Table 2 The optimal model accuracy under draft attack.

From: Aggregation algorithm based on consensus verification

 

MNIST (%)

CIFAR10 (%)

NoAttack

97.04

61.64

NoDefense

95.42

50.46

Bulyan

89.81

36.89

TrimmedMean

90.56

40.89

Krum

83.56

29.88

K-means

97.09

62.71

Loss_Baesd

93.3

48.66

Vector_Based

94.06

51.55

Back to article page