Skip to main content

Table 2 Classification error (%) on the first 1000 test samples (CIFAR-10) for the gray-box PGD 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

Attacked

Transferability

Transferability KDA

 
 

vanilla

vanilla

# channels · # classifiers

 
   

3

5

7

9

VGG 16

   

Original

10.7

11.7

11.6

9.9

9.5

9

PGD

16.1

15.2

14.25

12.16

11.75

11

ResNet 18

   

Original

9.5

10.6

11.7

9.3

8.8

8.1

PGD

17.9

14.9

14.7

11.29

10.67

9.7