Table 8 Performance of NHANet model with different epsilons under FGSM attack algorithm.

From: An incremental adversarial training method enables timeliness and rapid new knowledge acquisition

 

Original

\(\upepsilon\) = 1/255

\(\upepsilon\) = 2/255

\(\upepsilon\) = 3/255

\(\upepsilon\) = 4/255

\(\upepsilon\) = 5/255

Adversarial accuracy

0.9840

0.8893

0.5573

0.2880

0.1960

0.1307

Attack success rate

0.0

0.0962

0.4336

0.7073

0.8008

0.8672

Average L1 distance

0.0

1.1759

2.3518

3.5277

4.7035

5.8792

Average L2 distance

0.0

0.0679

0.1358

0.2037

0.2716

0.3395

Back to article page