Skip to main content
EURASIP Journal on Information Security

Table 6 Classification error (%) on the first 1000 test samples (CIFAR-10) for the multi-channel system against the direct black-box OnePixel attacks 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

Attacked KDA

 

# channels · # classifiers

 

3

5

7

VGG16

Original

11.7

9.5

9.3

OnePixel p=1

11.3

9.6

9

OnePixel p=3

11.5

9.8

8.9

OnePixel p=5

12

10.6

9.4

ResNet18

Original

11.1

9.7

8.8

OnePixel p=1

11.1

9.2

8.9

OnePixel p=3

11.4

9.6

8.8

OnePixel p=5

10.9

9.8

9.1

Back to article page