Feature Inversion on CIFAR-100
69.8SSIMUpConvNet
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UpConvNetBackbone=ResNet-502026.04 | 69.8 | 62.8 | — | — | — | |
| Adjoint InversionBackbone=ResNet-502026.04 | 67.6 | 85.4 | — | — | — | |
| UpConvNetBackbone=DenseNet-1212026.04 | 67.4 | 75.4 | — | — | — | |
| UpConvNetBackbone=ResNet-182026.04 | 63.4 | 73.8 | — | — | — | |
| Adjoint InversionBackbone=DenseNet-1212026.04 | 40.4 | 94.6 | — | — | — | |
| Adjoint InversionBackbone=ResNet-182026.04 | 36.8 | 93.8 | — | — | — | |
| MahendranBackbone=ResNet-502026.04 | 13.8 | 136.4 | — | — | — | |
| RobustBackbone=DenseNet-1212026.04 | 10.1 | 143 | — | — | — | |
| MahendranBackbone=DenseNet-1212026.04 | 9.8 | 142.7 | — | — | — | |
| RobustBackbone=ResNet-502026.04 | 8.7 | 142.7 | — | — | — | |
| RobustBackbone=ResNet-182026.04 | 7.4 | 165.4 | — | — | — | |
| MahendranBackbone=ResNet-182026.04 | 7.3 | 164 | — | — | — | |
| Vanilla GradientBackbone=ResNet-502026.04 | 4.5 | 140.8 | — | — | — | |
| Vanilla GradientBackbone=DenseNet-1212026.04 | 4.1 | 144.6 | — | — | — | |
| Vanilla GradientBackbone=ResNet-182026.04 | 2 | 173 | — | — | — | |
| NL-SMEE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.8642 | 0.1503 | 25.883 | 0.9255 | 0.0029 | |
| SMEE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.7286 | 0.2409 | 21.5807 | 0.8563 | 0.0484 | |
| NL-SMEE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.7225 | 0.2756 | 21.0483 | 0.8431 | 0.0069 | |
| DGGIE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.6784 | 0.2787 | 19.9262 | 0.8248 | 0.0521 | |
| SMEE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.5815 | 0.3654 | 18.4793 | 0.7807 | 0.061 | |
| DGGIE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.5761 | 0.3801 | 18.598 | 0.7721 | 0.0592 | |
| GI-NASE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.4448 | 0.4379 | 15.9999 | 0.7315 | 0.1586 | |
| IGE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.384 | 0.4606 | 15.0098 | 0.719 | 0.1385 | |
| FEDLEAKE (local epoch)=20, N (local data size)=50, B (batch size)=10, T (total local steps)=1002025.09 | 0.343 | 0.4788 | 14.5956 | 0.7028 | 0.14 | |
| GI-NASE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.2456 | 0.5089 | 12.6997 | 0.6621 | 0.2914 | |
| IGE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.2213 | 0.5174 | 12.2821 | 0.664 | 0.2681 | |
| FEDLEAKE (local epoch)=50, N (local data size)=50, B (batch size)=10, T (total local steps)=2502025.09 | 0.204 | 0.519 | 12.2429 | 0.6605 | 0.2575 |