Skip to main content
EURASIP Journal on Information Security

Table 5 Classification error (%) on the first 1000 test samples for the gray-box C&W transferability attacks from a single-channel model to a multi-channel model with randomly selected channels (the average results over 10 runs)

From: Machine learning through cryptographic glasses: combating adversarial attacks by key-based diversified aggregation

Data type

Transferability KDA

 

# channels · # classifiers

 

3

5

7

MNIST

Original

0.6

0.5

0.6

C&Wℓ2

5.06

4.77

4.44

C&Wℓ0

7.77

7.3

7.12

C&Wℓ∞

3.41

3.12

2.77

Fashion-MNIST

Original

8.2

8.2

8.1

C&Wℓ2

9.4

9.1

8.84

C&Wℓ0

10.58

10.52

10.27

C&Wℓ∞

9.33

9.09

8.83

CIFAR-10

Original

21.2

20.5

19.9

C&Wℓ2

22.92

21.22

21.1

C&Wℓ0

27.82

25.46

24.33

C&Wℓ∞

24.57

22.33

21.86

Back to article page