Table 5 Privacy evaluation metrics between original and reconstructed images under the iDLG attack.
From: ALDP-FL for adaptive local differential privacy in federated learning
Dataset | Approach | MSE | PSNR | MAE | SSIM |
---|---|---|---|---|---|
MNIST | FedAvg | 0.0659 | 11.81 | 0.1609 | 0.5368 |
Fed-DPA | 0.2025 | 6.93 | 0.2949 | 0.1891 | |
GFL-ALDPA | 0.2213 | 6.36 | 0.3234 | 0.0974 | |
LDP-Fed | 0.2088 | 6.80 | 0.2972 | 0.1090 | |
ALDP-FL | 0.2226 | 6.5 | 0.3004 | 0.0525 | |
Fashion MNIST | FedAvg | 0.0341 | 14.67 | 0.1306 | 0.5085 |
Fed-DPA | 0.1279 | 8.93 | 0.2624 | 0.2494 | |
GFL-ALDPA | 0.1317 | 8.8 | 0.2594 | 0.1468 | |
LDP-Fed | 0.1492 | 8.26 | 0.2745 | 0.1354 | |
ALDP-FL | 0.1852 | 7.32 | 0.2827 | 0.0882 | |
CIFAR-10 | FedAvg | 0.0310 | 0.0753 | 15.08 | 0.4895 |
Fed-DPA | 0.1003 | 0.2320 | 9.9 | 0.1870 | |
GFL-ALDPA | 0.1214 | 0.2716 | 9.16 | 0.1231 | |
LDP-Fed | 0.1062 | 0.2489 | 9.74 | 0.1403 | |
ALDP-FL | 0.1836 | 0.3352 | 7.36 | 0.0808 |