Table 4 Performance comparison via CNN networks.
From: Based on model randomization and adaptive defense for federated learning schemes
Dataset | MNIST | CIFAR-10 | Fashion-MNIST | ||||||
|---|---|---|---|---|---|---|---|---|---|
Number of DCA | \(N_p\)=5 | \(N_p\)=10 | \(N_p\)=15 | \(N_p\)=5 | \(N_p\)=10 | \(N_p\)=15 | \(N_p\)=5 | \(N_p\)=10 | \(N_p\)=15 |
Krum47 | 85.1% | 84.6% | 81.5% | 30.4% | 28.5% | 27.2% | 68.3% | 67.6% | 65.3% |
Trimmed mean47 | 75.7% | 70.9% | 66.8% | 26.5% | 25.3% | 25.6% | 62.2% | 60.0% | 57.6% |
Median47 | 68.3% | 65.2% | 63.9% | 27.3% | 24.5% | 22.3% | 60.3% | 59.3% | 58.2% |
HSFL | 87.3% | 82.3% | 79.1% | 31.2% | 27.5% | 26.5% | 75.4% | 73.2% | 69.0% |
DPRLDS48 | 88.3% | 85.6% | 85.0% | 30.2% | 30.5% | 29.3% | 69.2% | 69.2% | 68.3% |
RMCS49 | 87.3% | 84.6% | 83.0% | 29.2% | 28.5% | 28.3% | 69.2% | 68.9% | 66.3% |
PPHSFL | 89.2% | 87.3% | 86.6% | 33.1% | 32.1% | 32.5% | 72.1% | 69.3% | 68.7% |
Improve\(\uparrow\) | 1.9% | 2.7% | 3.6% | 3.9% | 4.1% | 4.2% | 2.9% | 1.4% | 2.4% |
