Table 2 Convergence accuracy with different parameters.
From: Based on model randomization and adaptive defense for federated learning schemes
Dataset | MNIST | CIFAR-10 | Fashion-MNIST | ||||||
|---|---|---|---|---|---|---|---|---|---|
Round | t=50 | t=100 | t=150 | t=50 | t=100 | t=150 | t=50 | t=100 | t=150 |
(1) \(\gamma\)=4, K=1, \(M_s\)=10 | 80.7% | 88.1% | 89.3% | 24.6% | 29.3% | 30.5% | 52.1% | 66.3% | 67.8% |
(2) \(\gamma\)=4, K=2, \(M_s\)=20 | 80.2% | 87.7% | 89.1% | 23.3% | 28.2% | 29.6% | 51.4% | 65.8% | 66.9% |
(3) \(\gamma\)=4, K=3, \(M_s\)=20 | 79.6% | 86.3% | 88.5% | 23.0% | 27.8% | 29.1% | 50.8% | 63.7% | 64.2% |
(4) \(\gamma\)=5, K=1, \(M_s\)=10 | 80.6% | 88.0% | 88.4% | 24.2% | 30.1% | 31.0% | 53.2% | 66.5% | 67.8% |
(5) \(\gamma\)=5, K=2, \(M_s\)=20 | 80.2% | 88.1% | 88.2% | 23.2% | 29.5% | 31.1% | 52.8% | 66.1% | 66.9% |
(6) \(\gamma\)=5, K=3, \(M_s\)=20 | 80.1% | 87.0% | 87.2% | 23.1% | 30.3% | 31.1% | 52.3% | 66.6% | 67.1% |
(7) \(\gamma\)=6, K=1, \(M_s\)=10 | 81.1% | 88.3% | 89.5% | 24.7% | 30.2% | 31.0% | 55.4% | 67.1% | 68.0% |
(8) \(\gamma\)=6, K=2, \(M_s\)=20 | 80.5% | 87.2% | 88.3% | 23.4% | 29.1% | 29.9% | 54.3% | 67.2% | 67.3% |
(9) \(\gamma\)=6, K=3, \(M_s\)=20 | 79.2% | 85.3% | 86.9% | 24.6% | 28.6% | 29.1% | 54.1% | 66.8% | 67.5% |
(10) \(\gamma\)=6, K=4, \(M_s\)=20 | 76.5% | 83.3% | 84.1% | 23.4% | 27.8% | 28.9% | 53.3% | 65.2% | 66.0% |
