyModel=MLP 5 × 100‡2021.03 | 16 | 0.7 | |
| CROWN/DeepPolyModel=ConvSmall2021.03 | 15.8 | 3 |