Reconstruction Robustness on CIFAR10 (train)
0.0013MSE (Decoder-based)No_defense
Evaluation results
| Method | Links | ||
|---|---|---|---|
| No_defenseBackbone=VGG11, CEM integration=false2025.03 | 0.0013 | 0.0014 | |
| BottleneckBackbone=VGG11, CEM integration=false2025.03 | 0.0036 | 0.0039 | |
| Bottleneck+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0054 | 0.0076 | |
| DistCorrBackbone=VGG11, CEM integration=false2025.03 | 0.0074 | 0.0079 | |
| DistCorr+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.009 | 0.0094 | |
| DropoutBackbone=VGG11, CEM integration=false2025.03 | 0.0098 | 0.0099 | |
| Dropout+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0129 | 0.0132 | |
| ResSFLBackbone=VGG11, CEM integration=false2025.03 | 0.0146 | 0.0158 | |
| Average w/o CEMBackbone=VGG11, CEM integration=false2025.03 | 0.0148 | 0.0156 | |
| Noise_NopeekBackbone=VGG11, CEM integration=false2025.03 | 0.0152 | 0.0156 | |
| Noise_Nopeek+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0174 | 0.0173 | |
| Average w/ CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0183 | 0.0194 | |
| ResSFL+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0201 | 0.0229 | |
| PATROLBackbone=VGG11, CEM integration=false2025.03 | 0.0245 | 0.0257 | |
| Noise_ARLBackbone=VGG11, CEM integration=false2025.03 | 0.029 | 0.0304 | |
| PATROL+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.0304 | 0.0313 | |
| Noise_ARL+CEMBackbone=VGG11, CEM integration=true2025.03 | 0.033 | 0.0341 |