Improving model robustness through hybrid adversarial training: integrating fgsm and pgd methods

HIGHLIGHTS

• What: The significant decrease in training loss shows that the model gradually learns the characteristics of the data, while the stability of the test loss shows that the model shows consistent performance on unseen data without obvious overfitting. The limitation of this study is that the selected model and dataset are relatively simple, and no experiments are conducted on more complex models and datasets, resulting in a small difference in the defensive performance comparison of the two different adversarial training strategies.
• Who: Hybrid Adversarial Training et al. from the College of Information Science and Engineering . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or E-mail:

  If you don't have an account, you can create one here.