Image Classification on MNIST (test) (Accuracy, Robustness, and Performance)
99.3Clean AccuracyERM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ERMrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 99.3 | 14.3 | 1.46 | 0.007 | 3.47 | |
| DALErho (ρ)=1.0, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 99.1 | 97.7 | 94.5 | 0.053 | 25.4 | |
| TRADESrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98.9 | 96.5 | 94 | 0.055 | 25.8 | |
| MARTrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98.9 | 96.1 | 93.5 | 0.043 | 20.4 | |
| FGSMrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98.3 | 98.1 | 13 | 0.011 | 5.48 | |
| PGDrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98.1 | 95.5 | 93.1 | 0.039 | 18.2 | |
| ALPrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98.1 | 95.5 | 92.5 | 0.048 | 22 | |
| CLPrho (ρ)=-, Architecture=Four-layer CNN, Optimizer=Adadelta2021.10 | 98 | 95.4 | 92.2 | 0.047 | 21.9 |